diff --git "a/exp/log/log-train-2023-03-27-14-47-20-3" "b/exp/log/log-train-2023-03-27-14-47-20-3"
new file mode 100644--- /dev/null
+++ "b/exp/log/log-train-2023-03-27-14-47-20-3"
@@ -0,0 +1,12238 @@
+2023-03-27 14:47:20,927 INFO [train.py:962] (3/4) Training started
+2023-03-27 14:47:20,927 INFO [train.py:972] (3/4) Device: cuda:3
+2023-03-27 14:47:20,931 INFO [train.py:981] (3/4) {'frame_shift_ms': 10.0, 'allowed_excess_duration_ratio': 0.1, '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': '1.23.4', 'k2-build-type': 'Release', 'k2-with-cuda': True, 'k2-git-sha1': '9426c9f730820d291f5dcb06be337662595fa7b4', 'k2-git-date': 'Sun Feb 5 17:35:01 2023', 'lhotse-version': '1.13.0.dev+git.4cbd1bde.clean', 'torch-version': '1.10.0+cu102', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'bbpe', 'icefall-git-sha1': 'e03c10a-dirty', 'icefall-git-date': 'Mon Mar 27 00:05:03 2023', 'icefall-path': '/ceph-kw/kangwei/code/icefall_bbpe', 'k2-path': '/ceph-hw/kangwei/code/k2_release/k2/k2/python/k2/__init__.py', 'lhotse-path': '/ceph-hw/kangwei/dev_tools/anaconda3/envs/rnnt2/lib/python3.8/site-packages/lhotse-1.13.0.dev0+git.4cbd1bde.clean-py3.8.egg/lhotse/__init__.py', 'hostname': 'de-74279-k2-train-9-0208143539-7dcb6bfd79-b6fdq', 'IP address': '10.177.13.150'}, 'world_size': 4, 'master_port': 12535, 'tensorboard': True, 'num_epochs': 50, 'start_epoch': 1, 'start_batch': 0, 'exp_dir': PosixPath('pruned_transducer_stateless7_bbpe/exp'), 'bbpe_model': 'data/lang_bbpe_500/bbpe.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, 'num_encoder_layers': '2,4,3,2,4', 'feedforward_dims': '1024,1024,2048,2048,1024', 'nhead': '8,8,8,8,8', 'encoder_dims': '384,384,384,384,384', 'attention_dims': '192,192,192,192,192', 'encoder_unmasked_dims': '256,256,256,256,256', 'zipformer_downsampling_factors': '1,2,4,8,2', 'cnn_module_kernels': '31,31,31,31,31', 'decoder_dim': 512, 'joiner_dim': 512, 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 800, 'bucketing_sampler': True, 'num_buckets': 300, 'concatenate_cuts': False, 'duration_factor': 1.0, 'gap': 1.0, 'on_the_fly_feats': False, 'shuffle': True, 'drop_last': True, 'return_cuts': True, 'num_workers': 2, 'enable_spec_aug': True, 'spec_aug_time_warp_factor': 80, 'enable_musan': True, 'input_strategy': 'PrecomputedFeatures', 'blank_id': 0, 'vocab_size': 500}
+2023-03-27 14:47:20,931 INFO [train.py:983] (3/4) About to create model
+2023-03-27 14:47:21,866 INFO [zipformer.py:178] (3/4) 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.
+2023-03-27 14:47:21,890 INFO [train.py:987] (3/4) Number of model parameters: 70369391
+2023-03-27 14:47:28,399 INFO [train.py:1002] (3/4) Using DDP
+2023-03-27 14:47:28,704 INFO [asr_datamodule.py:407] (3/4) About to get train cuts
+2023-03-27 14:47:28,707 INFO [train.py:1083] (3/4) Filtering short and long utterances.
+2023-03-27 14:47:28,707 INFO [train.py:1086] (3/4) Tokenizing and encoding texts in train cuts.
+2023-03-27 14:47:28,707 INFO [asr_datamodule.py:224] (3/4) About to get Musan cuts
+2023-03-27 14:47:32,013 INFO [asr_datamodule.py:229] (3/4) Enable MUSAN
+2023-03-27 14:47:32,013 INFO [asr_datamodule.py:252] (3/4) Enable SpecAugment
+2023-03-27 14:47:32,013 INFO [asr_datamodule.py:253] (3/4) Time warp factor: 80
+2023-03-27 14:47:32,013 INFO [asr_datamodule.py:263] (3/4) Num frame mask: 10
+2023-03-27 14:47:32,014 INFO [asr_datamodule.py:276] (3/4) About to create train dataset
+2023-03-27 14:47:32,014 INFO [asr_datamodule.py:303] (3/4) Using DynamicBucketingSampler.
+2023-03-27 14:47:42,539 INFO [asr_datamodule.py:320] (3/4) About to create train dataloader
+2023-03-27 14:47:42,540 INFO [asr_datamodule.py:414] (3/4) About to get dev cuts
+2023-03-27 14:47:42,542 INFO [train.py:1102] (3/4) Tokenizing and encoding texts in valid cuts.
+2023-03-27 14:47:42,542 INFO [asr_datamodule.py:351] (3/4) About to create dev dataset
+2023-03-27 14:47:43,376 INFO [asr_datamodule.py:370] (3/4) About to create dev dataloader
+2023-03-27 14:48:25,411 INFO [train.py:892] (3/4) Epoch 1, batch 0, loss[loss=7.391, simple_loss=6.697, pruned_loss=6.928, over 19912.00 frames. ], tot_loss[loss=7.391, simple_loss=6.697, pruned_loss=6.928, over 19912.00 frames. ], batch size: 116, lr: 2.50e-02, grad_scale: 2.0
+2023-03-27 14:48:25,411 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-27 14:48:53,175 INFO [train.py:926] (3/4) Epoch 1, validation: loss=6.85, simple_loss=6.179, pruned_loss=6.691, over 2883724.00 frames. 
+2023-03-27 14:48:53,177 INFO [train.py:927] (3/4) Maximum memory allocated so far is 13770MB
+2023-03-27 14:49:01,327 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5.0, num_to_drop=2, layers_to_drop={1, 2}
+2023-03-27 14:49:05,203 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=3.82 vs. limit=2.0
+2023-03-27 14:49:31,451 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 14:50:04,792 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=10.19 vs. limit=2.0
+2023-03-27 14:50:15,341 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=195.51 vs. limit=5.0
+2023-03-27 14:50:17,292 INFO [train.py:892] (3/4) Epoch 1, batch 50, loss[loss=1.282, simple_loss=1.135, pruned_loss=1.313, over 19749.00 frames. ], tot_loss[loss=2.201, simple_loss=1.995, pruned_loss=1.982, over 890355.70 frames. ], batch size: 253, lr: 2.75e-02, grad_scale: 0.5
+2023-03-27 14:51:13,781 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 14:51:39,746 INFO [train.py:892] (3/4) Epoch 1, batch 100, loss[loss=0.7812, simple_loss=0.6666, pruned_loss=0.9035, over 19810.00 frames. ], tot_loss[loss=1.554, simple_loss=1.381, pruned_loss=1.558, over 1567670.50 frames. ], batch size: 114, lr: 3.00e-02, grad_scale: 1.0
+2023-03-27 14:51:42,869 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.423e+01 2.042e+02 3.545e+02 1.360e+03 1.838e+04, threshold=7.089e+02, percent-clipped=0.0
+2023-03-27 14:51:50,747 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.5203, 5.5160, 5.5232, 5.5239, 5.5258, 5.5236, 5.5220, 5.5153],
+       device='cuda:3'), covar=tensor([0.0016, 0.0020, 0.0025, 0.0023, 0.0018, 0.0023, 0.0014, 0.0017],
+       device='cuda:3'), in_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009],
+       device='cuda:3'), out_proj_covar=tensor([9.4301e-06, 9.3490e-06, 9.2599e-06, 9.1772e-06, 9.1839e-06, 9.2881e-06,
+        9.1292e-06, 9.3122e-06], device='cuda:3')
+2023-03-27 14:52:51,616 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=144.0, num_to_drop=2, layers_to_drop={1, 2}
+2023-03-27 14:53:04,049 INFO [train.py:892] (3/4) Epoch 1, batch 150, loss[loss=0.811, simple_loss=0.6919, pruned_loss=0.8649, over 19737.00 frames. ], tot_loss[loss=1.259, simple_loss=1.106, pruned_loss=1.303, over 2096475.77 frames. ], batch size: 92, lr: 3.25e-02, grad_scale: 1.0
+2023-03-27 14:53:44,668 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.50 vs. limit=2.0
+2023-03-27 14:54:26,335 INFO [train.py:892] (3/4) Epoch 1, batch 200, loss[loss=0.7443, simple_loss=0.6368, pruned_loss=0.7326, over 19851.00 frames. ], tot_loss[loss=1.107, simple_loss=0.9645, pruned_loss=1.146, over 2508579.07 frames. ], batch size: 197, lr: 3.50e-02, grad_scale: 1.0
+2023-03-27 14:54:29,325 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 8.392e+01 1.785e+02 2.449e+02 3.359e+02 7.299e+02, threshold=4.898e+02, percent-clipped=1.0
+2023-03-27 14:55:44,974 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.75 vs. limit=5.0
+2023-03-27 14:55:46,820 INFO [train.py:892] (3/4) Epoch 1, batch 250, loss[loss=0.8294, simple_loss=0.6964, pruned_loss=0.8299, over 19867.00 frames. ], tot_loss[loss=1.013, simple_loss=0.877, pruned_loss=1.037, over 2827371.34 frames. ], batch size: 46, lr: 3.75e-02, grad_scale: 1.0
+2023-03-27 14:57:03,590 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 14:57:10,702 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300.0, num_to_drop=2, layers_to_drop={1, 2}
+2023-03-27 14:57:11,391 INFO [train.py:892] (3/4) Epoch 1, batch 300, loss[loss=0.8331, simple_loss=0.6949, pruned_loss=0.8092, over 19850.00 frames. ], tot_loss[loss=0.9501, simple_loss=0.8171, pruned_loss=0.9584, over 3076498.67 frames. ], batch size: 58, lr: 4.00e-02, grad_scale: 1.0
+2023-03-27 14:57:14,287 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 9.056e+01 1.487e+02 2.142e+02 2.914e+02 5.641e+02, threshold=4.285e+02, percent-clipped=2.0
+2023-03-27 14:58:31,274 INFO [train.py:892] (3/4) Epoch 1, batch 350, loss[loss=0.7089, simple_loss=0.594, pruned_loss=0.6473, over 19817.00 frames. ], tot_loss[loss=0.9079, simple_loss=0.7759, pruned_loss=0.9008, over 3269069.89 frames. ], batch size: 133, lr: 4.25e-02, grad_scale: 1.0
+2023-03-27 14:58:35,148 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.7175, 5.8043, 5.6806, 5.7820, 5.6970, 5.8171, 5.6611, 5.8176],
+       device='cuda:3'), covar=tensor([0.0046, 0.0053, 0.0116, 0.0049, 0.0076, 0.0049, 0.0106, 0.0045],
+       device='cuda:3'), in_proj_covar=tensor([0.0010, 0.0009, 0.0010, 0.0009, 0.0010, 0.0009, 0.0009, 0.0009],
+       device='cuda:3'), out_proj_covar=tensor([9.6279e-06, 9.8732e-06, 9.5947e-06, 9.3665e-06, 9.4341e-06, 9.5127e-06,
+        9.4523e-06, 9.5237e-06], device='cuda:3')
+2023-03-27 14:58:41,004 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=357.0, num_to_drop=2, layers_to_drop={0, 2}
+2023-03-27 14:59:30,245 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 14:59:51,339 INFO [train.py:892] (3/4) Epoch 1, batch 400, loss[loss=0.8137, simple_loss=0.6806, pruned_loss=0.7179, over 19763.00 frames. ], tot_loss[loss=0.8757, simple_loss=0.7441, pruned_loss=0.8522, over 3420227.75 frames. ], batch size: 226, lr: 4.50e-02, grad_scale: 2.0
+2023-03-27 14:59:54,323 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.004e+02 1.540e+02 2.069e+02 2.975e+02 6.292e+02, threshold=4.137e+02, percent-clipped=2.0
+2023-03-27 15:00:28,953 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1530, 5.0236, 3.8826, 3.9570, 4.4585, 5.9391, 4.1091, 5.9334],
+       device='cuda:3'), covar=tensor([0.1322, 0.1207, 0.2949, 0.2749, 0.1268, 0.0079, 0.1814, 0.0073],
+       device='cuda:3'), in_proj_covar=tensor([0.0012, 0.0011, 0.0013, 0.0012, 0.0012, 0.0011, 0.0012, 0.0011],
+       device='cuda:3'), out_proj_covar=tensor([1.1604e-05, 1.2292e-05, 1.2210e-05, 1.1661e-05, 1.1292e-05, 1.1146e-05,
+        1.2261e-05, 1.1158e-05], device='cuda:3')
+2023-03-27 15:00:48,922 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7913, 3.8387, 3.7104, 3.6304, 3.8902, 4.7327, 3.5848, 4.7111],
+       device='cuda:3'), covar=tensor([0.1004, 0.2488, 0.1838, 0.2487, 0.1548, 0.0105, 0.2407, 0.0117],
+       device='cuda:3'), in_proj_covar=tensor([0.0013, 0.0013, 0.0014, 0.0014, 0.0013, 0.0012, 0.0014, 0.0012],
+       device='cuda:3'), out_proj_covar=tensor([1.2831e-05, 1.3905e-05, 1.3732e-05, 1.3362e-05, 1.2633e-05, 1.2018e-05,
+        1.4092e-05, 1.2117e-05], device='cuda:3')
+2023-03-27 15:00:51,886 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=439.0, num_to_drop=2, layers_to_drop={2, 3}
+2023-03-27 15:00:55,013 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=5.63 vs. limit=2.0
+2023-03-27 15:01:07,673 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=448.0, num_to_drop=2, layers_to_drop={1, 3}
+2023-03-27 15:01:12,540 INFO [train.py:892] (3/4) Epoch 1, batch 450, loss[loss=0.6934, simple_loss=0.5801, pruned_loss=0.59, over 19817.00 frames. ], tot_loss[loss=0.8562, simple_loss=0.7245, pruned_loss=0.8132, over 3535896.68 frames. ], batch size: 143, lr: 4.75e-02, grad_scale: 2.0
+2023-03-27 15:01:22,638 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.94 vs. limit=2.0
+2023-03-27 15:02:30,383 INFO [train.py:892] (3/4) Epoch 1, batch 500, loss[loss=0.6489, simple_loss=0.547, pruned_loss=0.5237, over 19767.00 frames. ], tot_loss[loss=0.8324, simple_loss=0.7028, pruned_loss=0.7696, over 3627463.51 frames. ], batch size: 152, lr: 4.99e-02, grad_scale: 2.0
+2023-03-27 15:02:33,256 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.044e+02 2.386e+02 3.363e+02 4.760e+02 1.006e+03, threshold=6.727e+02, percent-clipped=34.0
+2023-03-27 15:03:17,524 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7609, 3.4905, 3.6805, 3.4334, 3.3351, 5.0783, 3.4326, 3.6016],
+       device='cuda:3'), covar=tensor([0.2344, 0.3564, 0.1939, 0.2248, 0.1637, 0.0099, 0.2027, 0.3465],
+       device='cuda:3'), in_proj_covar=tensor([0.0018, 0.0018, 0.0019, 0.0019, 0.0017, 0.0015, 0.0018, 0.0019],
+       device='cuda:3'), out_proj_covar=tensor([1.7115e-05, 1.9309e-05, 1.7639e-05, 1.8689e-05, 1.6561e-05, 1.5176e-05,
+        1.8375e-05, 1.8720e-05], device='cuda:3')
+2023-03-27 15:03:50,988 INFO [train.py:892] (3/4) Epoch 1, batch 550, loss[loss=0.6878, simple_loss=0.5819, pruned_loss=0.5347, over 19736.00 frames. ], tot_loss[loss=0.8128, simple_loss=0.6858, pruned_loss=0.7289, over 3696276.35 frames. ], batch size: 134, lr: 4.98e-02, grad_scale: 2.0
+2023-03-27 15:04:03,932 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=7.25 vs. limit=5.0
+2023-03-27 15:04:10,954 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=562.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 15:04:21,054 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=568.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 15:04:55,609 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=590.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 15:05:13,811 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600.0, num_to_drop=2, layers_to_drop={1, 2}
+2023-03-27 15:05:14,219 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.33 vs. limit=5.0
+2023-03-27 15:05:14,479 INFO [train.py:892] (3/4) Epoch 1, batch 600, loss[loss=0.704, simple_loss=0.5978, pruned_loss=0.5292, over 19722.00 frames. ], tot_loss[loss=0.7871, simple_loss=0.6649, pruned_loss=0.6838, over 3753735.74 frames. ], batch size: 81, lr: 4.98e-02, grad_scale: 2.0
+2023-03-27 15:05:17,272 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=3.21 vs. limit=2.0
+2023-03-27 15:05:17,530 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.786e+02 4.093e+02 5.434e+02 6.548e+02 1.823e+03, threshold=1.087e+03, percent-clipped=20.0
+2023-03-27 15:05:49,303 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=623.0, num_to_drop=2, layers_to_drop={0, 1}
+2023-03-27 15:05:59,533 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=629.0, num_to_drop=2, layers_to_drop={0, 2}
+2023-03-27 15:06:31,698 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=648.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 15:06:35,616 INFO [train.py:892] (3/4) Epoch 1, batch 650, loss[loss=0.6205, simple_loss=0.5313, pruned_loss=0.447, over 19887.00 frames. ], tot_loss[loss=0.7609, simple_loss=0.644, pruned_loss=0.6403, over 3798467.10 frames. ], batch size: 134, lr: 4.98e-02, grad_scale: 2.0
+2023-03-27 15:06:36,422 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651.0, num_to_drop=2, layers_to_drop={1, 2}
+2023-03-27 15:06:37,911 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=652.0, num_to_drop=2, layers_to_drop={2, 3}
+2023-03-27 15:07:56,921 INFO [train.py:892] (3/4) Epoch 1, batch 700, loss[loss=0.7152, simple_loss=0.6001, pruned_loss=0.5302, over 19735.00 frames. ], tot_loss[loss=0.7383, simple_loss=0.6264, pruned_loss=0.6017, over 3832423.51 frames. ], batch size: 118, lr: 4.98e-02, grad_scale: 2.0
+2023-03-27 15:07:59,945 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.633e+02 4.666e+02 6.058e+02 9.217e+02 2.342e+03, threshold=1.212e+03, percent-clipped=17.0
+2023-03-27 15:08:35,685 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5131, 1.3560, 2.7914, 2.4572, 2.4014, 2.9503, 2.2994, 1.9656],
+       device='cuda:3'), covar=tensor([0.4283, 0.9664, 0.2562, 0.3715, 0.4521, 0.2240, 0.4436, 0.5144],
+       device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0019, 0.0017, 0.0020, 0.0019, 0.0016, 0.0019, 0.0018],
+       device='cuda:3'), out_proj_covar=tensor([1.9393e-05, 1.7674e-05, 1.2753e-05, 1.6496e-05, 1.6590e-05, 1.3068e-05,
+        1.6663e-05, 1.4807e-05], device='cuda:3')
+2023-03-27 15:08:59,067 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=739.0, num_to_drop=2, layers_to_drop={0, 2}
+2023-03-27 15:09:06,935 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=743.0, num_to_drop=2, layers_to_drop={0, 3}
+2023-03-27 15:09:20,459 INFO [train.py:892] (3/4) Epoch 1, batch 750, loss[loss=0.6928, simple_loss=0.5991, pruned_loss=0.4709, over 19829.00 frames. ], tot_loss[loss=0.7215, simple_loss=0.6134, pruned_loss=0.5707, over 3859015.17 frames. ], batch size: 57, lr: 4.97e-02, grad_scale: 2.0
+2023-03-27 15:10:17,411 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=787.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 15:10:32,724 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.13 vs. limit=2.0
+2023-03-27 15:10:39,432 INFO [train.py:892] (3/4) Epoch 1, batch 800, loss[loss=0.5651, simple_loss=0.4938, pruned_loss=0.3696, over 19868.00 frames. ], tot_loss[loss=0.7003, simple_loss=0.5973, pruned_loss=0.5377, over 3879828.03 frames. ], batch size: 46, lr: 4.97e-02, grad_scale: 4.0
+2023-03-27 15:10:42,238 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.256e+02 4.881e+02 6.070e+02 8.660e+02 1.858e+03, threshold=1.214e+03, percent-clipped=11.0
+2023-03-27 15:11:40,083 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1238, 4.4100, 4.4234, 4.2223, 4.4531, 4.2636, 4.1617, 4.1700],
+       device='cuda:3'), covar=tensor([0.0850, 0.0672, 0.0794, 0.0857, 0.0599, 0.0754, 0.0982, 0.0800],
+       device='cuda:3'), in_proj_covar=tensor([0.0028, 0.0026, 0.0027, 0.0028, 0.0026, 0.0027, 0.0027, 0.0028],
+       device='cuda:3'), out_proj_covar=tensor([2.8597e-05, 2.7423e-05, 2.6352e-05, 2.7448e-05, 2.5853e-05, 2.6379e-05,
+        2.8313e-05, 2.6426e-05], device='cuda:3')
+2023-03-27 15:11:53,996 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=847.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 15:12:00,810 INFO [train.py:892] (3/4) Epoch 1, batch 850, loss[loss=0.6015, simple_loss=0.5222, pruned_loss=0.3943, over 19655.00 frames. ], tot_loss[loss=0.6815, simple_loss=0.5834, pruned_loss=0.5082, over 3894986.43 frames. ], batch size: 67, lr: 4.96e-02, grad_scale: 4.0
+2023-03-27 15:14:32,572 INFO [train.py:892] (3/4) Epoch 1, batch 900, loss[loss=0.6275, simple_loss=0.543, pruned_loss=0.4093, over 19821.00 frames. ], tot_loss[loss=0.6706, simple_loss=0.5748, pruned_loss=0.4883, over 3905710.68 frames. ], batch size: 229, lr: 4.96e-02, grad_scale: 4.0
+2023-03-27 15:14:38,993 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.144e+02 5.792e+02 7.433e+02 9.367e+02 4.103e+03, threshold=1.487e+03, percent-clipped=16.0
+2023-03-27 15:14:57,692 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=908.0, num_to_drop=2, layers_to_drop={0, 2}
+2023-03-27 15:15:27,128 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918.0, num_to_drop=2, layers_to_drop={2, 3}
+2023-03-27 15:15:45,084 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=924.0, num_to_drop=2, layers_to_drop={0, 3}
+2023-03-27 15:17:00,715 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=946.0, num_to_drop=2, layers_to_drop={0, 2}
+2023-03-27 15:17:13,608 INFO [train.py:892] (3/4) Epoch 1, batch 950, loss[loss=0.5278, simple_loss=0.4691, pruned_loss=0.3225, over 19872.00 frames. ], tot_loss[loss=0.6529, simple_loss=0.5618, pruned_loss=0.4631, over 3916042.81 frames. ], batch size: 108, lr: 4.96e-02, grad_scale: 4.0
+2023-03-27 15:17:17,195 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=952.0, num_to_drop=2, layers_to_drop={0, 1}
+2023-03-27 15:18:16,423 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0
+2023-03-27 15:18:37,718 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5247, 3.7861, 3.4318, 3.5212, 3.4050, 2.9702, 3.6074, 3.0412],
+       device='cuda:3'), covar=tensor([0.2208, 0.2000, 0.2543, 0.2334, 0.2419, 0.3456, 0.1889, 0.1765],
+       device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0046, 0.0050, 0.0046, 0.0048, 0.0055, 0.0049, 0.0049],
+       device='cuda:3'), out_proj_covar=tensor([4.6555e-05, 3.8635e-05, 4.3616e-05, 3.7834e-05, 4.2260e-05, 4.4320e-05,
+        4.4152e-05, 4.1393e-05], device='cuda:3')
+2023-03-27 15:19:15,008 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1000.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 15:19:15,789 INFO [train.py:892] (3/4) Epoch 1, batch 1000, loss[loss=0.5634, simple_loss=0.4942, pruned_loss=0.3505, over 19830.00 frames. ], tot_loss[loss=0.6435, simple_loss=0.5549, pruned_loss=0.4465, over 3921643.28 frames. ], batch size: 76, lr: 4.95e-02, grad_scale: 4.0
+2023-03-27 15:19:21,124 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.942e+02 5.404e+02 6.507e+02 8.567e+02 3.462e+03, threshold=1.301e+03, percent-clipped=3.0
+2023-03-27 15:21:24,807 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1043.0, num_to_drop=2, layers_to_drop={0, 1}
+2023-03-27 15:21:50,442 INFO [train.py:892] (3/4) Epoch 1, batch 1050, loss[loss=0.5607, simple_loss=0.4912, pruned_loss=0.3466, over 19800.00 frames. ], tot_loss[loss=0.6336, simple_loss=0.547, pruned_loss=0.4315, over 3929029.23 frames. ], batch size: 111, lr: 4.95e-02, grad_scale: 4.0
+2023-03-27 15:22:31,961 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7932, 3.6695, 3.6811, 3.5876, 3.5317, 3.4666, 3.7107, 3.7454],
+       device='cuda:3'), covar=tensor([0.0660, 0.0614, 0.0618, 0.0537, 0.0553, 0.0590, 0.0505, 0.0604],
+       device='cuda:3'), in_proj_covar=tensor([0.0034, 0.0029, 0.0034, 0.0035, 0.0031, 0.0035, 0.0032, 0.0029],
+       device='cuda:3'), out_proj_covar=tensor([3.1073e-05, 2.5853e-05, 3.0663e-05, 3.2891e-05, 2.9204e-05, 3.0326e-05,
+        2.8336e-05, 2.4846e-05], device='cuda:3')
+2023-03-27 15:23:44,420 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1091.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 15:24:13,072 INFO [train.py:892] (3/4) Epoch 1, batch 1100, loss[loss=0.5279, simple_loss=0.4646, pruned_loss=0.321, over 19774.00 frames. ], tot_loss[loss=0.6186, simple_loss=0.5363, pruned_loss=0.412, over 3933380.57 frames. ], batch size: 182, lr: 4.94e-02, grad_scale: 4.0
+2023-03-27 15:24:19,018 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.337e+02 6.074e+02 7.964e+02 1.002e+03 2.431e+03, threshold=1.593e+03, percent-clipped=21.0
+2023-03-27 15:26:40,046 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-27 15:26:44,150 INFO [train.py:892] (3/4) Epoch 1, batch 1150, loss[loss=0.5878, simple_loss=0.5073, pruned_loss=0.3665, over 19778.00 frames. ], tot_loss[loss=0.6071, simple_loss=0.5282, pruned_loss=0.3965, over 3936674.07 frames. ], batch size: 280, lr: 4.94e-02, grad_scale: 4.0
+2023-03-27 15:26:48,844 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1153.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 15:26:58,785 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6158, 3.8733, 3.6286, 3.7745, 3.2806, 2.9238, 3.6066, 2.6727],
+       device='cuda:3'), covar=tensor([0.1364, 0.1210, 0.1485, 0.1188, 0.1578, 0.2488, 0.1258, 0.1640],
+       device='cuda:3'), in_proj_covar=tensor([0.0052, 0.0045, 0.0049, 0.0045, 0.0048, 0.0051, 0.0049, 0.0047],
+       device='cuda:3'), out_proj_covar=tensor([4.2351e-05, 3.6302e-05, 4.1443e-05, 3.5507e-05, 4.0484e-05, 4.1444e-05,
+        4.0943e-05, 3.8858e-05], device='cuda:3')
+2023-03-27 15:28:34,694 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.1372, 1.9733, 2.1501, 2.4945, 2.0035, 2.0119, 2.2681, 2.0616],
+       device='cuda:3'), covar=tensor([0.1520, 0.1382, 0.1028, 0.0818, 0.1166, 0.1369, 0.1217, 0.2033],
+       device='cuda:3'), in_proj_covar=tensor([0.0048, 0.0055, 0.0054, 0.0050, 0.0057, 0.0050, 0.0051, 0.0054],
+       device='cuda:3'), out_proj_covar=tensor([4.2710e-05, 4.7493e-05, 4.6763e-05, 4.4382e-05, 5.2308e-05, 4.5827e-05,
+        4.5056e-05, 4.7236e-05], device='cuda:3')
+2023-03-27 15:28:37,489 INFO [train.py:892] (3/4) Epoch 1, batch 1200, loss[loss=0.5153, simple_loss=0.4672, pruned_loss=0.2941, over 19537.00 frames. ], tot_loss[loss=0.5997, simple_loss=0.5233, pruned_loss=0.385, over 3937799.75 frames. ], batch size: 46, lr: 4.93e-02, grad_scale: 8.0
+2023-03-27 15:28:40,927 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.874e+02 6.275e+02 7.572e+02 9.981e+02 2.448e+03, threshold=1.514e+03, percent-clipped=4.0
+2023-03-27 15:28:41,659 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1203.0, num_to_drop=2, layers_to_drop={0, 2}
+2023-03-27 15:29:12,834 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1214.0, num_to_drop=2, layers_to_drop={2, 3}
+2023-03-27 15:29:26,892 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1218.0, num_to_drop=2, layers_to_drop={0, 2}
+2023-03-27 15:29:46,000 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1224.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 15:30:51,253 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1246.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 15:31:03,313 INFO [train.py:892] (3/4) Epoch 1, batch 1250, loss[loss=0.5059, simple_loss=0.4472, pruned_loss=0.2995, over 19749.00 frames. ], tot_loss[loss=0.5876, simple_loss=0.5149, pruned_loss=0.3705, over 3941002.21 frames. ], batch size: 129, lr: 4.92e-02, grad_scale: 8.0
+2023-03-27 15:31:48,016 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1266.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 15:31:54,456 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.8471, 6.0785, 6.1258, 5.8812, 6.1538, 5.8551, 5.8223, 5.5587],
+       device='cuda:3'), covar=tensor([0.0327, 0.0341, 0.0401, 0.0307, 0.0369, 0.0429, 0.0422, 0.0528],
+       device='cuda:3'), in_proj_covar=tensor([0.0055, 0.0050, 0.0053, 0.0052, 0.0050, 0.0053, 0.0053, 0.0063],
+       device='cuda:3'), out_proj_covar=tensor([5.6752e-05, 5.2222e-05, 5.5125e-05, 5.4434e-05, 5.6824e-05, 5.1209e-05,
+        5.3302e-05, 6.3220e-05], device='cuda:3')
+2023-03-27 15:32:00,209 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 15:32:47,426 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 15:32:57,675 INFO [train.py:892] (3/4) Epoch 1, batch 1300, loss[loss=0.5901, simple_loss=0.5174, pruned_loss=0.3515, over 19750.00 frames. ], tot_loss[loss=0.5815, simple_loss=0.5107, pruned_loss=0.3615, over 3942950.30 frames. ], batch size: 250, lr: 4.92e-02, grad_scale: 8.0
+2023-03-27 15:33:00,840 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.916e+02 5.972e+02 8.796e+02 1.171e+03 2.494e+03, threshold=1.759e+03, percent-clipped=14.0
+2023-03-27 15:34:24,476 INFO [train.py:892] (3/4) Epoch 1, batch 1350, loss[loss=0.5414, simple_loss=0.4876, pruned_loss=0.3087, over 19744.00 frames. ], tot_loss[loss=0.572, simple_loss=0.5043, pruned_loss=0.3502, over 3945420.86 frames. ], batch size: 253, lr: 4.91e-02, grad_scale: 8.0
+2023-03-27 15:35:50,286 INFO [train.py:892] (3/4) Epoch 1, batch 1400, loss[loss=0.5045, simple_loss=0.454, pruned_loss=0.2869, over 19805.00 frames. ], tot_loss[loss=0.5608, simple_loss=0.4967, pruned_loss=0.3383, over 3946770.61 frames. ], batch size: 111, lr: 4.91e-02, grad_scale: 8.0
+2023-03-27 15:35:53,769 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.724e+02 6.220e+02 8.188e+02 1.056e+03 1.766e+03, threshold=1.638e+03, percent-clipped=1.0
+2023-03-27 15:36:50,026 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1434.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 15:37:19,445 INFO [train.py:892] (3/4) Epoch 1, batch 1450, loss[loss=0.5622, simple_loss=0.49, pruned_loss=0.332, over 19877.00 frames. ], tot_loss[loss=0.5543, simple_loss=0.4925, pruned_loss=0.3303, over 3946887.37 frames. ], batch size: 139, lr: 4.90e-02, grad_scale: 8.0
+2023-03-27 15:38:38,521 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1495.0, num_to_drop=2, layers_to_drop={2, 3}
+2023-03-27 15:38:47,161 INFO [train.py:892] (3/4) Epoch 1, batch 1500, loss[loss=0.4568, simple_loss=0.4212, pruned_loss=0.2498, over 19775.00 frames. ], tot_loss[loss=0.546, simple_loss=0.4871, pruned_loss=0.3213, over 3947259.44 frames. ], batch size: 46, lr: 4.89e-02, grad_scale: 8.0
+2023-03-27 15:38:51,708 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.400e+02 6.710e+02 8.347e+02 1.087e+03 2.003e+03, threshold=1.669e+03, percent-clipped=5.0
+2023-03-27 15:38:52,726 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1503.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 15:39:02,569 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1509.0, num_to_drop=2, layers_to_drop={2, 3}
+2023-03-27 15:39:47,408 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9155, 4.4164, 4.7099, 4.3452, 4.7525, 4.2936, 4.3963, 4.8672],
+       device='cuda:3'), covar=tensor([0.0598, 0.0412, 0.0400, 0.0423, 0.0397, 0.0500, 0.0446, 0.0412],
+       device='cuda:3'), in_proj_covar=tensor([0.0036, 0.0033, 0.0038, 0.0040, 0.0037, 0.0039, 0.0037, 0.0032],
+       device='cuda:3'), out_proj_covar=tensor([3.4061e-05, 2.6352e-05, 3.3753e-05, 3.6236e-05, 3.2800e-05, 3.3850e-05,
+        3.2107e-05, 2.7429e-05], device='cuda:3')
+2023-03-27 15:40:06,651 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2094, 4.4602, 4.5020, 4.4470, 4.5822, 4.2688, 4.3148, 4.1666],
+       device='cuda:3'), covar=tensor([0.0498, 0.0514, 0.0601, 0.0444, 0.0490, 0.0635, 0.0603, 0.0717],
+       device='cuda:3'), in_proj_covar=tensor([0.0068, 0.0061, 0.0068, 0.0063, 0.0063, 0.0065, 0.0068, 0.0082],
+       device='cuda:3'), out_proj_covar=tensor([6.9875e-05, 6.3146e-05, 7.2748e-05, 6.7291e-05, 7.6656e-05, 6.3061e-05,
+        6.8689e-05, 8.7013e-05], device='cuda:3')
+2023-03-27 15:40:15,406 INFO [train.py:892] (3/4) Epoch 1, batch 1550, loss[loss=0.5489, simple_loss=0.4885, pruned_loss=0.3131, over 19737.00 frames. ], tot_loss[loss=0.5414, simple_loss=0.4844, pruned_loss=0.3152, over 3946961.23 frames. ], batch size: 269, lr: 4.89e-02, grad_scale: 8.0
+2023-03-27 15:40:16,162 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 15:40:24,270 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.1292, 1.9718, 3.1603, 2.6667, 1.9338, 2.8493, 2.0224, 1.5089],
+       device='cuda:3'), covar=tensor([0.6015, 1.8541, 0.1185, 0.1710, 0.7876, 0.2564, 0.3807, 1.0049],
+       device='cuda:3'), in_proj_covar=tensor([0.0046, 0.0064, 0.0034, 0.0035, 0.0052, 0.0034, 0.0040, 0.0060],
+       device='cuda:3'), out_proj_covar=tensor([4.4713e-05, 6.3081e-05, 2.4715e-05, 2.6460e-05, 5.1114e-05, 2.4839e-05,
+        3.4745e-05, 5.5653e-05], device='cuda:3')
+2023-03-27 15:40:45,132 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1566.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 15:40:57,004 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0
+2023-03-27 15:41:45,849 INFO [train.py:892] (3/4) Epoch 1, batch 1600, loss[loss=0.534, simple_loss=0.4821, pruned_loss=0.2984, over 19769.00 frames. ], tot_loss[loss=0.5362, simple_loss=0.4819, pruned_loss=0.3087, over 3947027.14 frames. ], batch size: 66, lr: 4.88e-02, grad_scale: 8.0
+2023-03-27 15:41:50,276 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.877e+02 6.710e+02 8.151e+02 1.107e+03 2.056e+03, threshold=1.630e+03, percent-clipped=4.0
+2023-03-27 15:42:05,360 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1611.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 15:42:24,545 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5831, 3.6710, 3.6909, 3.7390, 3.6537, 3.7011, 3.5392, 3.2039],
+       device='cuda:3'), covar=tensor([0.0594, 0.0605, 0.0619, 0.0496, 0.0449, 0.0550, 0.0560, 0.0767],
+       device='cuda:3'), in_proj_covar=tensor([0.0047, 0.0043, 0.0047, 0.0045, 0.0045, 0.0046, 0.0048, 0.0055],
+       device='cuda:3'), out_proj_covar=tensor([4.6423e-05, 4.3213e-05, 4.5949e-05, 4.6230e-05, 4.1112e-05, 4.2896e-05,
+        4.2908e-05, 5.3167e-05], device='cuda:3')
+2023-03-27 15:42:30,937 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1627.0, num_to_drop=2, layers_to_drop={1, 2}
+2023-03-27 15:42:41,427 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.8935, 2.0306, 2.8523, 2.5258, 1.8755, 2.7851, 2.1652, 1.5710],
+       device='cuda:3'), covar=tensor([0.4197, 1.0387, 0.0956, 0.1144, 0.5894, 0.1662, 0.2366, 0.6475],
+       device='cuda:3'), in_proj_covar=tensor([0.0049, 0.0068, 0.0035, 0.0037, 0.0058, 0.0036, 0.0042, 0.0062],
+       device='cuda:3'), out_proj_covar=tensor([4.7505e-05, 6.7527e-05, 2.5086e-05, 2.7049e-05, 5.6441e-05, 2.6632e-05,
+        3.7222e-05, 5.8487e-05], device='cuda:3')
+2023-03-27 15:43:13,160 INFO [train.py:892] (3/4) Epoch 1, batch 1650, loss[loss=0.4266, simple_loss=0.3958, pruned_loss=0.2303, over 19754.00 frames. ], tot_loss[loss=0.532, simple_loss=0.4792, pruned_loss=0.3039, over 3947660.72 frames. ], batch size: 102, lr: 4.87e-02, grad_scale: 8.0
+2023-03-27 15:43:47,365 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1113, 3.1230, 3.2448, 3.0917, 2.5661, 2.8541, 2.6140, 2.4512],
+       device='cuda:3'), covar=tensor([0.0218, 0.0305, 0.0293, 0.0217, 0.0733, 0.0233, 0.0376, 0.0771],
+       device='cuda:3'), in_proj_covar=tensor([0.0026, 0.0026, 0.0027, 0.0027, 0.0034, 0.0026, 0.0030, 0.0035],
+       device='cuda:3'), out_proj_covar=tensor([2.0292e-05, 1.9594e-05, 2.1920e-05, 2.0952e-05, 3.1784e-05, 1.9670e-05,
+        2.5659e-05, 3.0395e-05], device='cuda:3')
+2023-03-27 15:43:50,630 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1672.0, num_to_drop=2, layers_to_drop={0, 1}
+2023-03-27 15:44:39,831 INFO [train.py:892] (3/4) Epoch 1, batch 1700, loss[loss=0.5228, simple_loss=0.4742, pruned_loss=0.2892, over 19779.00 frames. ], tot_loss[loss=0.5208, simple_loss=0.4722, pruned_loss=0.2939, over 3949574.08 frames. ], batch size: 247, lr: 4.86e-02, grad_scale: 8.0
+2023-03-27 15:44:43,066 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.702e+02 6.108e+02 7.327e+02 1.005e+03 2.757e+03, threshold=1.465e+03, percent-clipped=5.0
+2023-03-27 15:44:44,127 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.3830, 3.1767, 2.2933, 3.2613, 3.0723, 2.1111, 3.1008, 2.4544],
+       device='cuda:3'), covar=tensor([0.0930, 0.0387, 0.0916, 0.0384, 0.0593, 0.1484, 0.0692, 0.0829],
+       device='cuda:3'), in_proj_covar=tensor([0.0044, 0.0046, 0.0047, 0.0045, 0.0050, 0.0045, 0.0044, 0.0044],
+       device='cuda:3'), out_proj_covar=tensor([4.0011e-05, 4.0872e-05, 4.1125e-05, 3.8857e-05, 4.6704e-05, 4.0811e-05,
+        3.7288e-05, 4.0485e-05], device='cuda:3')
+2023-03-27 15:46:05,816 INFO [train.py:892] (3/4) Epoch 1, batch 1750, loss[loss=0.4809, simple_loss=0.4505, pruned_loss=0.2564, over 19764.00 frames. ], tot_loss[loss=0.5157, simple_loss=0.4694, pruned_loss=0.2886, over 3948255.15 frames. ], batch size: 70, lr: 4.86e-02, grad_scale: 8.0
+2023-03-27 15:47:05,138 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1790.0, num_to_drop=1, layers_to_drop={2}
+2023-03-27 15:47:13,750 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-27 15:47:20,401 INFO [train.py:892] (3/4) Epoch 1, batch 1800, loss[loss=0.4839, simple_loss=0.4622, pruned_loss=0.2524, over 19529.00 frames. ], tot_loss[loss=0.5109, simple_loss=0.4661, pruned_loss=0.2841, over 3947490.53 frames. ], batch size: 54, lr: 4.85e-02, grad_scale: 8.0
+2023-03-27 15:47:23,279 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.834e+02 6.979e+02 1.011e+03 1.306e+03 2.784e+03, threshold=2.021e+03, percent-clipped=17.0
+2023-03-27 15:47:28,270 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5193, 4.2123, 4.0783, 4.4204, 4.1223, 4.3049, 4.2979, 4.5463],
+       device='cuda:3'), covar=tensor([0.0215, 0.0317, 0.0421, 0.0199, 0.0340, 0.0208, 0.0241, 0.0247],
+       device='cuda:3'), in_proj_covar=tensor([0.0051, 0.0060, 0.0062, 0.0057, 0.0062, 0.0057, 0.0065, 0.0056],
+       device='cuda:3'), out_proj_covar=tensor([4.5044e-05, 5.8793e-05, 6.1899e-05, 5.5385e-05, 6.2508e-05, 5.8434e-05,
+        6.3131e-05, 5.8197e-05], device='cuda:3')
+2023-03-27 15:47:32,263 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1809.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 15:48:13,803 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1838.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 15:48:31,000 INFO [train.py:892] (3/4) Epoch 1, batch 1850, loss[loss=0.5027, simple_loss=0.478, pruned_loss=0.2636, over 19827.00 frames. ], tot_loss[loss=0.5052, simple_loss=0.4639, pruned_loss=0.2782, over 3947289.74 frames. ], batch size: 57, lr: 4.84e-02, grad_scale: 8.0
+2023-03-27 15:49:26,261 INFO [train.py:892] (3/4) Epoch 2, batch 0, loss[loss=0.468, simple_loss=0.4289, pruned_loss=0.2546, over 19786.00 frames. ], tot_loss[loss=0.468, simple_loss=0.4289, pruned_loss=0.2546, over 19786.00 frames. ], batch size: 193, lr: 4.75e-02, grad_scale: 8.0
+2023-03-27 15:49:26,261 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-27 15:49:35,665 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.1906, 2.5375, 2.2564, 2.9730, 2.1768, 2.3457, 2.2182, 2.2545],
+       device='cuda:3'), covar=tensor([0.1146, 0.0732, 0.3103, 0.0517, 0.1052, 0.0847, 0.0869, 0.0742],
+       device='cuda:3'), in_proj_covar=tensor([0.0033, 0.0030, 0.0031, 0.0031, 0.0034, 0.0036, 0.0039, 0.0035],
+       device='cuda:3'), out_proj_covar=tensor([2.6538e-05, 2.3124e-05, 2.6662e-05, 2.3965e-05, 2.5553e-05, 2.9377e-05,
+        3.2424e-05, 2.8234e-05], device='cuda:3')
+2023-03-27 15:49:49,016 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.4848, 0.7376, 1.4382, 0.9674, 1.2972, 1.4838, 1.4330, 1.3098],
+       device='cuda:3'), covar=tensor([0.5096, 1.5934, 0.4387, 1.3410, 0.5425, 0.3416, 0.2767, 0.3499],
+       device='cuda:3'), in_proj_covar=tensor([0.0026, 0.0049, 0.0025, 0.0052, 0.0028, 0.0032, 0.0027, 0.0027],
+       device='cuda:3'), out_proj_covar=tensor([1.9811e-05, 4.4370e-05, 1.9254e-05, 4.6820e-05, 2.2890e-05, 2.4864e-05,
+        2.0715e-05, 1.9702e-05], device='cuda:3')
+2023-03-27 15:49:52,687 INFO [train.py:926] (3/4) Epoch 2, validation: loss=0.3819, simple_loss=0.4085, pruned_loss=0.1743, over 2883724.00 frames. 
+2023-03-27 15:49:52,688 INFO [train.py:927] (3/4) Maximum memory allocated so far is 18229MB
+2023-03-27 15:49:56,678 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1857.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 15:49:56,787 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1857.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 15:50:22,549 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1870.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 15:51:16,007 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1899.0, num_to_drop=2, layers_to_drop={1, 2}
+2023-03-27 15:51:22,950 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.599e+02 7.082e+02 8.943e+02 1.129e+03 2.587e+03, threshold=1.789e+03, percent-clipped=3.0
+2023-03-27 15:51:29,934 INFO [train.py:892] (3/4) Epoch 2, batch 50, loss[loss=0.4131, simple_loss=0.3974, pruned_loss=0.2141, over 19900.00 frames. ], tot_loss[loss=0.4664, simple_loss=0.4381, pruned_loss=0.2476, over 891088.92 frames. ], batch size: 94, lr: 4.74e-02, grad_scale: 8.0
+2023-03-27 15:51:53,426 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1918.0, num_to_drop=2, layers_to_drop={0, 3}
+2023-03-27 15:51:59,865 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1922.0, num_to_drop=1, layers_to_drop={3}
+2023-03-27 15:52:16,598 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1931.0, num_to_drop=2, layers_to_drop={1, 3}
+2023-03-27 15:53:00,441 INFO [train.py:892] (3/4) Epoch 2, batch 100, loss[loss=0.4143, simple_loss=0.4123, pruned_loss=0.2077, over 19594.00 frames. ], tot_loss[loss=0.4719, simple_loss=0.4414, pruned_loss=0.2514, over 1569515.46 frames. ], batch size: 45, lr: 4.73e-02, grad_scale: 8.0
+2023-03-27 15:53:21,344 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 15:54:30,519 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.745e+02 7.981e+02 1.059e+03 1.451e+03 2.427e+03, threshold=2.118e+03, percent-clipped=7.0
+2023-03-27 15:54:36,142 INFO [train.py:892] (3/4) Epoch 2, batch 150, loss[loss=0.4199, simple_loss=0.4126, pruned_loss=0.2136, over 19910.00 frames. ], tot_loss[loss=0.4763, simple_loss=0.4447, pruned_loss=0.2541, over 2097065.24 frames. ], batch size: 53, lr: 4.72e-02, grad_scale: 8.0
+2023-03-27 15:55:54,400 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3178, 2.9779, 3.3264, 2.0872, 3.2133, 2.8091, 2.9517, 2.5438],
+       device='cuda:3'), covar=tensor([0.0286, 0.0673, 0.0260, 0.1048, 0.0317, 0.0878, 0.0372, 0.0925],
+       device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0036, 0.0035, 0.0034, 0.0032, 0.0032, 0.0032, 0.0035],
+       device='cuda:3'), out_proj_covar=tensor([2.7098e-05, 3.1050e-05, 2.9049e-05, 3.0348e-05, 2.7440e-05, 2.7391e-05,
+        2.7054e-05, 3.1112e-05], device='cuda:3')
+2023-03-27 15:56:17,767 INFO [train.py:892] (3/4) Epoch 2, batch 200, loss[loss=0.4025, simple_loss=0.396, pruned_loss=0.2046, over 19811.00 frames. ], tot_loss[loss=0.4642, simple_loss=0.4387, pruned_loss=0.2449, over 2508565.02 frames. ], batch size: 181, lr: 4.72e-02, grad_scale: 16.0
+2023-03-27 15:57:25,349 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2090.0, num_to_drop=2, layers_to_drop={0, 2}
+2023-03-27 15:57:51,094 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.526e+02 6.247e+02 7.217e+02 9.235e+02 2.365e+03, threshold=1.443e+03, percent-clipped=1.0
+2023-03-27 15:57:52,377 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.83 vs. limit=2.0
+2023-03-27 15:57:56,574 INFO [train.py:892] (3/4) Epoch 2, batch 250, loss[loss=0.4396, simple_loss=0.4254, pruned_loss=0.2269, over 19802.00 frames. ], tot_loss[loss=0.4556, simple_loss=0.4341, pruned_loss=0.2386, over 2827623.99 frames. ], batch size: 114, lr: 4.71e-02, grad_scale: 16.0
+2023-03-27 15:59:00,539 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 15:59:32,883 INFO [train.py:892] (3/4) Epoch 2, batch 300, loss[loss=0.4034, simple_loss=0.3919, pruned_loss=0.2075, over 19867.00 frames. ], tot_loss[loss=0.4524, simple_loss=0.4323, pruned_loss=0.2363, over 3076931.12 frames. ], batch size: 158, lr: 4.70e-02, grad_scale: 16.0
+2023-03-27 15:59:42,446 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0
+2023-03-27 16:00:44,822 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5958, 3.9911, 4.2704, 4.1978, 3.6718, 2.3334, 4.0377, 3.4417],
+       device='cuda:3'), covar=tensor([0.0498, 0.0318, 0.0146, 0.0232, 0.0329, 0.0780, 0.0187, 0.0238],
+       device='cuda:3'), in_proj_covar=tensor([0.0041, 0.0037, 0.0036, 0.0032, 0.0040, 0.0039, 0.0035, 0.0034],
+       device='cuda:3'), out_proj_covar=tensor([3.3026e-05, 2.8533e-05, 2.9083e-05, 2.4633e-05, 3.1269e-05, 3.2907e-05,
+        2.7889e-05, 2.6352e-05], device='cuda:3')
+2023-03-27 16:00:53,346 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:01:11,032 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.616e+02 6.231e+02 7.603e+02 9.238e+02 1.387e+03, threshold=1.521e+03, percent-clipped=0.0
+2023-03-27 16:01:18,840 INFO [train.py:892] (3/4) Epoch 2, batch 350, loss[loss=0.3921, simple_loss=0.3856, pruned_loss=0.1993, over 19878.00 frames. ], tot_loss[loss=0.4496, simple_loss=0.4318, pruned_loss=0.2338, over 3269777.01 frames. ], batch size: 136, lr: 4.69e-02, grad_scale: 16.0
+2023-03-27 16:01:23,959 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.07 vs. limit=2.0
+2023-03-27 16:01:32,866 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2213.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:01:49,481 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2222.0, num_to_drop=2, layers_to_drop={0, 1}
+2023-03-27 16:01:56,902 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2226.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:02:54,453 INFO [train.py:892] (3/4) Epoch 2, batch 400, loss[loss=0.4439, simple_loss=0.4204, pruned_loss=0.2337, over 19823.00 frames. ], tot_loss[loss=0.4439, simple_loss=0.4283, pruned_loss=0.2298, over 3420226.42 frames. ], batch size: 204, lr: 4.68e-02, grad_scale: 16.0
+2023-03-27 16:03:15,762 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2267.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 16:03:22,343 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:04:11,273 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0
+2023-03-27 16:04:26,665 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.159e+02 7.202e+02 8.951e+02 1.212e+03 2.125e+03, threshold=1.790e+03, percent-clipped=11.0
+2023-03-27 16:04:31,839 INFO [train.py:892] (3/4) Epoch 2, batch 450, loss[loss=0.4069, simple_loss=0.4077, pruned_loss=0.203, over 19849.00 frames. ], tot_loss[loss=0.4415, simple_loss=0.4281, pruned_loss=0.2275, over 3535916.60 frames. ], batch size: 78, lr: 4.67e-02, grad_scale: 16.0
+2023-03-27 16:04:52,556 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:04:54,292 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1030, 2.7921, 3.1392, 3.4135, 2.9023, 2.8782, 2.8339, 2.8657],
+       device='cuda:3'), covar=tensor([0.0242, 0.0413, 0.0260, 0.0135, 0.0308, 0.0243, 0.0321, 0.0500],
+       device='cuda:3'), in_proj_covar=tensor([0.0032, 0.0035, 0.0034, 0.0027, 0.0033, 0.0035, 0.0036, 0.0036],
+       device='cuda:3'), out_proj_covar=tensor([2.4629e-05, 2.9540e-05, 2.5618e-05, 2.0359e-05, 2.6259e-05, 2.7004e-05,
+        2.9100e-05, 2.9080e-05], device='cuda:3')
+2023-03-27 16:05:40,814 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2340.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 16:06:13,203 INFO [train.py:892] (3/4) Epoch 2, batch 500, loss[loss=0.3941, simple_loss=0.3974, pruned_loss=0.1954, over 19668.00 frames. ], tot_loss[loss=0.4347, simple_loss=0.424, pruned_loss=0.2227, over 3626956.02 frames. ], batch size: 64, lr: 4.66e-02, grad_scale: 16.0
+2023-03-27 16:06:42,046 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0
+2023-03-27 16:06:56,465 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.13 vs. limit=2.0
+2023-03-27 16:07:09,839 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7921, 3.8841, 4.5465, 4.2275, 3.9752, 4.5664, 4.4551, 4.4004],
+       device='cuda:3'), covar=tensor([0.0128, 0.0328, 0.0098, 0.0163, 0.0176, 0.0074, 0.0112, 0.0225],
+       device='cuda:3'), in_proj_covar=tensor([0.0027, 0.0029, 0.0026, 0.0027, 0.0025, 0.0024, 0.0025, 0.0029],
+       device='cuda:3'), out_proj_covar=tensor([2.6857e-05, 2.9056e-05, 2.5483e-05, 2.5903e-05, 2.3394e-05, 2.3996e-05,
+        2.2156e-05, 2.9005e-05], device='cuda:3')
+2023-03-27 16:07:43,928 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2401.0, num_to_drop=2, layers_to_drop={2, 3}
+2023-03-27 16:07:46,784 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.849e+02 6.111e+02 7.665e+02 9.809e+02 1.525e+03, threshold=1.533e+03, percent-clipped=0.0
+2023-03-27 16:07:51,947 INFO [train.py:892] (3/4) Epoch 2, batch 550, loss[loss=0.4228, simple_loss=0.4073, pruned_loss=0.2191, over 19856.00 frames. ], tot_loss[loss=0.4318, simple_loss=0.4218, pruned_loss=0.221, over 3699005.67 frames. ], batch size: 165, lr: 4.65e-02, grad_scale: 16.0
+2023-03-27 16:08:00,308 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8661, 4.8564, 5.1288, 4.6521, 4.8569, 5.1126, 4.7903, 5.3181],
+       device='cuda:3'), covar=tensor([0.0555, 0.0228, 0.0230, 0.0274, 0.0177, 0.0161, 0.0179, 0.0201],
+       device='cuda:3'), in_proj_covar=tensor([0.0063, 0.0066, 0.0059, 0.0061, 0.0056, 0.0059, 0.0057, 0.0052],
+       device='cuda:3'), out_proj_covar=tensor([6.3262e-05, 7.1101e-05, 6.8406e-05, 6.4986e-05, 5.6210e-05, 6.7763e-05,
+        5.8540e-05, 5.5500e-05], device='cuda:3')
+2023-03-27 16:09:31,343 INFO [train.py:892] (3/4) Epoch 2, batch 600, loss[loss=0.4209, simple_loss=0.4036, pruned_loss=0.2191, over 19805.00 frames. ], tot_loss[loss=0.4317, simple_loss=0.4219, pruned_loss=0.2207, over 3754484.70 frames. ], batch size: 132, lr: 4.64e-02, grad_scale: 16.0
+2023-03-27 16:10:35,568 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4168, 2.5781, 2.4406, 3.0655, 2.2104, 2.4856, 2.3020, 2.3822],
+       device='cuda:3'), covar=tensor([0.0436, 0.0252, 0.0827, 0.0118, 0.0329, 0.0304, 0.0311, 0.0323],
+       device='cuda:3'), in_proj_covar=tensor([0.0027, 0.0023, 0.0034, 0.0023, 0.0027, 0.0028, 0.0029, 0.0029],
+       device='cuda:3'), out_proj_covar=tensor([2.2109e-05, 1.8438e-05, 3.1263e-05, 1.7908e-05, 2.2495e-05, 2.4394e-05,
+        2.4173e-05, 2.4099e-05], device='cuda:3')
+2023-03-27 16:10:49,215 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2494.0, num_to_drop=1, layers_to_drop={2}
+2023-03-27 16:11:07,763 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.512e+02 7.276e+02 8.474e+02 1.052e+03 1.837e+03, threshold=1.695e+03, percent-clipped=3.0
+2023-03-27 16:11:13,425 INFO [train.py:892] (3/4) Epoch 2, batch 650, loss[loss=0.4214, simple_loss=0.4076, pruned_loss=0.2176, over 19592.00 frames. ], tot_loss[loss=0.4258, simple_loss=0.4178, pruned_loss=0.217, over 3799360.64 frames. ], batch size: 45, lr: 4.64e-02, grad_scale: 16.0
+2023-03-27 16:11:23,880 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=2.01 vs. limit=2.0
+2023-03-27 16:11:27,290 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:11:34,888 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0
+2023-03-27 16:11:38,205 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7552, 4.8735, 4.9764, 5.0142, 3.5620, 4.4467, 4.3929, 2.6598],
+       device='cuda:3'), covar=tensor([0.0116, 0.0189, 0.0136, 0.0054, 0.1418, 0.0117, 0.0235, 0.1793],
+       device='cuda:3'), in_proj_covar=tensor([0.0036, 0.0032, 0.0041, 0.0034, 0.0072, 0.0036, 0.0044, 0.0079],
+       device='cuda:3'), out_proj_covar=tensor([2.7818e-05, 2.6339e-05, 3.1898e-05, 2.4860e-05, 6.4600e-05, 2.7744e-05,
+        3.6618e-05, 6.8934e-05], device='cuda:3')
+2023-03-27 16:11:51,007 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:12:26,532 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:12:51,166 INFO [train.py:892] (3/4) Epoch 2, batch 700, loss[loss=0.3559, simple_loss=0.3675, pruned_loss=0.1722, over 19851.00 frames. ], tot_loss[loss=0.4231, simple_loss=0.4166, pruned_loss=0.2148, over 3833520.97 frames. ], batch size: 78, lr: 4.63e-02, grad_scale: 16.0
+2023-03-27 16:13:01,235 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2561.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:13:27,673 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:14:23,672 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.137e+02 6.142e+02 8.213e+02 1.055e+03 2.684e+03, threshold=1.643e+03, percent-clipped=4.0
+2023-03-27 16:14:30,632 INFO [train.py:892] (3/4) Epoch 2, batch 750, loss[loss=0.4312, simple_loss=0.4256, pruned_loss=0.2184, over 19817.00 frames. ], tot_loss[loss=0.4188, simple_loss=0.4142, pruned_loss=0.2117, over 3859455.43 frames. ], batch size: 202, lr: 4.62e-02, grad_scale: 16.0
+2023-03-27 16:14:43,238 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.2311, 5.2829, 5.6409, 5.6753, 5.7192, 5.2497, 5.3494, 5.2086],
+       device='cuda:3'), covar=tensor([0.0635, 0.0543, 0.0647, 0.0319, 0.0457, 0.0730, 0.0542, 0.1154],
+       device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0091, 0.0133, 0.0097, 0.0112, 0.0105, 0.0114, 0.0153],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002],
+       device='cuda:3')
+2023-03-27 16:16:13,081 INFO [train.py:892] (3/4) Epoch 2, batch 800, loss[loss=0.444, simple_loss=0.4557, pruned_loss=0.2162, over 19855.00 frames. ], tot_loss[loss=0.4185, simple_loss=0.4147, pruned_loss=0.2111, over 3879508.48 frames. ], batch size: 56, lr: 4.61e-02, grad_scale: 16.0
+2023-03-27 16:16:51,514 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.54 vs. limit=5.0
+2023-03-27 16:17:11,704 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2685.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 16:17:33,333 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2696.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 16:17:45,234 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.299e+02 7.441e+02 9.827e+02 1.217e+03 2.716e+03, threshold=1.965e+03, percent-clipped=6.0
+2023-03-27 16:17:51,072 INFO [train.py:892] (3/4) Epoch 2, batch 850, loss[loss=0.4682, simple_loss=0.4533, pruned_loss=0.2416, over 19653.00 frames. ], tot_loss[loss=0.4155, simple_loss=0.4126, pruned_loss=0.2093, over 3896811.79 frames. ], batch size: 299, lr: 4.60e-02, grad_scale: 16.0
+2023-03-27 16:19:12,056 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2746.0, num_to_drop=2, layers_to_drop={1, 2}
+2023-03-27 16:19:32,079 INFO [train.py:892] (3/4) Epoch 2, batch 900, loss[loss=0.3682, simple_loss=0.3665, pruned_loss=0.185, over 19767.00 frames. ], tot_loss[loss=0.4114, simple_loss=0.4098, pruned_loss=0.2065, over 3909608.61 frames. ], batch size: 152, lr: 4.59e-02, grad_scale: 16.0
+2023-03-27 16:20:35,588 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.2841, 4.5586, 5.1128, 4.6402, 5.1614, 4.1950, 4.6279, 4.9632],
+       device='cuda:3'), covar=tensor([0.0147, 0.0176, 0.0118, 0.0181, 0.0131, 0.0296, 0.0271, 0.0147],
+       device='cuda:3'), in_proj_covar=tensor([0.0041, 0.0041, 0.0044, 0.0049, 0.0047, 0.0055, 0.0046, 0.0041],
+       device='cuda:3'), out_proj_covar=tensor([4.7403e-05, 4.8745e-05, 4.9450e-05, 5.6105e-05, 5.8238e-05, 6.2881e-05,
+        5.2681e-05, 4.6412e-05], device='cuda:3')
+2023-03-27 16:21:02,190 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-27 16:21:04,242 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.961e+02 6.224e+02 7.989e+02 1.015e+03 2.345e+03, threshold=1.598e+03, percent-clipped=4.0
+2023-03-27 16:21:10,145 INFO [train.py:892] (3/4) Epoch 2, batch 950, loss[loss=0.5639, simple_loss=0.5212, pruned_loss=0.3033, over 19483.00 frames. ], tot_loss[loss=0.4121, simple_loss=0.4107, pruned_loss=0.2068, over 3918185.06 frames. ], batch size: 396, lr: 4.58e-02, grad_scale: 16.0
+2023-03-27 16:21:18,325 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3250, 4.4764, 4.4833, 4.5561, 4.4222, 4.5234, 4.0433, 4.0568],
+       device='cuda:3'), covar=tensor([0.0323, 0.0238, 0.0429, 0.0301, 0.0357, 0.0357, 0.0393, 0.0556],
+       device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0073, 0.0087, 0.0078, 0.0082, 0.0069, 0.0090, 0.0112],
+       device='cuda:3'), out_proj_covar=tensor([8.6071e-05, 8.3058e-05, 9.7536e-05, 8.9516e-05, 9.0709e-05, 7.6796e-05,
+        9.8915e-05, 1.3023e-04], device='cuda:3')
+2023-03-27 16:21:45,495 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2824.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 16:22:47,335 INFO [train.py:892] (3/4) Epoch 2, batch 1000, loss[loss=0.3821, simple_loss=0.3995, pruned_loss=0.1823, over 19655.00 frames. ], tot_loss[loss=0.4102, simple_loss=0.4095, pruned_loss=0.2055, over 3926791.53 frames. ], batch size: 72, lr: 4.57e-02, grad_scale: 16.0
+2023-03-27 16:23:26,877 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.2738, 3.5080, 4.6280, 4.4500, 4.3198, 5.0108, 4.6396, 4.8115],
+       device='cuda:3'), covar=tensor([0.0102, 0.0597, 0.0113, 0.0205, 0.0161, 0.0076, 0.0113, 0.0136],
+       device='cuda:3'), in_proj_covar=tensor([0.0028, 0.0031, 0.0026, 0.0028, 0.0025, 0.0023, 0.0025, 0.0027],
+       device='cuda:3'), out_proj_covar=tensor([3.3221e-05, 3.9120e-05, 2.9875e-05, 3.3386e-05, 2.7853e-05, 2.9110e-05,
+        2.8030e-05, 3.2263e-05], device='cuda:3')
+2023-03-27 16:23:47,472 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2885.0, num_to_drop=2, layers_to_drop={0, 2}
+2023-03-27 16:24:08,101 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0
+2023-03-27 16:24:23,443 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.209e+02 6.834e+02 8.062e+02 1.027e+03 2.036e+03, threshold=1.612e+03, percent-clipped=4.0
+2023-03-27 16:24:29,000 INFO [train.py:892] (3/4) Epoch 2, batch 1050, loss[loss=0.3572, simple_loss=0.3684, pruned_loss=0.1729, over 19740.00 frames. ], tot_loss[loss=0.4087, simple_loss=0.4086, pruned_loss=0.2044, over 3931923.62 frames. ], batch size: 77, lr: 4.56e-02, grad_scale: 16.0
+2023-03-27 16:25:02,683 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2123, 2.2826, 4.1330, 3.8910, 3.4660, 3.7923, 4.0715, 3.4098],
+       device='cuda:3'), covar=tensor([0.0208, 0.1204, 0.0213, 0.0276, 0.0578, 0.0180, 0.0322, 0.0992],
+       device='cuda:3'), in_proj_covar=tensor([0.0037, 0.0056, 0.0038, 0.0033, 0.0036, 0.0040, 0.0040, 0.0042],
+       device='cuda:3'), out_proj_covar=tensor([3.2324e-05, 5.8891e-05, 3.2660e-05, 2.7489e-05, 3.4416e-05, 3.5928e-05,
+        3.8486e-05, 4.1790e-05], device='cuda:3')
+2023-03-27 16:25:31,346 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2937.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 16:26:09,032 INFO [train.py:892] (3/4) Epoch 2, batch 1100, loss[loss=0.3852, simple_loss=0.4077, pruned_loss=0.1814, over 19575.00 frames. ], tot_loss[loss=0.4061, simple_loss=0.4072, pruned_loss=0.2025, over 3935359.20 frames. ], batch size: 53, lr: 4.55e-02, grad_scale: 16.0
+2023-03-27 16:26:24,858 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.17 vs. limit=5.0
+2023-03-27 16:26:43,824 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0045, 5.0889, 5.4301, 5.2539, 5.5254, 4.9519, 5.1943, 5.0349],
+       device='cuda:3'), covar=tensor([0.0743, 0.0645, 0.0774, 0.0388, 0.0480, 0.0878, 0.0794, 0.1496],
+       device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0096, 0.0150, 0.0099, 0.0117, 0.0113, 0.0121, 0.0166],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002],
+       device='cuda:3')
+2023-03-27 16:27:09,427 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.41 vs. limit=5.0
+2023-03-27 16:27:28,917 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2996.0, num_to_drop=1, layers_to_drop={2}
+2023-03-27 16:27:32,696 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2998.0, num_to_drop=2, layers_to_drop={0, 3}
+2023-03-27 16:27:41,289 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.835e+02 7.115e+02 9.003e+02 1.112e+03 2.091e+03, threshold=1.801e+03, percent-clipped=6.0
+2023-03-27 16:27:46,924 INFO [train.py:892] (3/4) Epoch 2, batch 1150, loss[loss=0.3845, simple_loss=0.3836, pruned_loss=0.1927, over 19768.00 frames. ], tot_loss[loss=0.4009, simple_loss=0.4034, pruned_loss=0.1992, over 3939625.66 frames. ], batch size: 155, lr: 4.54e-02, grad_scale: 16.0
+2023-03-27 16:28:07,194 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.2819, 5.6800, 5.8922, 5.8632, 5.9205, 5.4388, 5.6120, 5.4443],
+       device='cuda:3'), covar=tensor([0.0764, 0.0579, 0.0697, 0.0354, 0.0465, 0.0706, 0.0642, 0.1431],
+       device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0100, 0.0156, 0.0103, 0.0124, 0.0116, 0.0127, 0.0176],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 16:28:17,957 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3019.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 16:29:00,276 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:29:05,930 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3044.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 16:29:22,137 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0985, 3.6680, 3.9385, 3.7185, 3.9803, 3.3368, 3.5906, 3.9450],
+       device='cuda:3'), covar=tensor([0.0139, 0.0195, 0.0142, 0.0176, 0.0171, 0.0393, 0.0348, 0.0153],
+       device='cuda:3'), in_proj_covar=tensor([0.0041, 0.0043, 0.0045, 0.0049, 0.0049, 0.0057, 0.0051, 0.0044],
+       device='cuda:3'), out_proj_covar=tensor([5.0133e-05, 5.7036e-05, 5.5892e-05, 6.1558e-05, 6.7079e-05, 7.2034e-05,
+        6.4820e-05, 5.3519e-05], device='cuda:3')
+2023-03-27 16:29:26,153 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4009, 3.9541, 4.9758, 4.5155, 4.5683, 4.9723, 4.5374, 5.1557],
+       device='cuda:3'), covar=tensor([0.0087, 0.0372, 0.0101, 0.0331, 0.0113, 0.0086, 0.0125, 0.0081],
+       device='cuda:3'), in_proj_covar=tensor([0.0029, 0.0032, 0.0027, 0.0032, 0.0026, 0.0024, 0.0027, 0.0028],
+       device='cuda:3'), out_proj_covar=tensor([3.7336e-05, 4.4119e-05, 3.3391e-05, 4.2470e-05, 3.1104e-05, 3.2311e-05,
+        3.2271e-05, 3.5675e-05], device='cuda:3')
+2023-03-27 16:29:30,578 INFO [train.py:892] (3/4) Epoch 2, batch 1200, loss[loss=0.6468, simple_loss=0.5786, pruned_loss=0.3575, over 19211.00 frames. ], tot_loss[loss=0.4014, simple_loss=0.4037, pruned_loss=0.1995, over 3942114.24 frames. ], batch size: 452, lr: 4.53e-02, grad_scale: 16.0
+2023-03-27 16:30:18,071 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3080.0, num_to_drop=2, layers_to_drop={1, 2}
+2023-03-27 16:31:04,257 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.272e+02 6.636e+02 8.301e+02 1.092e+03 3.055e+03, threshold=1.660e+03, percent-clipped=1.0
+2023-03-27 16:31:10,207 INFO [train.py:892] (3/4) Epoch 2, batch 1250, loss[loss=0.3752, simple_loss=0.3861, pruned_loss=0.1821, over 19759.00 frames. ], tot_loss[loss=0.3991, simple_loss=0.4028, pruned_loss=0.1977, over 3943053.46 frames. ], batch size: 113, lr: 4.52e-02, grad_scale: 16.0
+2023-03-27 16:32:23,456 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.8030, 2.1050, 2.4310, 1.7004, 2.4191, 2.0293, 2.3275, 1.8455],
+       device='cuda:3'), covar=tensor([0.0204, 0.0365, 0.0192, 0.0487, 0.0196, 0.0521, 0.0217, 0.0929],
+       device='cuda:3'), in_proj_covar=tensor([0.0020, 0.0025, 0.0024, 0.0026, 0.0023, 0.0022, 0.0021, 0.0022],
+       device='cuda:3'), out_proj_covar=tensor([2.0556e-05, 2.4882e-05, 2.3696e-05, 2.6713e-05, 2.2938e-05, 2.2478e-05,
+        2.1906e-05, 2.3864e-05], device='cuda:3')
+2023-03-27 16:32:23,490 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9126, 4.5415, 4.2236, 4.4291, 2.8476, 4.2932, 4.2089, 2.4450],
+       device='cuda:3'), covar=tensor([0.0199, 0.0253, 0.0253, 0.0110, 0.1924, 0.0124, 0.0229, 0.2016],
+       device='cuda:3'), in_proj_covar=tensor([0.0048, 0.0050, 0.0055, 0.0044, 0.0109, 0.0048, 0.0056, 0.0115],
+       device='cuda:3'), out_proj_covar=tensor([3.8588e-05, 4.4989e-05, 4.5924e-05, 3.2633e-05, 9.3137e-05, 3.8160e-05,
+        4.6517e-05, 9.8414e-05], device='cuda:3')
+2023-03-27 16:32:49,634 INFO [train.py:892] (3/4) Epoch 2, batch 1300, loss[loss=0.3876, simple_loss=0.4068, pruned_loss=0.1842, over 19618.00 frames. ], tot_loss[loss=0.3954, simple_loss=0.4003, pruned_loss=0.1953, over 3945277.29 frames. ], batch size: 52, lr: 4.51e-02, grad_scale: 16.0
+2023-03-27 16:33:01,252 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0
+2023-03-27 16:33:37,200 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3180.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:33:49,223 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0
+2023-03-27 16:34:23,392 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.299e+02 7.276e+02 8.986e+02 1.118e+03 2.110e+03, threshold=1.797e+03, percent-clipped=4.0
+2023-03-27 16:34:28,595 INFO [train.py:892] (3/4) Epoch 2, batch 1350, loss[loss=0.3675, simple_loss=0.3632, pruned_loss=0.1859, over 19866.00 frames. ], tot_loss[loss=0.3965, simple_loss=0.4014, pruned_loss=0.1958, over 3945098.39 frames. ], batch size: 154, lr: 4.50e-02, grad_scale: 16.0
+2023-03-27 16:35:19,999 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.7203, 1.6974, 2.7241, 1.4582, 2.5991, 2.0137, 1.7274, 2.4983],
+       device='cuda:3'), covar=tensor([0.0205, 0.0593, 0.0158, 0.0719, 0.0134, 0.0435, 0.0277, 0.0176],
+       device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0024, 0.0021, 0.0025, 0.0022, 0.0021, 0.0019, 0.0020],
+       device='cuda:3'), out_proj_covar=tensor([1.9074e-05, 2.4491e-05, 2.1201e-05, 2.5812e-05, 2.1640e-05, 2.2153e-05,
+        2.0413e-05, 2.1782e-05], device='cuda:3')
+2023-03-27 16:36:05,210 INFO [train.py:892] (3/4) Epoch 2, batch 1400, loss[loss=0.3896, simple_loss=0.3999, pruned_loss=0.1897, over 19849.00 frames. ], tot_loss[loss=0.3945, simple_loss=0.3995, pruned_loss=0.1948, over 3946000.25 frames. ], batch size: 81, lr: 4.49e-02, grad_scale: 16.0
+2023-03-27 16:36:38,195 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9911, 3.2111, 3.8462, 3.6156, 3.2657, 2.3677, 3.2492, 3.1439],
+       device='cuda:3'), covar=tensor([0.0890, 0.0335, 0.0411, 0.0285, 0.0522, 0.0718, 0.0345, 0.0522],
+       device='cuda:3'), in_proj_covar=tensor([0.0035, 0.0027, 0.0030, 0.0029, 0.0034, 0.0028, 0.0028, 0.0028],
+       device='cuda:3'), out_proj_covar=tensor([3.7876e-05, 2.8043e-05, 3.1989e-05, 2.8992e-05, 3.5789e-05, 3.1189e-05,
+        2.8563e-05, 2.9364e-05], device='cuda:3')
+2023-03-27 16:37:09,969 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3288.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 16:37:18,966 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:37:38,105 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.923e+02 7.473e+02 1.007e+03 1.301e+03 1.953e+03, threshold=2.013e+03, percent-clipped=1.0
+2023-03-27 16:37:43,355 INFO [train.py:892] (3/4) Epoch 2, batch 1450, loss[loss=0.3686, simple_loss=0.3842, pruned_loss=0.1765, over 19567.00 frames. ], tot_loss[loss=0.3931, simple_loss=0.3988, pruned_loss=0.1937, over 3946324.47 frames. ], batch size: 42, lr: 4.48e-02, grad_scale: 16.0
+2023-03-27 16:38:55,739 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:39:13,578 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3349.0, num_to_drop=2, layers_to_drop={1, 3}
+2023-03-27 16:39:26,357 INFO [train.py:892] (3/4) Epoch 2, batch 1500, loss[loss=0.3415, simple_loss=0.348, pruned_loss=0.1675, over 19825.00 frames. ], tot_loss[loss=0.3933, simple_loss=0.3996, pruned_loss=0.1935, over 3944781.13 frames. ], batch size: 127, lr: 4.47e-02, grad_scale: 16.0
+2023-03-27 16:40:03,978 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3375.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:40:34,100 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:40:59,891 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.490e+02 6.824e+02 8.232e+02 9.819e+02 1.797e+03, threshold=1.646e+03, percent-clipped=0.0
+2023-03-27 16:41:05,677 INFO [train.py:892] (3/4) Epoch 2, batch 1550, loss[loss=0.3718, simple_loss=0.384, pruned_loss=0.1798, over 19840.00 frames. ], tot_loss[loss=0.389, simple_loss=0.3968, pruned_loss=0.1906, over 3945266.90 frames. ], batch size: 160, lr: 4.46e-02, grad_scale: 16.0
+2023-03-27 16:41:54,346 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.9273, 2.0484, 2.1548, 1.3247, 1.9008, 2.0452, 2.1682, 1.7911],
+       device='cuda:3'), covar=tensor([0.0186, 0.0268, 0.0187, 0.0505, 0.0393, 0.0234, 0.0213, 0.0441],
+       device='cuda:3'), in_proj_covar=tensor([0.0019, 0.0022, 0.0020, 0.0025, 0.0023, 0.0021, 0.0019, 0.0021],
+       device='cuda:3'), out_proj_covar=tensor([2.0278e-05, 2.3711e-05, 2.1283e-05, 2.7371e-05, 2.3898e-05, 2.2712e-05,
+        2.0973e-05, 2.2949e-05], device='cuda:3')
+2023-03-27 16:42:03,871 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.07 vs. limit=2.0
+2023-03-27 16:42:33,518 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.87 vs. limit=2.0
+2023-03-27 16:42:46,604 INFO [train.py:892] (3/4) Epoch 2, batch 1600, loss[loss=0.522, simple_loss=0.4948, pruned_loss=0.2746, over 19595.00 frames. ], tot_loss[loss=0.3878, simple_loss=0.3958, pruned_loss=0.1899, over 3946940.48 frames. ], batch size: 367, lr: 4.45e-02, grad_scale: 16.0
+2023-03-27 16:43:33,870 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-27 16:43:35,475 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3480.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:44:16,601 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8539, 2.4830, 1.9795, 2.4842, 2.8104, 1.4697, 2.7112, 2.6197],
+       device='cuda:3'), covar=tensor([0.0668, 0.1174, 0.3681, 0.0476, 0.0334, 0.5023, 0.0902, 0.0462],
+       device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0060, 0.0119, 0.0040, 0.0041, 0.0125, 0.0066, 0.0052],
+       device='cuda:3'), out_proj_covar=tensor([4.7642e-05, 5.5741e-05, 1.0606e-04, 3.3642e-05, 3.2441e-05, 1.0913e-04,
+        5.7167e-05, 3.9904e-05], device='cuda:3')
+2023-03-27 16:44:24,962 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.425e+02 7.682e+02 9.382e+02 1.262e+03 3.177e+03, threshold=1.876e+03, percent-clipped=11.0
+2023-03-27 16:44:28,481 INFO [train.py:892] (3/4) Epoch 2, batch 1650, loss[loss=0.3962, simple_loss=0.3904, pruned_loss=0.201, over 19746.00 frames. ], tot_loss[loss=0.3852, simple_loss=0.3939, pruned_loss=0.1882, over 3947780.26 frames. ], batch size: 134, lr: 4.44e-02, grad_scale: 8.0
+2023-03-27 16:44:58,421 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.84 vs. limit=2.0
+2023-03-27 16:45:11,987 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:45:50,178 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3546.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 16:45:53,585 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9375, 3.1070, 2.5763, 3.8106, 3.2433, 3.5484, 3.4211, 3.0992],
+       device='cuda:3'), covar=tensor([0.0681, 0.0401, 0.2121, 0.0341, 0.0472, 0.0643, 0.0359, 0.0273],
+       device='cuda:3'), in_proj_covar=tensor([0.0041, 0.0035, 0.0069, 0.0034, 0.0037, 0.0035, 0.0035, 0.0033],
+       device='cuda:3'), out_proj_covar=tensor([4.2799e-05, 3.6609e-05, 8.8467e-05, 3.6552e-05, 3.8614e-05, 4.1031e-05,
+        3.7415e-05, 3.5524e-05], device='cuda:3')
+2023-03-27 16:46:07,497 INFO [train.py:892] (3/4) Epoch 2, batch 1700, loss[loss=0.3912, simple_loss=0.394, pruned_loss=0.1942, over 19858.00 frames. ], tot_loss[loss=0.3853, simple_loss=0.3942, pruned_loss=0.1882, over 3948916.24 frames. ], batch size: 157, lr: 4.43e-02, grad_scale: 8.0
+2023-03-27 16:47:22,729 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:47:42,005 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 5.530e+02 9.023e+02 1.103e+03 1.410e+03 2.321e+03, threshold=2.205e+03, percent-clipped=10.0
+2023-03-27 16:47:45,749 INFO [train.py:892] (3/4) Epoch 2, batch 1750, loss[loss=0.364, simple_loss=0.3679, pruned_loss=0.1801, over 19794.00 frames. ], tot_loss[loss=0.3855, simple_loss=0.3942, pruned_loss=0.1884, over 3947865.47 frames. ], batch size: 162, lr: 4.42e-02, grad_scale: 8.0
+2023-03-27 16:47:48,514 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3607.0, num_to_drop=2, layers_to_drop={1, 2}
+2023-03-27 16:47:57,407 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0
+2023-03-27 16:48:50,102 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:48:55,288 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3644.0, num_to_drop=1, layers_to_drop={2}
+2023-03-27 16:49:15,987 INFO [train.py:892] (3/4) Epoch 2, batch 1800, loss[loss=0.331, simple_loss=0.3459, pruned_loss=0.158, over 19763.00 frames. ], tot_loss[loss=0.3873, simple_loss=0.3956, pruned_loss=0.1895, over 3945353.18 frames. ], batch size: 102, lr: 4.41e-02, grad_scale: 8.0
+2023-03-27 16:49:49,890 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3675.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:49:51,502 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3676.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 16:50:37,051 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.302e+02 8.227e+02 1.028e+03 1.238e+03 2.427e+03, threshold=2.056e+03, percent-clipped=3.0
+2023-03-27 16:50:40,376 INFO [train.py:892] (3/4) Epoch 2, batch 1850, loss[loss=0.3425, simple_loss=0.3714, pruned_loss=0.1568, over 19832.00 frames. ], tot_loss[loss=0.3849, simple_loss=0.3952, pruned_loss=0.1873, over 3945951.63 frames. ], batch size: 57, lr: 4.39e-02, grad_scale: 8.0
+2023-03-27 16:51:36,783 INFO [train.py:892] (3/4) Epoch 3, batch 0, loss[loss=0.3282, simple_loss=0.3558, pruned_loss=0.1503, over 19845.00 frames. ], tot_loss[loss=0.3282, simple_loss=0.3558, pruned_loss=0.1503, over 19845.00 frames. ], batch size: 81, lr: 4.17e-02, grad_scale: 8.0
+2023-03-27 16:51:36,784 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-27 16:52:02,988 INFO [train.py:926] (3/4) Epoch 3, validation: loss=0.2594, simple_loss=0.3267, pruned_loss=0.09605, over 2883724.00 frames. 
+2023-03-27 16:52:02,989 INFO [train.py:927] (3/4) Maximum memory allocated so far is 20449MB
+2023-03-27 16:52:30,237 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3723.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 16:52:58,013 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3737.0, num_to_drop=2, layers_to_drop={2, 3}
+2023-03-27 16:53:13,777 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-27 16:53:47,373 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7472, 4.1864, 4.6941, 4.2479, 4.7454, 3.5677, 3.8664, 4.3958],
+       device='cuda:3'), covar=tensor([0.0129, 0.0178, 0.0093, 0.0154, 0.0125, 0.0424, 0.0755, 0.0189],
+       device='cuda:3'), in_proj_covar=tensor([0.0044, 0.0048, 0.0052, 0.0056, 0.0055, 0.0066, 0.0073, 0.0050],
+       device='cuda:3'), out_proj_covar=tensor([6.3975e-05, 7.9338e-05, 7.4855e-05, 8.4068e-05, 9.0885e-05, 1.0242e-04,
+        1.1805e-04, 7.4657e-05], device='cuda:3')
+2023-03-27 16:53:50,416 INFO [train.py:892] (3/4) Epoch 3, batch 50, loss[loss=0.3181, simple_loss=0.3482, pruned_loss=0.144, over 19902.00 frames. ], tot_loss[loss=0.3558, simple_loss=0.3734, pruned_loss=0.1691, over 891939.29 frames. ], batch size: 94, lr: 4.16e-02, grad_scale: 8.0
+2023-03-27 16:54:08,837 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-27 16:54:27,232 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2697, 3.0935, 2.1785, 3.2969, 3.2028, 1.7976, 3.1646, 2.6511],
+       device='cuda:3'), covar=tensor([0.0546, 0.0662, 0.2749, 0.0172, 0.0202, 0.3033, 0.0445, 0.0417],
+       device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0062, 0.0114, 0.0039, 0.0039, 0.0116, 0.0063, 0.0052],
+       device='cuda:3'), out_proj_covar=tensor([4.7938e-05, 5.7746e-05, 1.0337e-04, 3.4056e-05, 3.2408e-05, 1.0088e-04,
+        5.5273e-05, 4.0368e-05], device='cuda:3')
+2023-03-27 16:55:16,582 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.227e+02 7.628e+02 9.313e+02 1.232e+03 3.200e+03, threshold=1.863e+03, percent-clipped=4.0
+2023-03-27 16:55:31,509 INFO [train.py:892] (3/4) Epoch 3, batch 100, loss[loss=0.3727, simple_loss=0.3842, pruned_loss=0.1806, over 19768.00 frames. ], tot_loss[loss=0.3641, simple_loss=0.3795, pruned_loss=0.1743, over 1569919.11 frames. ], batch size: 247, lr: 4.15e-02, grad_scale: 8.0
+2023-03-27 16:55:58,103 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.3092, 1.9590, 1.9623, 2.2544, 1.7449, 1.7225, 1.9611, 2.0720],
+       device='cuda:3'), covar=tensor([0.0335, 0.0576, 0.0490, 0.0183, 0.0598, 0.0554, 0.0466, 0.0580],
+       device='cuda:3'), in_proj_covar=tensor([0.0027, 0.0031, 0.0032, 0.0027, 0.0030, 0.0031, 0.0031, 0.0031],
+       device='cuda:3'), out_proj_covar=tensor([3.5352e-05, 3.7379e-05, 4.1582e-05, 3.3804e-05, 3.9726e-05, 4.0664e-05,
+        4.0425e-05, 4.1496e-05], device='cuda:3')
+2023-03-27 16:57:05,637 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.27 vs. limit=5.0
+2023-03-27 16:57:14,244 INFO [train.py:892] (3/4) Epoch 3, batch 150, loss[loss=0.3734, simple_loss=0.3868, pruned_loss=0.18, over 19780.00 frames. ], tot_loss[loss=0.3692, simple_loss=0.3844, pruned_loss=0.177, over 2092586.59 frames. ], batch size: 65, lr: 4.14e-02, grad_scale: 8.0
+2023-03-27 16:57:38,449 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6373, 3.3850, 2.5171, 3.6744, 3.6324, 1.9574, 3.3286, 3.2667],
+       device='cuda:3'), covar=tensor([0.0356, 0.0435, 0.1992, 0.0083, 0.0129, 0.2281, 0.0376, 0.0265],
+       device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0062, 0.0113, 0.0037, 0.0039, 0.0114, 0.0065, 0.0052],
+       device='cuda:3'), out_proj_covar=tensor([4.9090e-05, 5.8662e-05, 1.0296e-04, 3.3160e-05, 3.2426e-05, 9.9558e-05,
+        5.7364e-05, 4.1299e-05], device='cuda:3')
+2023-03-27 16:58:41,880 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3902.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 16:58:44,801 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.184e+02 7.831e+02 9.821e+02 1.147e+03 2.106e+03, threshold=1.964e+03, percent-clipped=1.0
+2023-03-27 16:58:59,048 INFO [train.py:892] (3/4) Epoch 3, batch 200, loss[loss=0.4008, simple_loss=0.4024, pruned_loss=0.1996, over 19738.00 frames. ], tot_loss[loss=0.365, simple_loss=0.3809, pruned_loss=0.1745, over 2505672.53 frames. ], batch size: 291, lr: 4.13e-02, grad_scale: 8.0
+2023-03-27 16:59:30,197 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 17:00:09,690 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3944.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 17:00:41,361 INFO [train.py:892] (3/4) Epoch 3, batch 250, loss[loss=0.3689, simple_loss=0.3801, pruned_loss=0.1788, over 19804.00 frames. ], tot_loss[loss=0.3623, simple_loss=0.3794, pruned_loss=0.1726, over 2825843.64 frames. ], batch size: 172, lr: 4.12e-02, grad_scale: 8.0
+2023-03-27 17:01:17,203 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4528, 3.2466, 4.3408, 3.6755, 4.0343, 4.3147, 4.2597, 4.3593],
+       device='cuda:3'), covar=tensor([0.0150, 0.0480, 0.0085, 0.1207, 0.0116, 0.0104, 0.0133, 0.0109],
+       device='cuda:3'), in_proj_covar=tensor([0.0047, 0.0051, 0.0042, 0.0106, 0.0043, 0.0038, 0.0047, 0.0041],
+       device='cuda:3'), out_proj_covar=tensor([7.9301e-05, 9.2351e-05, 6.6195e-05, 1.7672e-04, 6.9456e-05, 6.9478e-05,
+        8.0017e-05, 6.9809e-05], device='cuda:3')
+2023-03-27 17:01:37,877 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 17:01:47,663 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3992.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 17:02:03,000 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5250, 1.4654, 2.5817, 2.6817, 2.3494, 2.4152, 2.5466, 2.2928],
+       device='cuda:3'), covar=tensor([0.0306, 0.1722, 0.0286, 0.0183, 0.0369, 0.0256, 0.0268, 0.0648],
+       device='cuda:3'), in_proj_covar=tensor([0.0052, 0.0104, 0.0055, 0.0046, 0.0048, 0.0053, 0.0048, 0.0057],
+       device='cuda:3'), out_proj_covar=tensor([5.7951e-05, 1.1906e-04, 6.0551e-05, 5.2158e-05, 6.0285e-05, 5.9766e-05,
+        5.8689e-05, 7.2447e-05], device='cuda:3')
+2023-03-27 17:02:15,836 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.914e+02 7.402e+02 9.885e+02 1.286e+03 2.496e+03, threshold=1.977e+03, percent-clipped=5.0
+2023-03-27 17:02:31,809 INFO [train.py:892] (3/4) Epoch 3, batch 300, loss[loss=0.3654, simple_loss=0.3771, pruned_loss=0.1769, over 19822.00 frames. ], tot_loss[loss=0.3626, simple_loss=0.3797, pruned_loss=0.1727, over 3074354.13 frames. ], batch size: 187, lr: 4.11e-02, grad_scale: 8.0
+2023-03-27 17:03:15,130 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4032.0, num_to_drop=1, layers_to_drop={2}
+2023-03-27 17:04:16,800 INFO [train.py:892] (3/4) Epoch 3, batch 350, loss[loss=0.4777, simple_loss=0.4625, pruned_loss=0.2465, over 19619.00 frames. ], tot_loss[loss=0.3613, simple_loss=0.3782, pruned_loss=0.1722, over 3268982.70 frames. ], batch size: 387, lr: 4.10e-02, grad_scale: 8.0
+2023-03-27 17:05:19,343 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5972, 2.8126, 4.2770, 4.0413, 3.4167, 4.3189, 2.9002, 2.5423],
+       device='cuda:3'), covar=tensor([0.0713, 0.5142, 0.0352, 0.0284, 0.1537, 0.0314, 0.0871, 0.1420],
+       device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0156, 0.0062, 0.0065, 0.0107, 0.0054, 0.0070, 0.0088],
+       device='cuda:3'), out_proj_covar=tensor([8.6075e-05, 1.6417e-04, 5.4171e-05, 5.5139e-05, 1.0496e-04, 5.2470e-05,
+        6.4818e-05, 8.2033e-05], device='cuda:3')
+2023-03-27 17:05:45,545 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.173e+02 7.083e+02 8.600e+02 1.034e+03 2.171e+03, threshold=1.720e+03, percent-clipped=1.0
+2023-03-27 17:05:59,473 INFO [train.py:892] (3/4) Epoch 3, batch 400, loss[loss=0.308, simple_loss=0.3385, pruned_loss=0.1388, over 19778.00 frames. ], tot_loss[loss=0.357, simple_loss=0.3753, pruned_loss=0.1694, over 3418827.88 frames. ], batch size: 66, lr: 4.09e-02, grad_scale: 8.0
+2023-03-27 17:07:02,441 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 17:07:46,002 INFO [train.py:892] (3/4) Epoch 3, batch 450, loss[loss=0.5427, simple_loss=0.5513, pruned_loss=0.267, over 17921.00 frames. ], tot_loss[loss=0.3593, simple_loss=0.3771, pruned_loss=0.1707, over 3535024.52 frames. ], batch size: 633, lr: 4.08e-02, grad_scale: 8.0
+2023-03-27 17:08:10,362 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2432, 2.7315, 3.6451, 3.9288, 4.9888, 3.9119, 4.7665, 4.7338],
+       device='cuda:3'), covar=tensor([0.0287, 0.1115, 0.0453, 0.1060, 0.0171, 0.0284, 0.0237, 0.0083],
+       device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0091, 0.0059, 0.0102, 0.0043, 0.0061, 0.0045, 0.0045],
+       device='cuda:3'), out_proj_covar=tensor([5.8510e-05, 8.3894e-05, 5.7088e-05, 9.9324e-05, 4.4334e-05, 5.7474e-05,
+        4.0388e-05, 4.0797e-05], device='cuda:3')
+2023-03-27 17:08:45,292 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-27 17:09:11,878 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4202.0, num_to_drop=1, layers_to_drop={2}
+2023-03-27 17:09:12,041 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 17:09:14,943 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.069e+02 7.856e+02 9.394e+02 1.124e+03 2.140e+03, threshold=1.879e+03, percent-clipped=3.0
+2023-03-27 17:09:28,188 INFO [train.py:892] (3/4) Epoch 3, batch 500, loss[loss=0.3474, simple_loss=0.3675, pruned_loss=0.1636, over 19775.00 frames. ], tot_loss[loss=0.3587, simple_loss=0.3765, pruned_loss=0.1705, over 3627547.59 frames. ], batch size: 130, lr: 4.07e-02, grad_scale: 8.0
+2023-03-27 17:09:37,127 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0205, 1.6836, 2.9361, 3.1867, 3.0270, 2.8869, 2.9876, 2.9179],
+       device='cuda:3'), covar=tensor([0.0215, 0.1719, 0.0261, 0.0180, 0.0307, 0.0204, 0.0274, 0.0612],
+       device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0110, 0.0059, 0.0049, 0.0049, 0.0055, 0.0051, 0.0060],
+       device='cuda:3'), out_proj_covar=tensor([6.2955e-05, 1.2807e-04, 6.7306e-05, 5.9368e-05, 6.3539e-05, 6.4165e-05,
+        6.4765e-05, 7.8330e-05], device='cuda:3')
+2023-03-27 17:10:50,210 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4250.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 17:11:12,796 INFO [train.py:892] (3/4) Epoch 3, batch 550, loss[loss=0.3263, simple_loss=0.3508, pruned_loss=0.1509, over 19741.00 frames. ], tot_loss[loss=0.3576, simple_loss=0.3756, pruned_loss=0.1698, over 3699489.95 frames. ], batch size: 89, lr: 4.06e-02, grad_scale: 8.0
+2023-03-27 17:11:30,135 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3388, 3.2463, 4.0057, 3.5094, 3.8222, 4.1537, 4.0596, 3.9301],
+       device='cuda:3'), covar=tensor([0.0148, 0.0503, 0.0129, 0.1297, 0.0153, 0.0112, 0.0162, 0.0150],
+       device='cuda:3'), in_proj_covar=tensor([0.0046, 0.0051, 0.0043, 0.0109, 0.0043, 0.0039, 0.0047, 0.0041],
+       device='cuda:3'), out_proj_covar=tensor([7.9661e-05, 9.6036e-05, 7.1113e-05, 1.8908e-04, 7.3738e-05, 7.6342e-05,
+        8.6992e-05, 7.4465e-05], device='cuda:3')
+2023-03-27 17:11:56,650 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 17:12:08,109 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 17:12:42,158 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.575e+02 7.145e+02 8.933e+02 1.130e+03 1.909e+03, threshold=1.787e+03, percent-clipped=2.0
+2023-03-27 17:12:45,220 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.5479, 5.7515, 5.6008, 5.5181, 5.5343, 5.8238, 4.9663, 5.0141],
+       device='cuda:3'), covar=tensor([0.0402, 0.0281, 0.0816, 0.0422, 0.0446, 0.0434, 0.0498, 0.0927],
+       device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0094, 0.0135, 0.0104, 0.0111, 0.0090, 0.0121, 0.0159],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 17:12:45,503 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4974, 2.8229, 3.9462, 3.6055, 3.5622, 3.9815, 3.1435, 3.0295],
+       device='cuda:3'), covar=tensor([0.0823, 0.6621, 0.0445, 0.0662, 0.1531, 0.0397, 0.0931, 0.1576],
+       device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0185, 0.0072, 0.0074, 0.0127, 0.0063, 0.0079, 0.0102],
+       device='cuda:3'), out_proj_covar=tensor([9.6717e-05, 1.9701e-04, 6.5688e-05, 6.3888e-05, 1.2600e-04, 6.3626e-05,
+        7.4422e-05, 9.6474e-05], device='cuda:3')
+2023-03-27 17:12:57,542 INFO [train.py:892] (3/4) Epoch 3, batch 600, loss[loss=0.3467, simple_loss=0.3681, pruned_loss=0.1626, over 19773.00 frames. ], tot_loss[loss=0.3546, simple_loss=0.3736, pruned_loss=0.1678, over 3755506.81 frames. ], batch size: 130, lr: 4.05e-02, grad_scale: 8.0
+2023-03-27 17:13:04,862 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2801, 3.0356, 2.8865, 3.5166, 3.3898, 3.8669, 3.7664, 3.1037],
+       device='cuda:3'), covar=tensor([0.0442, 0.0418, 0.1139, 0.0485, 0.0386, 0.0255, 0.0346, 0.0259],
+       device='cuda:3'), in_proj_covar=tensor([0.0049, 0.0042, 0.0073, 0.0042, 0.0043, 0.0037, 0.0040, 0.0036],
+       device='cuda:3'), out_proj_covar=tensor([6.0698e-05, 5.5778e-05, 1.0090e-04, 5.6666e-05, 5.4683e-05, 5.1734e-05,
+        5.4164e-05, 4.9027e-05], device='cuda:3')
+2023-03-27 17:13:41,285 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4332.0, num_to_drop=1, layers_to_drop={2}
+2023-03-27 17:14:16,429 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 17:14:18,247 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0402, 3.5026, 4.9461, 3.9014, 4.2565, 4.8816, 4.4403, 4.5241],
+       device='cuda:3'), covar=tensor([0.0105, 0.0443, 0.0071, 0.1400, 0.0142, 0.0091, 0.0144, 0.0094],
+       device='cuda:3'), in_proj_covar=tensor([0.0048, 0.0052, 0.0043, 0.0114, 0.0044, 0.0041, 0.0049, 0.0042],
+       device='cuda:3'), out_proj_covar=tensor([8.7324e-05, 1.0095e-04, 7.2500e-05, 2.0072e-04, 7.6987e-05, 8.1430e-05,
+        9.3084e-05, 7.7129e-05], device='cuda:3')
+2023-03-27 17:14:31,491 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4544, 5.8343, 5.8895, 5.8389, 5.8729, 5.6594, 5.6258, 5.6242],
+       device='cuda:3'), covar=tensor([0.0931, 0.0722, 0.0891, 0.0366, 0.0540, 0.0845, 0.1125, 0.1920],
+       device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0119, 0.0180, 0.0129, 0.0147, 0.0142, 0.0159, 0.0210],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 17:14:38,642 INFO [train.py:892] (3/4) Epoch 3, batch 650, loss[loss=0.4076, simple_loss=0.4076, pruned_loss=0.2038, over 19769.00 frames. ], tot_loss[loss=0.3525, simple_loss=0.3721, pruned_loss=0.1665, over 3797192.55 frames. ], batch size: 273, lr: 4.04e-02, grad_scale: 8.0
+2023-03-27 17:15:21,827 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4380.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 17:16:08,628 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.540e+02 7.551e+02 9.171e+02 1.079e+03 2.038e+03, threshold=1.834e+03, percent-clipped=1.0
+2023-03-27 17:16:24,580 INFO [train.py:892] (3/4) Epoch 3, batch 700, loss[loss=0.3051, simple_loss=0.3358, pruned_loss=0.1372, over 19710.00 frames. ], tot_loss[loss=0.3565, simple_loss=0.375, pruned_loss=0.169, over 3830695.12 frames. ], batch size: 78, lr: 4.03e-02, grad_scale: 8.0
+2023-03-27 17:16:42,030 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0
+2023-03-27 17:18:06,990 INFO [train.py:892] (3/4) Epoch 3, batch 750, loss[loss=0.3378, simple_loss=0.3715, pruned_loss=0.1521, over 19652.00 frames. ], tot_loss[loss=0.3552, simple_loss=0.374, pruned_loss=0.1682, over 3857608.88 frames. ], batch size: 68, lr: 4.02e-02, grad_scale: 8.0
+2023-03-27 17:19:22,158 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 17:19:29,963 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5734, 3.3208, 4.3837, 3.7176, 4.0780, 4.3926, 4.2157, 4.0192],
+       device='cuda:3'), covar=tensor([0.0112, 0.0426, 0.0082, 0.1189, 0.0119, 0.0097, 0.0129, 0.0121],
+       device='cuda:3'), in_proj_covar=tensor([0.0048, 0.0051, 0.0044, 0.0117, 0.0045, 0.0041, 0.0049, 0.0043],
+       device='cuda:3'), out_proj_covar=tensor([8.7445e-05, 1.0241e-04, 7.6084e-05, 2.0828e-04, 8.0364e-05, 8.4605e-05,
+        9.3404e-05, 8.2001e-05], device='cuda:3')
+2023-03-27 17:19:34,746 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.458e+02 7.939e+02 9.251e+02 1.113e+03 1.728e+03, threshold=1.850e+03, percent-clipped=0.0
+2023-03-27 17:19:48,315 INFO [train.py:892] (3/4) Epoch 3, batch 800, loss[loss=0.367, simple_loss=0.3835, pruned_loss=0.1752, over 19766.00 frames. ], tot_loss[loss=0.3532, simple_loss=0.3723, pruned_loss=0.167, over 3879749.75 frames. ], batch size: 217, lr: 4.01e-02, grad_scale: 8.0
+2023-03-27 17:21:33,146 INFO [train.py:892] (3/4) Epoch 3, batch 850, loss[loss=0.49, simple_loss=0.5009, pruned_loss=0.2396, over 18766.00 frames. ], tot_loss[loss=0.351, simple_loss=0.3712, pruned_loss=0.1654, over 3894825.90 frames. ], batch size: 564, lr: 4.00e-02, grad_scale: 8.0
+2023-03-27 17:22:13,954 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 17:22:58,523 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.263e+02 6.963e+02 8.935e+02 1.119e+03 2.174e+03, threshold=1.787e+03, percent-clipped=2.0
+2023-03-27 17:23:13,499 INFO [train.py:892] (3/4) Epoch 3, batch 900, loss[loss=0.3267, simple_loss=0.3587, pruned_loss=0.1473, over 19747.00 frames. ], tot_loss[loss=0.3525, simple_loss=0.3723, pruned_loss=0.1663, over 3905693.76 frames. ], batch size: 97, lr: 3.99e-02, grad_scale: 8.0
+2023-03-27 17:23:53,307 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 17:24:01,812 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9919, 2.6075, 4.0153, 3.6671, 4.6083, 3.7901, 3.6820, 4.0819],
+       device='cuda:3'), covar=tensor([0.0434, 0.1487, 0.0396, 0.1215, 0.0278, 0.0352, 0.0315, 0.0454],
+       device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0104, 0.0073, 0.0118, 0.0053, 0.0072, 0.0054, 0.0057],
+       device='cuda:3'), out_proj_covar=tensor([7.5772e-05, 1.0160e-04, 7.6952e-05, 1.1982e-04, 6.0192e-05, 7.3622e-05,
+        5.2758e-05, 5.7168e-05], device='cuda:3')
+2023-03-27 17:24:23,408 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4644.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 17:24:56,909 INFO [train.py:892] (3/4) Epoch 3, batch 950, loss[loss=0.3495, simple_loss=0.3674, pruned_loss=0.1658, over 19793.00 frames. ], tot_loss[loss=0.3526, simple_loss=0.3723, pruned_loss=0.1665, over 3915847.51 frames. ], batch size: 120, lr: 3.98e-02, grad_scale: 8.0
+2023-03-27 17:25:20,254 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-27 17:26:26,127 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.384e+02 7.071e+02 8.023e+02 9.803e+02 2.669e+03, threshold=1.605e+03, percent-clipped=2.0
+2023-03-27 17:26:42,508 INFO [train.py:892] (3/4) Epoch 3, batch 1000, loss[loss=0.3331, simple_loss=0.351, pruned_loss=0.1576, over 19740.00 frames. ], tot_loss[loss=0.3481, simple_loss=0.3693, pruned_loss=0.1635, over 3924438.88 frames. ], batch size: 134, lr: 3.97e-02, grad_scale: 8.0
+2023-03-27 17:27:59,175 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0
+2023-03-27 17:28:22,547 INFO [train.py:892] (3/4) Epoch 3, batch 1050, loss[loss=0.38, simple_loss=0.3919, pruned_loss=0.1841, over 19721.00 frames. ], tot_loss[loss=0.3462, simple_loss=0.3677, pruned_loss=0.1623, over 3930524.56 frames. ], batch size: 63, lr: 3.96e-02, grad_scale: 8.0
+2023-03-27 17:29:37,216 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 17:29:50,537 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.737e+02 7.391e+02 9.077e+02 1.100e+03 2.003e+03, threshold=1.815e+03, percent-clipped=3.0
+2023-03-27 17:30:03,721 INFO [train.py:892] (3/4) Epoch 3, batch 1100, loss[loss=0.3884, simple_loss=0.3776, pruned_loss=0.1996, over 19640.00 frames. ], tot_loss[loss=0.3459, simple_loss=0.3673, pruned_loss=0.1622, over 3935175.01 frames. ], batch size: 72, lr: 3.95e-02, grad_scale: 8.0
+2023-03-27 17:31:13,968 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3153, 2.5852, 3.2293, 2.5427, 3.0323, 3.1061, 2.9876, 3.1899],
+       device='cuda:3'), covar=tensor([0.0179, 0.0388, 0.0156, 0.1220, 0.0151, 0.0194, 0.0236, 0.0124],
+       device='cuda:3'), in_proj_covar=tensor([0.0048, 0.0051, 0.0047, 0.0121, 0.0046, 0.0043, 0.0050, 0.0044],
+       device='cuda:3'), out_proj_covar=tensor([9.1726e-05, 1.0870e-04, 8.5227e-05, 2.2325e-04, 8.5846e-05, 9.2642e-05,
+        1.0183e-04, 8.7616e-05], device='cuda:3')
+2023-03-27 17:31:15,749 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 17:31:50,561 INFO [train.py:892] (3/4) Epoch 3, batch 1150, loss[loss=0.299, simple_loss=0.3251, pruned_loss=0.1365, over 19757.00 frames. ], tot_loss[loss=0.3456, simple_loss=0.367, pruned_loss=0.1621, over 3937636.21 frames. ], batch size: 125, lr: 3.95e-02, grad_scale: 8.0
+2023-03-27 17:33:20,477 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.536e+02 7.167e+02 9.190e+02 1.175e+03 2.796e+03, threshold=1.838e+03, percent-clipped=7.0
+2023-03-27 17:33:33,646 INFO [train.py:892] (3/4) Epoch 3, batch 1200, loss[loss=0.3435, simple_loss=0.3698, pruned_loss=0.1586, over 19733.00 frames. ], tot_loss[loss=0.3423, simple_loss=0.3648, pruned_loss=0.1599, over 3941412.03 frames. ], batch size: 76, lr: 3.94e-02, grad_scale: 8.0
+2023-03-27 17:34:13,529 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-27 17:34:43,353 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 17:35:18,119 INFO [train.py:892] (3/4) Epoch 3, batch 1250, loss[loss=0.3252, simple_loss=0.3419, pruned_loss=0.1543, over 19875.00 frames. ], tot_loss[loss=0.3437, simple_loss=0.3652, pruned_loss=0.1611, over 3944266.95 frames. ], batch size: 159, lr: 3.93e-02, grad_scale: 8.0
+2023-03-27 17:36:20,314 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 17:36:42,698 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0
+2023-03-27 17:36:47,192 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.843e+02 6.781e+02 8.883e+02 1.088e+03 2.699e+03, threshold=1.777e+03, percent-clipped=1.0
+2023-03-27 17:37:00,636 INFO [train.py:892] (3/4) Epoch 3, batch 1300, loss[loss=0.3666, simple_loss=0.3843, pruned_loss=0.1744, over 19784.00 frames. ], tot_loss[loss=0.3416, simple_loss=0.3638, pruned_loss=0.1597, over 3946837.73 frames. ], batch size: 191, lr: 3.92e-02, grad_scale: 8.0
+2023-03-27 17:37:28,385 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8358, 4.3301, 4.6989, 4.3329, 4.8025, 3.5594, 3.8619, 4.3409],
+       device='cuda:3'), covar=tensor([0.0133, 0.0143, 0.0108, 0.0144, 0.0108, 0.0485, 0.0916, 0.0233],
+       device='cuda:3'), in_proj_covar=tensor([0.0051, 0.0054, 0.0057, 0.0066, 0.0060, 0.0078, 0.0092, 0.0062],
+       device='cuda:3'), out_proj_covar=tensor([9.3113e-05, 1.1141e-04, 1.0320e-04, 1.2489e-04, 1.2484e-04, 1.4957e-04,
+        1.6839e-04, 1.1593e-04], device='cuda:3')
+2023-03-27 17:38:43,017 INFO [train.py:892] (3/4) Epoch 3, batch 1350, loss[loss=0.3164, simple_loss=0.3379, pruned_loss=0.1475, over 19888.00 frames. ], tot_loss[loss=0.3426, simple_loss=0.3647, pruned_loss=0.1602, over 3948272.41 frames. ], batch size: 176, lr: 3.91e-02, grad_scale: 8.0
+2023-03-27 17:39:13,707 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.64 vs. limit=5.0
+2023-03-27 17:40:10,987 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.510e+02 6.737e+02 7.871e+02 9.316e+02 1.602e+03, threshold=1.574e+03, percent-clipped=0.0
+2023-03-27 17:40:26,686 INFO [train.py:892] (3/4) Epoch 3, batch 1400, loss[loss=0.3345, simple_loss=0.3715, pruned_loss=0.1487, over 19604.00 frames. ], tot_loss[loss=0.3409, simple_loss=0.3634, pruned_loss=0.1591, over 3949656.77 frames. ], batch size: 50, lr: 3.90e-02, grad_scale: 8.0
+2023-03-27 17:42:06,528 INFO [train.py:892] (3/4) Epoch 3, batch 1450, loss[loss=0.3713, simple_loss=0.3797, pruned_loss=0.1815, over 19755.00 frames. ], tot_loss[loss=0.3423, simple_loss=0.3645, pruned_loss=0.1601, over 3950110.27 frames. ], batch size: 253, lr: 3.89e-02, grad_scale: 8.0
+2023-03-27 17:42:22,077 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.2590, 2.3171, 3.0330, 2.4540, 2.9594, 2.2807, 2.5380, 3.1806],
+       device='cuda:3'), covar=tensor([0.0845, 0.0342, 0.0306, 0.0409, 0.0288, 0.0416, 0.0406, 0.0306],
+       device='cuda:3'), in_proj_covar=tensor([0.0039, 0.0032, 0.0035, 0.0045, 0.0035, 0.0032, 0.0030, 0.0030],
+       device='cuda:3'), out_proj_covar=tensor([6.5354e-05, 5.4118e-05, 5.8266e-05, 6.8075e-05, 5.8690e-05, 5.5217e-05,
+        5.0588e-05, 5.0425e-05], device='cuda:3')
+2023-03-27 17:42:52,400 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0
+2023-03-27 17:43:29,281 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.83 vs. limit=5.0
+2023-03-27 17:43:35,887 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 5.438e+02 7.354e+02 9.251e+02 1.233e+03 2.096e+03, threshold=1.850e+03, percent-clipped=6.0
+2023-03-27 17:43:49,221 INFO [train.py:892] (3/4) Epoch 3, batch 1500, loss[loss=0.328, simple_loss=0.3711, pruned_loss=0.1424, over 19664.00 frames. ], tot_loss[loss=0.3429, simple_loss=0.3649, pruned_loss=0.1604, over 3949820.64 frames. ], batch size: 57, lr: 3.88e-02, grad_scale: 8.0
+2023-03-27 17:45:34,251 INFO [train.py:892] (3/4) Epoch 3, batch 1550, loss[loss=0.3329, simple_loss=0.3579, pruned_loss=0.154, over 19845.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3641, pruned_loss=0.1592, over 3948976.32 frames. ], batch size: 128, lr: 3.87e-02, grad_scale: 8.0
+2023-03-27 17:47:02,176 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.559e+02 6.524e+02 7.733e+02 9.539e+02 1.385e+03, threshold=1.547e+03, percent-clipped=0.0
+2023-03-27 17:47:15,533 INFO [train.py:892] (3/4) Epoch 3, batch 1600, loss[loss=0.3053, simple_loss=0.3394, pruned_loss=0.1356, over 19803.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.3642, pruned_loss=0.1592, over 3949875.63 frames. ], batch size: 68, lr: 3.86e-02, grad_scale: 8.0
+2023-03-27 17:48:56,941 INFO [train.py:892] (3/4) Epoch 3, batch 1650, loss[loss=0.354, simple_loss=0.3629, pruned_loss=0.1725, over 19803.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3623, pruned_loss=0.1581, over 3951069.72 frames. ], batch size: 224, lr: 3.85e-02, grad_scale: 8.0
+2023-03-27 17:50:27,888 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.447e+02 7.036e+02 8.570e+02 1.037e+03 1.553e+03, threshold=1.714e+03, percent-clipped=1.0
+2023-03-27 17:50:41,756 INFO [train.py:892] (3/4) Epoch 3, batch 1700, loss[loss=0.3384, simple_loss=0.3777, pruned_loss=0.1496, over 19730.00 frames. ], tot_loss[loss=0.3396, simple_loss=0.3633, pruned_loss=0.1579, over 3949840.49 frames. ], batch size: 63, lr: 3.84e-02, grad_scale: 8.0
+2023-03-27 17:50:53,737 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6375, 2.8393, 2.2305, 3.1944, 2.9723, 3.1424, 3.3212, 2.6852],
+       device='cuda:3'), covar=tensor([0.0681, 0.0544, 0.1274, 0.0691, 0.0432, 0.0726, 0.0800, 0.0387],
+       device='cuda:3'), in_proj_covar=tensor([0.0064, 0.0050, 0.0086, 0.0054, 0.0051, 0.0046, 0.0053, 0.0047],
+       device='cuda:3'), out_proj_covar=tensor([9.6243e-05, 8.0150e-05, 1.3367e-04, 8.8044e-05, 7.8885e-05, 7.6554e-05,
+        8.6671e-05, 7.7903e-05], device='cuda:3')
+2023-03-27 17:50:55,479 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6957, 3.3401, 3.4307, 3.7324, 3.4517, 3.5467, 3.6747, 3.8393],
+       device='cuda:3'), covar=tensor([0.0507, 0.0373, 0.0407, 0.0274, 0.0518, 0.0567, 0.0275, 0.0366],
+       device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0097, 0.0107, 0.0093, 0.0097, 0.0077, 0.0105, 0.0117],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0001, 0.0002],
+       device='cuda:3')
+2023-03-27 17:51:15,212 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4273, 2.0285, 2.8875, 2.5344, 2.8725, 2.3973, 2.4896, 3.0531],
+       device='cuda:3'), covar=tensor([0.0594, 0.0444, 0.0472, 0.0405, 0.0448, 0.0354, 0.0428, 0.0221],
+       device='cuda:3'), in_proj_covar=tensor([0.0043, 0.0034, 0.0038, 0.0049, 0.0040, 0.0033, 0.0032, 0.0031],
+       device='cuda:3'), out_proj_covar=tensor([7.4169e-05, 5.9467e-05, 6.5770e-05, 7.6842e-05, 6.9189e-05, 6.0070e-05,
+        5.7655e-05, 5.4514e-05], device='cuda:3')
+2023-03-27 17:52:02,951 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0025, 2.2333, 3.2603, 2.6596, 2.6137, 3.3096, 2.2667, 2.2084],
+       device='cuda:3'), covar=tensor([0.0551, 0.3140, 0.0321, 0.0675, 0.1120, 0.0340, 0.0743, 0.1213],
+       device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0219, 0.0097, 0.0107, 0.0174, 0.0087, 0.0111, 0.0128],
+       device='cuda:3'), out_proj_covar=tensor([1.3875e-04, 2.3636e-04, 9.9659e-05, 1.0437e-04, 1.8058e-04, 9.5120e-05,
+        1.1453e-04, 1.2969e-04], device='cuda:3')
+2023-03-27 17:52:19,755 INFO [train.py:892] (3/4) Epoch 3, batch 1750, loss[loss=0.3887, simple_loss=0.396, pruned_loss=0.1907, over 19700.00 frames. ], tot_loss[loss=0.3377, simple_loss=0.362, pruned_loss=0.1567, over 3947582.44 frames. ], batch size: 265, lr: 3.83e-02, grad_scale: 8.0
+2023-03-27 17:53:15,298 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.4415, 1.7974, 1.5575, 1.1801, 1.4762, 1.8057, 1.3998, 1.9413],
+       device='cuda:3'), covar=tensor([0.0291, 0.0348, 0.0360, 0.0630, 0.0778, 0.0411, 0.0351, 0.0292],
+       device='cuda:3'), in_proj_covar=tensor([0.0028, 0.0028, 0.0028, 0.0041, 0.0041, 0.0030, 0.0025, 0.0029],
+       device='cuda:3'), out_proj_covar=tensor([4.4016e-05, 4.5819e-05, 4.4036e-05, 6.7577e-05, 6.5822e-05, 4.7991e-05,
+        4.3536e-05, 4.7865e-05], device='cuda:3')
+2023-03-27 17:53:34,940 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.884e+02 6.493e+02 7.809e+02 9.948e+02 1.676e+03, threshold=1.562e+03, percent-clipped=0.0
+2023-03-27 17:53:46,189 INFO [train.py:892] (3/4) Epoch 3, batch 1800, loss[loss=0.2932, simple_loss=0.3181, pruned_loss=0.1341, over 19793.00 frames. ], tot_loss[loss=0.3364, simple_loss=0.3607, pruned_loss=0.1561, over 3949146.81 frames. ], batch size: 120, lr: 3.82e-02, grad_scale: 16.0
+2023-03-27 17:55:08,420 INFO [train.py:892] (3/4) Epoch 3, batch 1850, loss[loss=0.3194, simple_loss=0.3566, pruned_loss=0.1411, over 19833.00 frames. ], tot_loss[loss=0.3378, simple_loss=0.363, pruned_loss=0.1563, over 3948919.66 frames. ], batch size: 57, lr: 3.81e-02, grad_scale: 16.0
+2023-03-27 17:56:05,352 INFO [train.py:892] (3/4) Epoch 4, batch 0, loss[loss=0.2921, simple_loss=0.3235, pruned_loss=0.1303, over 19789.00 frames. ], tot_loss[loss=0.2921, simple_loss=0.3235, pruned_loss=0.1303, over 19789.00 frames. ], batch size: 79, lr: 3.56e-02, grad_scale: 16.0
+2023-03-27 17:56:05,353 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-27 17:56:21,815 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5534, 3.6079, 4.4864, 4.9519, 3.5829, 4.0287, 4.0486, 3.1473],
+       device='cuda:3'), covar=tensor([0.0273, 0.2403, 0.0423, 0.0093, 0.1865, 0.0374, 0.0459, 0.1861],
+       device='cuda:3'), in_proj_covar=tensor([0.0110, 0.0263, 0.0133, 0.0088, 0.0189, 0.0103, 0.0128, 0.0184],
+       device='cuda:3'), out_proj_covar=tensor([1.1510e-04, 2.5885e-04, 1.3836e-04, 8.4456e-05, 1.8556e-04, 9.9827e-05,
+        1.2950e-04, 1.8085e-04], device='cuda:3')
+2023-03-27 17:56:31,644 INFO [train.py:926] (3/4) Epoch 4, validation: loss=0.2293, simple_loss=0.3025, pruned_loss=0.07807, over 2883724.00 frames. 
+2023-03-27 17:56:31,645 INFO [train.py:927] (3/4) Maximum memory allocated so far is 21910MB
+2023-03-27 17:57:55,093 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.716e+02 7.146e+02 8.558e+02 9.901e+02 2.056e+03, threshold=1.712e+03, percent-clipped=2.0
+2023-03-27 17:58:16,485 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3371, 2.5719, 3.8636, 3.2983, 3.3210, 3.9433, 2.6118, 2.5537],
+       device='cuda:3'), covar=tensor([0.0553, 0.4665, 0.0249, 0.0422, 0.1177, 0.0298, 0.0687, 0.1429],
+       device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0227, 0.0100, 0.0110, 0.0180, 0.0090, 0.0113, 0.0133],
+       device='cuda:3'), out_proj_covar=tensor([1.4208e-04, 2.4411e-04, 1.0412e-04, 1.0747e-04, 1.8790e-04, 9.9310e-05,
+        1.1496e-04, 1.3595e-04], device='cuda:3')
+2023-03-27 17:58:21,180 INFO [train.py:892] (3/4) Epoch 4, batch 50, loss[loss=0.3088, simple_loss=0.3351, pruned_loss=0.1413, over 19838.00 frames. ], tot_loss[loss=0.3244, simple_loss=0.3515, pruned_loss=0.1487, over 890335.31 frames. ], batch size: 161, lr: 3.55e-02, grad_scale: 16.0
+2023-03-27 17:59:36,047 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5650.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 18:00:07,039 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4709, 3.1156, 3.2173, 3.4980, 3.2209, 3.2076, 3.4227, 3.6325],
+       device='cuda:3'), covar=tensor([0.0463, 0.0367, 0.0396, 0.0268, 0.0509, 0.0731, 0.0315, 0.0381],
+       device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0096, 0.0105, 0.0092, 0.0093, 0.0075, 0.0101, 0.0116],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0001, 0.0002],
+       device='cuda:3')
+2023-03-27 18:00:10,206 INFO [train.py:892] (3/4) Epoch 4, batch 100, loss[loss=0.2467, simple_loss=0.303, pruned_loss=0.09524, over 19916.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3532, pruned_loss=0.1497, over 1569772.63 frames. ], batch size: 53, lr: 3.54e-02, grad_scale: 16.0
+2023-03-27 18:01:31,609 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.543e+02 7.031e+02 8.200e+02 9.251e+02 1.434e+03, threshold=1.640e+03, percent-clipped=0.0
+2023-03-27 18:01:46,510 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5711.0, num_to_drop=1, layers_to_drop={3}
+2023-03-27 18:01:55,776 INFO [train.py:892] (3/4) Epoch 4, batch 150, loss[loss=0.3009, simple_loss=0.3264, pruned_loss=0.1377, over 19849.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.352, pruned_loss=0.1471, over 2097189.03 frames. ], batch size: 142, lr: 3.54e-02, grad_scale: 16.0
+2023-03-27 18:03:18,858 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:03:42,976 INFO [train.py:892] (3/4) Epoch 4, batch 200, loss[loss=0.3185, simple_loss=0.3415, pruned_loss=0.1477, over 19786.00 frames. ], tot_loss[loss=0.3264, simple_loss=0.3548, pruned_loss=0.149, over 2509296.13 frames. ], batch size: 193, lr: 3.53e-02, grad_scale: 16.0
+2023-03-27 18:04:05,925 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.13 vs. limit=2.0
+2023-03-27 18:04:58,698 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.3700, 2.2185, 3.1439, 2.4181, 3.0011, 1.8041, 2.7790, 3.1478],
+       device='cuda:3'), covar=tensor([0.1524, 0.0440, 0.0464, 0.0556, 0.0624, 0.0546, 0.0487, 0.0377],
+       device='cuda:3'), in_proj_covar=tensor([0.0042, 0.0033, 0.0037, 0.0050, 0.0037, 0.0032, 0.0032, 0.0029],
+       device='cuda:3'), out_proj_covar=tensor([7.7108e-05, 6.1307e-05, 6.6638e-05, 8.1308e-05, 6.8242e-05, 6.1921e-05,
+        6.1109e-05, 5.4558e-05], device='cuda:3')
+2023-03-27 18:05:07,192 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.424e+02 6.665e+02 7.849e+02 9.625e+02 1.890e+03, threshold=1.570e+03, percent-clipped=1.0
+2023-03-27 18:05:31,203 INFO [train.py:892] (3/4) Epoch 4, batch 250, loss[loss=0.2631, simple_loss=0.3153, pruned_loss=0.1054, over 19564.00 frames. ], tot_loss[loss=0.3223, simple_loss=0.3518, pruned_loss=0.1464, over 2827988.85 frames. ], batch size: 41, lr: 3.52e-02, grad_scale: 16.0
+2023-03-27 18:05:32,111 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:06:59,873 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:07:17,595 INFO [train.py:892] (3/4) Epoch 4, batch 300, loss[loss=0.3337, simple_loss=0.3702, pruned_loss=0.1486, over 19686.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.352, pruned_loss=0.1467, over 3076910.26 frames. ], batch size: 59, lr: 3.51e-02, grad_scale: 16.0
+2023-03-27 18:08:37,832 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.638e+02 6.552e+02 7.972e+02 9.789e+02 1.946e+03, threshold=1.594e+03, percent-clipped=3.0
+2023-03-27 18:09:05,480 INFO [train.py:892] (3/4) Epoch 4, batch 350, loss[loss=0.2883, simple_loss=0.3268, pruned_loss=0.1249, over 19875.00 frames. ], tot_loss[loss=0.3248, simple_loss=0.3534, pruned_loss=0.1481, over 3270138.54 frames. ], batch size: 159, lr: 3.50e-02, grad_scale: 16.0
+2023-03-27 18:09:10,801 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.78 vs. limit=5.0
+2023-03-27 18:09:12,518 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:09:21,851 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6622, 4.2307, 4.6367, 4.2587, 4.6836, 3.3604, 3.7898, 4.1825],
+       device='cuda:3'), covar=tensor([0.0178, 0.0184, 0.0115, 0.0166, 0.0120, 0.0676, 0.0988, 0.0304],
+       device='cuda:3'), in_proj_covar=tensor([0.0053, 0.0061, 0.0063, 0.0071, 0.0064, 0.0089, 0.0100, 0.0074],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 18:10:48,434 INFO [train.py:892] (3/4) Epoch 4, batch 400, loss[loss=0.49, simple_loss=0.4746, pruned_loss=0.2527, over 19572.00 frames. ], tot_loss[loss=0.3227, simple_loss=0.3517, pruned_loss=0.1469, over 3421935.76 frames. ], batch size: 376, lr: 3.49e-02, grad_scale: 16.0
+2023-03-27 18:11:41,742 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5990.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:12:12,455 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.258e+02 6.552e+02 7.732e+02 9.637e+02 1.932e+03, threshold=1.546e+03, percent-clipped=3.0
+2023-03-27 18:12:13,389 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4938, 5.0359, 5.2486, 4.9838, 4.6648, 5.2084, 5.1040, 5.6929],
+       device='cuda:3'), covar=tensor([0.1248, 0.0224, 0.0243, 0.0209, 0.0315, 0.0202, 0.0169, 0.0143],
+       device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0121, 0.0114, 0.0112, 0.0120, 0.0109, 0.0099, 0.0099],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 18:12:17,360 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6006.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 18:12:36,855 INFO [train.py:892] (3/4) Epoch 4, batch 450, loss[loss=0.4648, simple_loss=0.4834, pruned_loss=0.2231, over 18988.00 frames. ], tot_loss[loss=0.3233, simple_loss=0.3521, pruned_loss=0.1472, over 3539263.51 frames. ], batch size: 514, lr: 3.48e-02, grad_scale: 16.0
+2023-03-27 18:13:33,594 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:13:52,284 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6051.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:14:22,252 INFO [train.py:892] (3/4) Epoch 4, batch 500, loss[loss=0.3394, simple_loss=0.3563, pruned_loss=0.1612, over 19836.00 frames. ], tot_loss[loss=0.3276, simple_loss=0.3551, pruned_loss=0.1501, over 3628977.47 frames. ], batch size: 171, lr: 3.47e-02, grad_scale: 16.0
+2023-03-27 18:14:33,363 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9883, 2.5259, 3.1777, 3.3706, 4.1773, 3.7299, 3.9113, 4.1797],
+       device='cuda:3'), covar=tensor([0.0398, 0.1338, 0.0694, 0.1182, 0.0731, 0.0460, 0.0195, 0.0203],
+       device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0133, 0.0107, 0.0154, 0.0090, 0.0107, 0.0074, 0.0079],
+       device='cuda:3'), out_proj_covar=tensor([1.1835e-04, 1.4257e-04, 1.3080e-04, 1.6883e-04, 1.1204e-04, 1.2648e-04,
+        8.5936e-05, 9.3942e-05], device='cuda:3')
+2023-03-27 18:15:43,597 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.065e+02 6.254e+02 7.910e+02 9.308e+02 1.450e+03, threshold=1.582e+03, percent-clipped=0.0
+2023-03-27 18:15:44,548 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:15:58,502 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:16:07,743 INFO [train.py:892] (3/4) Epoch 4, batch 550, loss[loss=0.3286, simple_loss=0.3629, pruned_loss=0.1471, over 19698.00 frames. ], tot_loss[loss=0.3256, simple_loss=0.3542, pruned_loss=0.1485, over 3700755.44 frames. ], batch size: 59, lr: 3.47e-02, grad_scale: 16.0
+2023-03-27 18:16:46,275 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0577, 3.1816, 2.1921, 3.7636, 3.0993, 3.5178, 3.6930, 2.9721],
+       device='cuda:3'), covar=tensor([0.0603, 0.0430, 0.1539, 0.0627, 0.0656, 0.0563, 0.0518, 0.0418],
+       device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0058, 0.0093, 0.0062, 0.0059, 0.0050, 0.0058, 0.0055],
+       device='cuda:3'), out_proj_covar=tensor([1.1604e-04, 9.8484e-05, 1.5216e-04, 1.0876e-04, 9.8723e-05, 8.9428e-05,
+        1.0259e-04, 9.8192e-05], device='cuda:3')
+2023-03-27 18:16:51,681 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6135.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:17:55,484 INFO [train.py:892] (3/4) Epoch 4, batch 600, loss[loss=0.3235, simple_loss=0.352, pruned_loss=0.1474, over 19892.00 frames. ], tot_loss[loss=0.3273, simple_loss=0.3554, pruned_loss=0.1496, over 3755156.42 frames. ], batch size: 77, lr: 3.46e-02, grad_scale: 16.0
+2023-03-27 18:18:03,579 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:18:20,623 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.3793, 5.6602, 5.6659, 5.5830, 5.4524, 5.5943, 4.9641, 5.1038],
+       device='cuda:3'), covar=tensor([0.0282, 0.0253, 0.0474, 0.0330, 0.0468, 0.0539, 0.0490, 0.0734],
+       device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0109, 0.0157, 0.0121, 0.0125, 0.0106, 0.0140, 0.0175],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 18:18:56,322 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:19:15,165 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:19:16,015 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.339e+02 6.549e+02 8.017e+02 9.808e+02 1.883e+03, threshold=1.603e+03, percent-clipped=1.0
+2023-03-27 18:19:30,044 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:19:37,410 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6214.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:19:40,632 INFO [train.py:892] (3/4) Epoch 4, batch 650, loss[loss=0.3205, simple_loss=0.3537, pruned_loss=0.1436, over 19856.00 frames. ], tot_loss[loss=0.3252, simple_loss=0.3539, pruned_loss=0.1482, over 3797465.98 frames. ], batch size: 81, lr: 3.45e-02, grad_scale: 16.0
+2023-03-27 18:20:11,607 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:21:09,651 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0
+2023-03-27 18:21:22,740 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:21:25,689 INFO [train.py:892] (3/4) Epoch 4, batch 700, loss[loss=0.3412, simple_loss=0.3602, pruned_loss=0.1611, over 19759.00 frames. ], tot_loss[loss=0.3235, simple_loss=0.3526, pruned_loss=0.1472, over 3832105.60 frames. ], batch size: 256, lr: 3.44e-02, grad_scale: 16.0
+2023-03-27 18:21:26,835 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6266.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:21:42,050 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:21:58,798 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0445, 4.1878, 4.4855, 4.2867, 4.4368, 4.0263, 4.1154, 4.1225],
+       device='cuda:3'), covar=tensor([0.1172, 0.0891, 0.0798, 0.0709, 0.0752, 0.0894, 0.1778, 0.2260],
+       device='cuda:3'), in_proj_covar=tensor([0.0184, 0.0151, 0.0216, 0.0170, 0.0173, 0.0160, 0.0199, 0.0249],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-27 18:22:47,949 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.869e+02 6.690e+02 8.159e+02 9.859e+02 1.679e+03, threshold=1.632e+03, percent-clipped=2.0
+2023-03-27 18:22:54,134 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6306.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 18:23:08,949 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-27 18:23:15,289 INFO [train.py:892] (3/4) Epoch 4, batch 750, loss[loss=0.2845, simple_loss=0.3205, pruned_loss=0.1243, over 19835.00 frames. ], tot_loss[loss=0.3236, simple_loss=0.3527, pruned_loss=0.1473, over 3856014.29 frames. ], batch size: 76, lr: 3.43e-02, grad_scale: 8.0
+2023-03-27 18:23:38,538 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6327.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:24:16,272 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:24:34,078 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6354.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 18:24:57,112 INFO [train.py:892] (3/4) Epoch 4, batch 800, loss[loss=0.3642, simple_loss=0.3807, pruned_loss=0.1739, over 19837.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.3527, pruned_loss=0.1474, over 3877499.27 frames. ], batch size: 145, lr: 3.42e-02, grad_scale: 8.0
+2023-03-27 18:26:04,122 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3409, 3.3080, 4.3136, 3.5773, 3.8048, 4.1876, 4.0732, 4.1689],
+       device='cuda:3'), covar=tensor([0.0119, 0.0384, 0.0097, 0.1218, 0.0113, 0.0144, 0.0164, 0.0100],
+       device='cuda:3'), in_proj_covar=tensor([0.0052, 0.0055, 0.0048, 0.0127, 0.0049, 0.0048, 0.0050, 0.0046],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-27 18:26:05,820 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:26:17,456 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.240e+02 6.935e+02 8.296e+02 1.069e+03 2.251e+03, threshold=1.659e+03, percent-clipped=2.0
+2023-03-27 18:26:29,540 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:26:38,546 INFO [train.py:892] (3/4) Epoch 4, batch 850, loss[loss=0.3274, simple_loss=0.3501, pruned_loss=0.1523, over 19735.00 frames. ], tot_loss[loss=0.3237, simple_loss=0.353, pruned_loss=0.1472, over 3893612.04 frames. ], batch size: 221, lr: 3.42e-02, grad_scale: 8.0
+2023-03-27 18:26:40,882 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4247, 4.0225, 4.1372, 4.0069, 4.2971, 3.2570, 3.5043, 3.8702],
+       device='cuda:3'), covar=tensor([0.0150, 0.0171, 0.0122, 0.0144, 0.0103, 0.0609, 0.0961, 0.0305],
+       device='cuda:3'), in_proj_covar=tensor([0.0053, 0.0061, 0.0063, 0.0072, 0.0064, 0.0085, 0.0097, 0.0073],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 18:28:00,378 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2224, 2.4888, 3.1028, 3.5286, 2.2065, 2.7869, 2.5946, 1.9364],
+       device='cuda:3'), covar=tensor([0.0310, 0.2585, 0.0625, 0.0218, 0.2332, 0.0471, 0.0760, 0.2424],
+       device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0275, 0.0146, 0.0090, 0.0199, 0.0111, 0.0138, 0.0186],
+       device='cuda:3'), out_proj_covar=tensor([1.2838e-04, 2.7657e-04, 1.5582e-04, 9.1341e-05, 2.0021e-04, 1.1405e-04,
+        1.4516e-04, 1.9011e-04], device='cuda:3')
+2023-03-27 18:28:04,686 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-27 18:28:08,186 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:28:14,576 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:28:15,242 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-27 18:28:23,855 INFO [train.py:892] (3/4) Epoch 4, batch 900, loss[loss=0.3587, simple_loss=0.3929, pruned_loss=0.1622, over 19831.00 frames. ], tot_loss[loss=0.3239, simple_loss=0.3531, pruned_loss=0.1473, over 3905136.43 frames. ], batch size: 57, lr: 3.41e-02, grad_scale: 8.0
+2023-03-27 18:28:26,711 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:29:14,976 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:29:44,758 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.785e+02 6.152e+02 8.046e+02 9.426e+02 1.704e+03, threshold=1.609e+03, percent-clipped=1.0
+2023-03-27 18:30:02,702 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:30:05,648 INFO [train.py:892] (3/4) Epoch 4, batch 950, loss[loss=0.3349, simple_loss=0.3596, pruned_loss=0.1551, over 19653.00 frames. ], tot_loss[loss=0.3225, simple_loss=0.352, pruned_loss=0.1465, over 3915679.40 frames. ], batch size: 69, lr: 3.40e-02, grad_scale: 8.0
+2023-03-27 18:30:19,443 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6523.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 18:30:24,503 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:30:29,934 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6528.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:30:59,015 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.66 vs. limit=5.0
+2023-03-27 18:31:13,796 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6172, 3.8838, 4.4939, 3.7149, 3.9357, 4.3419, 4.2742, 4.3476],
+       device='cuda:3'), covar=tensor([0.0141, 0.0294, 0.0086, 0.1322, 0.0128, 0.0141, 0.0148, 0.0115],
+       device='cuda:3'), in_proj_covar=tensor([0.0053, 0.0057, 0.0050, 0.0130, 0.0049, 0.0050, 0.0051, 0.0048],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0003, 0.0001, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-27 18:31:32,899 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:31:38,665 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:31:46,898 INFO [train.py:892] (3/4) Epoch 4, batch 1000, loss[loss=0.2867, simple_loss=0.3334, pruned_loss=0.12, over 19797.00 frames. ], tot_loss[loss=0.3206, simple_loss=0.3504, pruned_loss=0.1454, over 3923709.62 frames. ], batch size: 83, lr: 3.39e-02, grad_scale: 8.0
+2023-03-27 18:31:49,595 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:31:57,941 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0
+2023-03-27 18:32:32,888 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6538, 2.1286, 2.7879, 2.4795, 2.3210, 2.8129, 2.1526, 2.0347],
+       device='cuda:3'), covar=tensor([0.0556, 0.2371, 0.0402, 0.0578, 0.1100, 0.0265, 0.0790, 0.1191],
+       device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0245, 0.0122, 0.0128, 0.0207, 0.0106, 0.0134, 0.0153],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0002],
+       device='cuda:3')
+2023-03-27 18:32:47,437 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0
+2023-03-27 18:33:08,466 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.155e+02 6.376e+02 7.596e+02 9.023e+02 1.357e+03, threshold=1.519e+03, percent-clipped=0.0
+2023-03-27 18:33:31,582 INFO [train.py:892] (3/4) Epoch 4, batch 1050, loss[loss=0.3013, simple_loss=0.3361, pruned_loss=0.1333, over 19653.00 frames. ], tot_loss[loss=0.3203, simple_loss=0.3502, pruned_loss=0.1452, over 3928906.96 frames. ], batch size: 67, lr: 3.38e-02, grad_scale: 8.0
+2023-03-27 18:33:45,135 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:33:58,978 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:34:35,700 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:35:14,541 INFO [train.py:892] (3/4) Epoch 4, batch 1100, loss[loss=0.3096, simple_loss=0.3348, pruned_loss=0.1422, over 19823.00 frames. ], tot_loss[loss=0.3219, simple_loss=0.3515, pruned_loss=0.1461, over 3931115.24 frames. ], batch size: 202, lr: 3.37e-02, grad_scale: 8.0
+2023-03-27 18:35:38,038 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7993, 2.3540, 2.9536, 2.9745, 3.7789, 3.1989, 3.8825, 4.0492],
+       device='cuda:3'), covar=tensor([0.0383, 0.1295, 0.0692, 0.1217, 0.0636, 0.0802, 0.0236, 0.0223],
+       device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0137, 0.0119, 0.0160, 0.0105, 0.0121, 0.0081, 0.0083],
+       device='cuda:3'), out_proj_covar=tensor([1.3264e-04, 1.5124e-04, 1.4698e-04, 1.8053e-04, 1.3163e-04, 1.4605e-04,
+        9.8969e-05, 1.0034e-04], device='cuda:3')
+2023-03-27 18:35:48,449 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8902, 2.8320, 1.5332, 3.5965, 3.3578, 3.3939, 3.4245, 2.7333],
+       device='cuda:3'), covar=tensor([0.0786, 0.0524, 0.2344, 0.0410, 0.0559, 0.0596, 0.1000, 0.0581],
+       device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0064, 0.0099, 0.0067, 0.0063, 0.0054, 0.0063, 0.0063],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-27 18:36:06,386 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:36:14,098 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:36:23,773 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:36:34,438 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.080e+02 6.475e+02 7.910e+02 9.923e+02 2.054e+03, threshold=1.582e+03, percent-clipped=5.0
+2023-03-27 18:36:43,028 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.78 vs. limit=5.0
+2023-03-27 18:36:57,929 INFO [train.py:892] (3/4) Epoch 4, batch 1150, loss[loss=0.3114, simple_loss=0.3317, pruned_loss=0.1456, over 19740.00 frames. ], tot_loss[loss=0.322, simple_loss=0.3515, pruned_loss=0.1462, over 3934127.48 frames. ], batch size: 134, lr: 3.37e-02, grad_scale: 8.0
+2023-03-27 18:38:01,833 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:38:41,081 INFO [train.py:892] (3/4) Epoch 4, batch 1200, loss[loss=0.2964, simple_loss=0.3381, pruned_loss=0.1274, over 19661.00 frames. ], tot_loss[loss=0.3212, simple_loss=0.3513, pruned_loss=0.1455, over 3937012.03 frames. ], batch size: 67, lr: 3.36e-02, grad_scale: 8.0
+2023-03-27 18:38:48,360 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0
+2023-03-27 18:39:32,808 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:39:59,926 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.559e+02 6.894e+02 8.217e+02 1.051e+03 2.121e+03, threshold=1.643e+03, percent-clipped=4.0
+2023-03-27 18:40:05,047 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-27 18:40:22,609 INFO [train.py:892] (3/4) Epoch 4, batch 1250, loss[loss=0.4823, simple_loss=0.4765, pruned_loss=0.244, over 19401.00 frames. ], tot_loss[loss=0.3214, simple_loss=0.3515, pruned_loss=0.1456, over 3938564.36 frames. ], batch size: 412, lr: 3.35e-02, grad_scale: 8.0
+2023-03-27 18:40:26,923 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6818.0, num_to_drop=1, layers_to_drop={2}
+2023-03-27 18:40:36,559 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:40:42,319 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:40:45,807 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9707, 5.2473, 5.5559, 5.4285, 5.5033, 4.8978, 5.1658, 5.2442],
+       device='cuda:3'), covar=tensor([0.1522, 0.0951, 0.1232, 0.0875, 0.0733, 0.1177, 0.2233, 0.2385],
+       device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0152, 0.0220, 0.0173, 0.0166, 0.0161, 0.0197, 0.0249],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-27 18:41:10,819 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:41:50,027 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:42:04,705 INFO [train.py:892] (3/4) Epoch 4, batch 1300, loss[loss=0.3041, simple_loss=0.3285, pruned_loss=0.1399, over 19839.00 frames. ], tot_loss[loss=0.3185, simple_loss=0.3492, pruned_loss=0.1438, over 3940867.18 frames. ], batch size: 166, lr: 3.34e-02, grad_scale: 8.0
+2023-03-27 18:42:07,290 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:42:11,319 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.2493, 4.7470, 5.0097, 4.7092, 5.1153, 3.5683, 3.9847, 4.4353],
+       device='cuda:3'), covar=tensor([0.0148, 0.0121, 0.0119, 0.0143, 0.0114, 0.0585, 0.1005, 0.0312],
+       device='cuda:3'), in_proj_covar=tensor([0.0052, 0.0060, 0.0063, 0.0071, 0.0064, 0.0086, 0.0098, 0.0074],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 18:42:21,480 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:42:29,212 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6878.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 18:42:37,554 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0
+2023-03-27 18:42:57,466 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6951, 1.7672, 3.1538, 3.1339, 3.0528, 3.3561, 3.2973, 3.1869],
+       device='cuda:3'), covar=tensor([0.0549, 0.2221, 0.0272, 0.0260, 0.0365, 0.0158, 0.0237, 0.0365],
+       device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0142, 0.0081, 0.0074, 0.0066, 0.0066, 0.0059, 0.0065],
+       device='cuda:3'), out_proj_covar=tensor([1.1515e-04, 1.9218e-04, 1.1697e-04, 1.1166e-04, 1.0502e-04, 9.7042e-05,
+        9.6374e-05, 1.0505e-04], device='cuda:3')
+2023-03-27 18:43:24,638 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.702e+02 6.185e+02 7.634e+02 1.003e+03 1.964e+03, threshold=1.527e+03, percent-clipped=1.0
+2023-03-27 18:43:29,149 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:43:34,614 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:43:45,438 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:43:46,682 INFO [train.py:892] (3/4) Epoch 4, batch 1350, loss[loss=0.3325, simple_loss=0.3584, pruned_loss=0.1533, over 19788.00 frames. ], tot_loss[loss=0.3164, simple_loss=0.3479, pruned_loss=0.1425, over 3942945.79 frames. ], batch size: 236, lr: 3.33e-02, grad_scale: 8.0
+2023-03-27 18:43:58,833 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:44:32,714 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6939.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 18:45:27,453 INFO [train.py:892] (3/4) Epoch 4, batch 1400, loss[loss=0.304, simple_loss=0.3356, pruned_loss=0.1362, over 19872.00 frames. ], tot_loss[loss=0.3153, simple_loss=0.3473, pruned_loss=0.1417, over 3944563.14 frames. ], batch size: 138, lr: 3.33e-02, grad_scale: 8.0
+2023-03-27 18:45:37,277 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:45:37,473 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:46:09,137 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:46:09,684 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.12 vs. limit=2.0
+2023-03-27 18:46:49,377 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.908e+02 6.198e+02 7.481e+02 9.100e+02 1.903e+03, threshold=1.496e+03, percent-clipped=2.0
+2023-03-27 18:47:13,276 INFO [train.py:892] (3/4) Epoch 4, batch 1450, loss[loss=0.3317, simple_loss=0.356, pruned_loss=0.1537, over 19775.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.348, pruned_loss=0.1415, over 3944959.97 frames. ], batch size: 224, lr: 3.32e-02, grad_scale: 8.0
+2023-03-27 18:47:20,478 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.73 vs. limit=5.0
+2023-03-27 18:48:17,370 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7048.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 18:48:20,507 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8975, 3.0213, 3.7859, 3.1891, 3.2200, 3.5635, 3.5796, 3.6292],
+       device='cuda:3'), covar=tensor([0.0120, 0.0365, 0.0086, 0.1153, 0.0117, 0.0167, 0.0156, 0.0123],
+       device='cuda:3'), in_proj_covar=tensor([0.0054, 0.0058, 0.0051, 0.0133, 0.0050, 0.0050, 0.0053, 0.0048],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0001, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-27 18:48:54,063 INFO [train.py:892] (3/4) Epoch 4, batch 1500, loss[loss=0.2808, simple_loss=0.3223, pruned_loss=0.1197, over 19709.00 frames. ], tot_loss[loss=0.3173, simple_loss=0.3489, pruned_loss=0.1428, over 3946094.96 frames. ], batch size: 78, lr: 3.31e-02, grad_scale: 8.0
+2023-03-27 18:49:54,267 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0716, 2.5999, 2.9601, 3.4877, 3.8895, 3.6326, 4.0763, 4.2842],
+       device='cuda:3'), covar=tensor([0.0412, 0.1323, 0.0877, 0.1217, 0.0802, 0.0772, 0.0239, 0.0172],
+       device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0147, 0.0131, 0.0169, 0.0122, 0.0133, 0.0090, 0.0089],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-27 18:50:13,928 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.314e+02 6.331e+02 7.635e+02 9.535e+02 1.700e+03, threshold=1.527e+03, percent-clipped=1.0
+2023-03-27 18:50:22,730 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7109.0, num_to_drop=1, layers_to_drop={2}
+2023-03-27 18:50:36,737 INFO [train.py:892] (3/4) Epoch 4, batch 1550, loss[loss=0.2735, simple_loss=0.323, pruned_loss=0.1121, over 19670.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.3488, pruned_loss=0.1423, over 3946728.46 frames. ], batch size: 51, lr: 3.30e-02, grad_scale: 8.0
+2023-03-27 18:50:41,227 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7118.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 18:50:51,496 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:51:19,573 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-27 18:51:59,734 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0926, 2.0013, 4.2186, 3.8579, 4.0019, 4.5227, 4.1415, 4.1715],
+       device='cuda:3'), covar=tensor([0.0527, 0.2586, 0.0230, 0.0278, 0.0243, 0.0104, 0.0264, 0.0306],
+       device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0152, 0.0086, 0.0080, 0.0070, 0.0070, 0.0065, 0.0068],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-27 18:52:15,306 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:52:19,898 INFO [train.py:892] (3/4) Epoch 4, batch 1600, loss[loss=0.3089, simple_loss=0.3349, pruned_loss=0.1414, over 19541.00 frames. ], tot_loss[loss=0.3161, simple_loss=0.3481, pruned_loss=0.1421, over 3945963.41 frames. ], batch size: 46, lr: 3.30e-02, grad_scale: 8.0
+2023-03-27 18:52:20,689 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:52:31,422 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:53:39,226 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.151e+02 6.559e+02 7.813e+02 1.032e+03 1.739e+03, threshold=1.563e+03, percent-clipped=2.0
+2023-03-27 18:53:46,230 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.01 vs. limit=5.0
+2023-03-27 18:53:48,964 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.5546, 5.7455, 5.8299, 5.6284, 5.6808, 5.8094, 4.8470, 5.0052],
+       device='cuda:3'), covar=tensor([0.0318, 0.0391, 0.0607, 0.0469, 0.0495, 0.0553, 0.0720, 0.1111],
+       device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0116, 0.0176, 0.0131, 0.0128, 0.0115, 0.0148, 0.0182],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 18:54:01,351 INFO [train.py:892] (3/4) Epoch 4, batch 1650, loss[loss=0.2794, simple_loss=0.3283, pruned_loss=0.1152, over 19731.00 frames. ], tot_loss[loss=0.3154, simple_loss=0.3481, pruned_loss=0.1413, over 3947584.41 frames. ], batch size: 95, lr: 3.29e-02, grad_scale: 8.0
+2023-03-27 18:54:18,837 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7224.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 18:54:30,161 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6850, 2.9074, 3.4691, 4.0243, 2.5260, 3.0930, 2.7794, 2.2150],
+       device='cuda:3'), covar=tensor([0.0296, 0.2451, 0.0589, 0.0115, 0.1936, 0.0423, 0.0745, 0.1956],
+       device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0284, 0.0155, 0.0092, 0.0202, 0.0116, 0.0147, 0.0186],
+       device='cuda:3'), out_proj_covar=tensor([1.3745e-04, 2.8957e-04, 1.6820e-04, 9.6758e-05, 2.0971e-04, 1.2081e-04,
+        1.5526e-04, 1.9534e-04], device='cuda:3')
+2023-03-27 18:54:36,544 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0
+2023-03-27 18:54:39,109 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7234.0, num_to_drop=1, layers_to_drop={3}
+2023-03-27 18:55:42,233 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:55:43,224 INFO [train.py:892] (3/4) Epoch 4, batch 1700, loss[loss=0.3738, simple_loss=0.3918, pruned_loss=0.1779, over 19702.00 frames. ], tot_loss[loss=0.3176, simple_loss=0.3497, pruned_loss=0.1427, over 3947312.06 frames. ], batch size: 315, lr: 3.28e-02, grad_scale: 8.0
+2023-03-27 18:56:03,017 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:56:07,217 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3758, 2.6705, 3.2825, 3.6788, 2.3920, 3.0858, 2.7584, 2.0438],
+       device='cuda:3'), covar=tensor([0.0360, 0.2301, 0.0623, 0.0162, 0.2167, 0.0372, 0.0680, 0.2040],
+       device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0289, 0.0160, 0.0095, 0.0208, 0.0119, 0.0148, 0.0189],
+       device='cuda:3'), out_proj_covar=tensor([1.4046e-04, 2.9430e-04, 1.7358e-04, 9.8637e-05, 2.1575e-04, 1.2425e-04,
+        1.5604e-04, 1.9900e-04], device='cuda:3')
+2023-03-27 18:56:23,725 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:56:34,829 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9353, 1.7843, 3.9507, 3.6607, 3.8928, 4.1673, 4.1576, 3.8810],
+       device='cuda:3'), covar=tensor([0.0644, 0.2698, 0.0262, 0.0342, 0.0291, 0.0114, 0.0208, 0.0382],
+       device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0145, 0.0083, 0.0078, 0.0066, 0.0068, 0.0063, 0.0068],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-27 18:57:00,731 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.185e+02 6.685e+02 7.833e+02 9.153e+02 1.678e+03, threshold=1.567e+03, percent-clipped=4.0
+2023-03-27 18:57:20,326 INFO [train.py:892] (3/4) Epoch 4, batch 1750, loss[loss=0.2375, simple_loss=0.2942, pruned_loss=0.09037, over 19781.00 frames. ], tot_loss[loss=0.313, simple_loss=0.3462, pruned_loss=0.1399, over 3948589.41 frames. ], batch size: 52, lr: 3.27e-02, grad_scale: 8.0
+2023-03-27 18:57:45,866 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8210, 1.9011, 3.9249, 3.7452, 3.8720, 4.1400, 4.2859, 4.0056],
+       device='cuda:3'), covar=tensor([0.0880, 0.2879, 0.0309, 0.0371, 0.0341, 0.0166, 0.0202, 0.0498],
+       device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0154, 0.0088, 0.0083, 0.0070, 0.0071, 0.0066, 0.0071],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-27 18:57:51,838 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:57:57,484 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 18:58:49,017 INFO [train.py:892] (3/4) Epoch 4, batch 1800, loss[loss=0.2735, simple_loss=0.3032, pruned_loss=0.1219, over 19781.00 frames. ], tot_loss[loss=0.3126, simple_loss=0.3457, pruned_loss=0.1398, over 3949415.31 frames. ], batch size: 131, lr: 3.27e-02, grad_scale: 8.0
+2023-03-27 18:59:44,746 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-27 18:59:52,487 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7404.0, num_to_drop=1, layers_to_drop={2}
+2023-03-27 18:59:53,631 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.607e+02 6.051e+02 7.381e+02 9.131e+02 1.849e+03, threshold=1.476e+03, percent-clipped=3.0
+2023-03-27 18:59:57,441 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:00:00,711 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.0421, 2.0578, 2.0280, 1.7218, 1.3038, 1.4179, 1.5920, 1.9791],
+       device='cuda:3'), covar=tensor([0.0358, 0.0453, 0.0300, 0.0380, 0.0502, 0.0645, 0.0600, 0.0839],
+       device='cuda:3'), in_proj_covar=tensor([0.0028, 0.0028, 0.0030, 0.0026, 0.0028, 0.0031, 0.0036, 0.0028],
+       device='cuda:3'), out_proj_covar=tensor([5.5489e-05, 5.3930e-05, 5.7052e-05, 5.1217e-05, 5.6225e-05, 5.9407e-05,
+        6.9274e-05, 5.5368e-05], device='cuda:3')
+2023-03-27 19:00:11,534 INFO [train.py:892] (3/4) Epoch 4, batch 1850, loss[loss=0.3118, simple_loss=0.366, pruned_loss=0.1288, over 19699.00 frames. ], tot_loss[loss=0.3128, simple_loss=0.3475, pruned_loss=0.139, over 3947741.98 frames. ], batch size: 56, lr: 3.26e-02, grad_scale: 8.0
+2023-03-27 19:01:08,060 INFO [train.py:892] (3/4) Epoch 5, batch 0, loss[loss=0.2844, simple_loss=0.3291, pruned_loss=0.1198, over 19884.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3291, pruned_loss=0.1198, over 19884.00 frames. ], batch size: 71, lr: 3.03e-02, grad_scale: 8.0
+2023-03-27 19:01:08,061 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-27 19:01:34,467 INFO [train.py:926] (3/4) Epoch 5, validation: loss=0.2154, simple_loss=0.2917, pruned_loss=0.06955, over 2883724.00 frames. 
+2023-03-27 19:01:34,468 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22042MB
+2023-03-27 19:03:19,109 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:03:24,016 INFO [train.py:892] (3/4) Epoch 5, batch 50, loss[loss=0.4357, simple_loss=0.435, pruned_loss=0.2182, over 19585.00 frames. ], tot_loss[loss=0.3019, simple_loss=0.3385, pruned_loss=0.1327, over 890787.12 frames. ], batch size: 387, lr: 3.03e-02, grad_scale: 8.0
+2023-03-27 19:04:08,908 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6735, 4.2438, 4.3056, 4.1527, 3.8594, 4.3109, 4.2279, 4.4998],
+       device='cuda:3'), covar=tensor([0.1474, 0.0268, 0.0429, 0.0300, 0.0553, 0.0337, 0.0234, 0.0330],
+       device='cuda:3'), in_proj_covar=tensor([0.0183, 0.0133, 0.0124, 0.0124, 0.0132, 0.0117, 0.0104, 0.0111],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 19:04:33,770 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.1056, 3.5641, 4.7397, 4.0527, 4.3980, 4.7343, 4.9659, 4.6983],
+       device='cuda:3'), covar=tensor([0.0080, 0.0393, 0.0085, 0.1065, 0.0112, 0.0104, 0.0084, 0.0100],
+       device='cuda:3'), in_proj_covar=tensor([0.0053, 0.0058, 0.0052, 0.0128, 0.0048, 0.0052, 0.0052, 0.0047],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0001, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-27 19:04:36,700 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.390e+02 6.285e+02 7.498e+02 8.995e+02 1.568e+03, threshold=1.500e+03, percent-clipped=1.0
+2023-03-27 19:05:06,275 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7519.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 19:05:09,579 INFO [train.py:892] (3/4) Epoch 5, batch 100, loss[loss=0.2693, simple_loss=0.3174, pruned_loss=0.1106, over 19796.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3419, pruned_loss=0.1355, over 1566467.74 frames. ], batch size: 111, lr: 3.02e-02, grad_scale: 8.0
+2023-03-27 19:05:37,876 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7534.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 19:06:28,807 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0
+2023-03-27 19:06:43,592 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:06:53,842 INFO [train.py:892] (3/4) Epoch 5, batch 150, loss[loss=0.3081, simple_loss=0.3365, pruned_loss=0.1399, over 19742.00 frames. ], tot_loss[loss=0.3106, simple_loss=0.3443, pruned_loss=0.1385, over 2094453.18 frames. ], batch size: 106, lr: 3.01e-02, grad_scale: 8.0
+2023-03-27 19:07:19,293 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7582.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 19:08:06,879 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.365e+02 5.876e+02 7.241e+02 9.334e+02 1.719e+03, threshold=1.448e+03, percent-clipped=1.0
+2023-03-27 19:08:20,976 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.69 vs. limit=5.0
+2023-03-27 19:08:24,630 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:08:42,166 INFO [train.py:892] (3/4) Epoch 5, batch 200, loss[loss=0.2995, simple_loss=0.3414, pruned_loss=0.1288, over 19735.00 frames. ], tot_loss[loss=0.3113, simple_loss=0.345, pruned_loss=0.1388, over 2503944.52 frames. ], batch size: 99, lr: 3.01e-02, grad_scale: 8.0
+2023-03-27 19:08:49,120 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:09:02,475 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:09:45,774 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-27 19:09:50,841 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6404, 3.9374, 4.1500, 3.8263, 3.8486, 4.0639, 3.8958, 4.3167],
+       device='cuda:3'), covar=tensor([0.1241, 0.0288, 0.0339, 0.0324, 0.0572, 0.0338, 0.0260, 0.0235],
+       device='cuda:3'), in_proj_covar=tensor([0.0188, 0.0136, 0.0128, 0.0128, 0.0135, 0.0121, 0.0108, 0.0115],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 19:10:28,533 INFO [train.py:892] (3/4) Epoch 5, batch 250, loss[loss=0.261, simple_loss=0.3146, pruned_loss=0.1037, over 19769.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3449, pruned_loss=0.138, over 2824964.54 frames. ], batch size: 113, lr: 3.00e-02, grad_scale: 8.0
+2023-03-27 19:10:55,924 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:11:20,178 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-27 19:11:37,127 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7704.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 19:11:38,138 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.842e+02 5.939e+02 7.915e+02 9.500e+02 2.384e+03, threshold=1.583e+03, percent-clipped=4.0
+2023-03-27 19:12:10,450 INFO [train.py:892] (3/4) Epoch 5, batch 300, loss[loss=0.2863, simple_loss=0.3504, pruned_loss=0.111, over 19530.00 frames. ], tot_loss[loss=0.3075, simple_loss=0.3431, pruned_loss=0.136, over 3071521.41 frames. ], batch size: 54, lr: 2.99e-02, grad_scale: 8.0
+2023-03-27 19:13:00,776 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:13:18,457 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7752.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 19:13:42,659 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:13:58,906 INFO [train.py:892] (3/4) Epoch 5, batch 350, loss[loss=0.3161, simple_loss=0.344, pruned_loss=0.1441, over 19722.00 frames. ], tot_loss[loss=0.3103, simple_loss=0.3453, pruned_loss=0.1376, over 3263305.81 frames. ], batch size: 219, lr: 2.98e-02, grad_scale: 8.0
+2023-03-27 19:14:29,128 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2719, 4.1503, 2.7725, 4.8108, 4.9334, 2.0839, 4.2224, 4.0259],
+       device='cuda:3'), covar=tensor([0.0443, 0.0676, 0.2142, 0.0275, 0.0072, 0.2612, 0.0590, 0.0421],
+       device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0141, 0.0171, 0.0091, 0.0071, 0.0172, 0.0161, 0.0103],
+       device='cuda:3'), out_proj_covar=tensor([1.4020e-04, 1.6064e-04, 1.8567e-04, 1.1090e-04, 8.2621e-05, 1.8107e-04,
+        1.7930e-04, 1.1442e-04], device='cuda:3')
+2023-03-27 19:15:04,466 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:15:08,676 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:15:09,458 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.882e+02 6.339e+02 7.549e+02 9.099e+02 1.830e+03, threshold=1.510e+03, percent-clipped=2.0
+2023-03-27 19:15:39,029 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7819.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 19:15:41,267 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0
+2023-03-27 19:15:42,006 INFO [train.py:892] (3/4) Epoch 5, batch 400, loss[loss=0.3491, simple_loss=0.3808, pruned_loss=0.1587, over 19711.00 frames. ], tot_loss[loss=0.3089, simple_loss=0.3447, pruned_loss=0.1366, over 3413037.55 frames. ], batch size: 265, lr: 2.98e-02, grad_scale: 8.0
+2023-03-27 19:16:10,427 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.1607, 5.4283, 5.5143, 5.4408, 5.3924, 5.4917, 4.8173, 4.9013],
+       device='cuda:3'), covar=tensor([0.0352, 0.0330, 0.0428, 0.0327, 0.0358, 0.0412, 0.0489, 0.0715],
+       device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0118, 0.0176, 0.0131, 0.0129, 0.0118, 0.0149, 0.0186],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 19:17:11,357 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:17:19,252 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:17:26,095 INFO [train.py:892] (3/4) Epoch 5, batch 450, loss[loss=0.3111, simple_loss=0.3455, pruned_loss=0.1384, over 19839.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.343, pruned_loss=0.1351, over 3532492.44 frames. ], batch size: 90, lr: 2.97e-02, grad_scale: 8.0
+2023-03-27 19:18:37,150 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.709e+02 6.480e+02 7.693e+02 9.079e+02 2.029e+03, threshold=1.539e+03, percent-clipped=2.0
+2023-03-27 19:19:12,463 INFO [train.py:892] (3/4) Epoch 5, batch 500, loss[loss=0.3046, simple_loss=0.3354, pruned_loss=0.1369, over 19872.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3425, pruned_loss=0.1351, over 3625966.33 frames. ], batch size: 138, lr: 2.96e-02, grad_scale: 8.0
+2023-03-27 19:19:34,533 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7931.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:20:57,008 INFO [train.py:892] (3/4) Epoch 5, batch 550, loss[loss=0.2789, simple_loss=0.3217, pruned_loss=0.118, over 19867.00 frames. ], tot_loss[loss=0.3046, simple_loss=0.3411, pruned_loss=0.1341, over 3697518.41 frames. ], batch size: 46, lr: 2.96e-02, grad_scale: 8.0
+2023-03-27 19:21:13,916 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:21:15,777 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:22:12,795 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.154e+02 6.076e+02 7.580e+02 9.097e+02 2.073e+03, threshold=1.516e+03, percent-clipped=3.0
+2023-03-27 19:22:44,345 INFO [train.py:892] (3/4) Epoch 5, batch 600, loss[loss=0.2874, simple_loss=0.3232, pruned_loss=0.1258, over 19669.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3407, pruned_loss=0.134, over 3754118.84 frames. ], batch size: 73, lr: 2.95e-02, grad_scale: 8.0
+2023-03-27 19:22:45,478 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1904, 4.1096, 2.7246, 4.9294, 4.9653, 2.1805, 4.2320, 3.7867],
+       device='cuda:3'), covar=tensor([0.0501, 0.0578, 0.2039, 0.0212, 0.0159, 0.2384, 0.0514, 0.0444],
+       device='cuda:3'), in_proj_covar=tensor([0.0124, 0.0142, 0.0171, 0.0092, 0.0072, 0.0173, 0.0163, 0.0105],
+       device='cuda:3'), out_proj_covar=tensor([1.4458e-04, 1.6260e-04, 1.8678e-04, 1.1278e-04, 8.4666e-05, 1.8247e-04,
+        1.8209e-04, 1.1706e-04], device='cuda:3')
+2023-03-27 19:23:39,086 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.2289, 2.4808, 2.7491, 2.0087, 2.6403, 2.5862, 2.4582, 2.9653],
+       device='cuda:3'), covar=tensor([0.0832, 0.0373, 0.0381, 0.0653, 0.0397, 0.0301, 0.0398, 0.0194],
+       device='cuda:3'), in_proj_covar=tensor([0.0042, 0.0039, 0.0041, 0.0061, 0.0040, 0.0034, 0.0036, 0.0033],
+       device='cuda:3'), out_proj_covar=tensor([9.6765e-05, 9.0013e-05, 9.3085e-05, 1.2384e-04, 9.1311e-05, 8.0302e-05,
+        8.6278e-05, 7.5765e-05], device='cuda:3')
+2023-03-27 19:24:12,736 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:24:14,669 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2131, 3.2244, 4.0377, 3.3928, 3.6104, 4.0675, 3.9809, 3.8829],
+       device='cuda:3'), covar=tensor([0.0131, 0.0424, 0.0093, 0.1218, 0.0116, 0.0168, 0.0124, 0.0129],
+       device='cuda:3'), in_proj_covar=tensor([0.0055, 0.0062, 0.0054, 0.0133, 0.0051, 0.0055, 0.0055, 0.0049],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0001, 0.0002, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-27 19:24:28,884 INFO [train.py:892] (3/4) Epoch 5, batch 650, loss[loss=0.2892, simple_loss=0.3159, pruned_loss=0.1313, over 19866.00 frames. ], tot_loss[loss=0.3038, simple_loss=0.3398, pruned_loss=0.1339, over 3796775.91 frames. ], batch size: 129, lr: 2.94e-02, grad_scale: 8.0
+2023-03-27 19:25:31,110 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:25:43,547 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.014e+02 6.583e+02 7.852e+02 9.012e+02 1.490e+03, threshold=1.570e+03, percent-clipped=0.0
+2023-03-27 19:25:50,472 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3388, 3.8387, 3.9327, 4.4032, 3.9745, 4.3524, 4.3616, 4.5473],
+       device='cuda:3'), covar=tensor([0.0532, 0.0334, 0.0415, 0.0227, 0.0490, 0.0242, 0.0334, 0.0274],
+       device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0111, 0.0125, 0.0109, 0.0108, 0.0088, 0.0113, 0.0125],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 19:25:56,285 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8111.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:26:16,767 INFO [train.py:892] (3/4) Epoch 5, batch 700, loss[loss=0.2654, simple_loss=0.3126, pruned_loss=0.1092, over 19746.00 frames. ], tot_loss[loss=0.3076, simple_loss=0.3431, pruned_loss=0.1361, over 3828346.90 frames. ], batch size: 89, lr: 2.94e-02, grad_scale: 8.0
+2023-03-27 19:27:17,366 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0
+2023-03-27 19:27:38,111 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:28:05,568 INFO [train.py:892] (3/4) Epoch 5, batch 750, loss[loss=0.2502, simple_loss=0.2927, pruned_loss=0.1039, over 19691.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.343, pruned_loss=0.1359, over 3855093.53 frames. ], batch size: 75, lr: 2.93e-02, grad_scale: 8.0
+2023-03-27 19:28:46,909 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9200, 1.8792, 3.1217, 3.0124, 3.3456, 3.5053, 3.6889, 3.5384],
+       device='cuda:3'), covar=tensor([0.0490, 0.2130, 0.0430, 0.0375, 0.0272, 0.0187, 0.0184, 0.0370],
+       device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0153, 0.0091, 0.0085, 0.0071, 0.0074, 0.0065, 0.0073],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-27 19:29:16,504 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.022e+02 6.531e+02 7.679e+02 9.153e+02 1.674e+03, threshold=1.536e+03, percent-clipped=1.0
+2023-03-27 19:29:50,716 INFO [train.py:892] (3/4) Epoch 5, batch 800, loss[loss=0.2791, simple_loss=0.3173, pruned_loss=0.1204, over 19870.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3403, pruned_loss=0.1341, over 3876936.34 frames. ], batch size: 136, lr: 2.92e-02, grad_scale: 8.0
+2023-03-27 19:31:19,697 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4877, 2.6025, 3.8324, 2.9095, 3.3005, 3.9616, 2.2926, 2.3814],
+       device='cuda:3'), covar=tensor([0.0554, 0.2730, 0.0341, 0.0528, 0.0978, 0.0298, 0.0940, 0.1424],
+       device='cuda:3'), in_proj_covar=tensor([0.0213, 0.0279, 0.0160, 0.0153, 0.0250, 0.0139, 0.0172, 0.0183],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 19:31:25,535 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:31:36,261 INFO [train.py:892] (3/4) Epoch 5, batch 850, loss[loss=0.2738, simple_loss=0.3055, pruned_loss=0.1211, over 19876.00 frames. ], tot_loss[loss=0.303, simple_loss=0.3397, pruned_loss=0.1332, over 3892867.95 frames. ], batch size: 158, lr: 2.92e-02, grad_scale: 8.0
+2023-03-27 19:31:57,977 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8280.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:32:49,196 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.707e+02 5.455e+02 6.919e+02 8.386e+02 1.759e+03, threshold=1.384e+03, percent-clipped=2.0
+2023-03-27 19:33:22,218 INFO [train.py:892] (3/4) Epoch 5, batch 900, loss[loss=0.2846, simple_loss=0.318, pruned_loss=0.1256, over 19734.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3384, pruned_loss=0.1317, over 3905119.44 frames. ], batch size: 47, lr: 2.91e-02, grad_scale: 16.0
+2023-03-27 19:33:34,361 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:33:38,042 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:35:10,587 INFO [train.py:892] (3/4) Epoch 5, batch 950, loss[loss=0.2978, simple_loss=0.326, pruned_loss=0.1348, over 19875.00 frames. ], tot_loss[loss=0.3006, simple_loss=0.3384, pruned_loss=0.1314, over 3913418.09 frames. ], batch size: 159, lr: 2.91e-02, grad_scale: 16.0
+2023-03-27 19:35:28,461 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3987, 3.1597, 4.3522, 3.5048, 3.8594, 4.2405, 4.1819, 4.1188],
+       device='cuda:3'), covar=tensor([0.0116, 0.0405, 0.0079, 0.1188, 0.0120, 0.0117, 0.0115, 0.0099],
+       device='cuda:3'), in_proj_covar=tensor([0.0055, 0.0064, 0.0055, 0.0132, 0.0050, 0.0055, 0.0056, 0.0049],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0001, 0.0003, 0.0001, 0.0002, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-27 19:36:10,071 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:36:10,283 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1005, 3.2179, 2.1928, 3.4367, 3.5746, 1.5334, 2.8063, 2.7805],
+       device='cuda:3'), covar=tensor([0.0580, 0.0614, 0.2099, 0.0293, 0.0130, 0.2680, 0.0828, 0.0460],
+       device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0148, 0.0176, 0.0100, 0.0072, 0.0174, 0.0172, 0.0109],
+       device='cuda:3'), out_proj_covar=tensor([1.5182e-04, 1.7017e-04, 1.9326e-04, 1.2339e-04, 8.6477e-05, 1.8621e-04,
+        1.9357e-04, 1.2115e-04], device='cuda:3')
+2023-03-27 19:36:23,191 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.522e+02 6.186e+02 7.459e+02 8.919e+02 2.032e+03, threshold=1.492e+03, percent-clipped=3.0
+2023-03-27 19:36:55,732 INFO [train.py:892] (3/4) Epoch 5, batch 1000, loss[loss=0.2802, simple_loss=0.3135, pruned_loss=0.1234, over 19822.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3375, pruned_loss=0.1304, over 3921836.38 frames. ], batch size: 147, lr: 2.90e-02, grad_scale: 16.0
+2023-03-27 19:37:53,238 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:37:53,696 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0
+2023-03-27 19:38:15,153 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:38:41,299 INFO [train.py:892] (3/4) Epoch 5, batch 1050, loss[loss=0.3257, simple_loss=0.3582, pruned_loss=0.1466, over 19644.00 frames. ], tot_loss[loss=0.2992, simple_loss=0.3375, pruned_loss=0.1304, over 3928298.82 frames. ], batch size: 68, lr: 2.89e-02, grad_scale: 16.0
+2023-03-27 19:39:52,154 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.605e+02 6.353e+02 7.329e+02 8.982e+02 1.532e+03, threshold=1.466e+03, percent-clipped=3.0
+2023-03-27 19:39:56,408 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:40:27,774 INFO [train.py:892] (3/4) Epoch 5, batch 1100, loss[loss=0.3005, simple_loss=0.3494, pruned_loss=0.1258, over 19701.00 frames. ], tot_loss[loss=0.2988, simple_loss=0.337, pruned_loss=0.1303, over 3932754.77 frames. ], batch size: 85, lr: 2.89e-02, grad_scale: 16.0
+2023-03-27 19:42:10,851 INFO [train.py:892] (3/4) Epoch 5, batch 1150, loss[loss=0.285, simple_loss=0.3176, pruned_loss=0.1262, over 19885.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3365, pruned_loss=0.1299, over 3937292.78 frames. ], batch size: 77, lr: 2.88e-02, grad_scale: 16.0
+2023-03-27 19:43:21,775 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.231e+02 6.114e+02 7.456e+02 8.704e+02 1.962e+03, threshold=1.491e+03, percent-clipped=1.0
+2023-03-27 19:43:56,823 INFO [train.py:892] (3/4) Epoch 5, batch 1200, loss[loss=0.2761, simple_loss=0.3299, pruned_loss=0.1112, over 19803.00 frames. ], tot_loss[loss=0.2963, simple_loss=0.3351, pruned_loss=0.1288, over 3941010.44 frames. ], batch size: 67, lr: 2.87e-02, grad_scale: 16.0
+2023-03-27 19:43:57,679 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:45:41,163 INFO [train.py:892] (3/4) Epoch 5, batch 1250, loss[loss=0.3756, simple_loss=0.3818, pruned_loss=0.1847, over 19653.00 frames. ], tot_loss[loss=0.2955, simple_loss=0.3344, pruned_loss=0.1283, over 3941899.39 frames. ], batch size: 299, lr: 2.87e-02, grad_scale: 16.0
+2023-03-27 19:46:21,680 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0497, 4.4370, 4.9283, 4.6945, 4.9999, 3.3522, 3.8304, 3.8659],
+       device='cuda:3'), covar=tensor([0.0243, 0.0199, 0.0163, 0.0164, 0.0196, 0.0722, 0.1071, 0.0585],
+       device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0072, 0.0069, 0.0078, 0.0070, 0.0095, 0.0105, 0.0085],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 19:46:22,090 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-27 19:46:53,148 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.783e+02 6.822e+02 8.338e+02 1.055e+03 2.150e+03, threshold=1.668e+03, percent-clipped=7.0
+2023-03-27 19:47:24,981 INFO [train.py:892] (3/4) Epoch 5, batch 1300, loss[loss=0.2627, simple_loss=0.3056, pruned_loss=0.1099, over 19742.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3357, pruned_loss=0.1291, over 3942285.52 frames. ], batch size: 95, lr: 2.86e-02, grad_scale: 16.0
+2023-03-27 19:48:48,202 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:49:09,896 INFO [train.py:892] (3/4) Epoch 5, batch 1350, loss[loss=0.3215, simple_loss=0.3488, pruned_loss=0.1471, over 19778.00 frames. ], tot_loss[loss=0.2981, simple_loss=0.3366, pruned_loss=0.1298, over 3942983.86 frames. ], batch size: 247, lr: 2.86e-02, grad_scale: 16.0
+2023-03-27 19:49:48,269 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.1497, 1.4606, 2.4348, 2.5876, 2.5990, 2.9792, 2.9426, 2.9047],
+       device='cuda:3'), covar=tensor([0.0815, 0.2291, 0.0575, 0.0422, 0.0402, 0.0203, 0.0227, 0.0380],
+       device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0154, 0.0097, 0.0090, 0.0075, 0.0076, 0.0068, 0.0076],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-27 19:50:22,594 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.065e+02 5.700e+02 7.084e+02 8.473e+02 1.458e+03, threshold=1.417e+03, percent-clipped=0.0
+2023-03-27 19:50:58,491 INFO [train.py:892] (3/4) Epoch 5, batch 1400, loss[loss=0.2725, simple_loss=0.3118, pruned_loss=0.1165, over 19838.00 frames. ], tot_loss[loss=0.2964, simple_loss=0.3347, pruned_loss=0.129, over 3945031.46 frames. ], batch size: 144, lr: 2.85e-02, grad_scale: 16.0
+2023-03-27 19:50:59,410 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:52:11,149 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.56 vs. limit=5.0
+2023-03-27 19:52:40,280 INFO [train.py:892] (3/4) Epoch 5, batch 1450, loss[loss=0.3211, simple_loss=0.3545, pruned_loss=0.1439, over 19750.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.3351, pruned_loss=0.1295, over 3945413.48 frames. ], batch size: 250, lr: 2.84e-02, grad_scale: 16.0
+2023-03-27 19:52:44,955 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.4819, 1.4914, 1.4798, 1.3387, 0.9694, 1.2104, 1.2508, 1.4225],
+       device='cuda:3'), covar=tensor([0.0330, 0.0329, 0.0260, 0.0268, 0.0468, 0.0455, 0.0523, 0.0409],
+       device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0033, 0.0033, 0.0027, 0.0033, 0.0034, 0.0040, 0.0031],
+       device='cuda:3'), out_proj_covar=tensor([6.4566e-05, 6.6332e-05, 6.6740e-05, 5.6404e-05, 6.8985e-05, 6.9620e-05,
+        7.9971e-05, 6.6434e-05], device='cuda:3')
+2023-03-27 19:53:29,657 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5688, 3.7428, 4.0903, 3.7206, 3.6139, 3.8752, 3.6920, 4.1659],
+       device='cuda:3'), covar=tensor([0.1242, 0.0343, 0.0335, 0.0331, 0.0781, 0.0422, 0.0333, 0.0268],
+       device='cuda:3'), in_proj_covar=tensor([0.0200, 0.0149, 0.0141, 0.0141, 0.0148, 0.0133, 0.0122, 0.0125],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 19:53:52,859 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.344e+02 6.588e+02 7.767e+02 1.033e+03 1.538e+03, threshold=1.553e+03, percent-clipped=3.0
+2023-03-27 19:54:13,107 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2248, 3.3024, 3.9101, 4.6206, 2.7813, 3.5367, 3.1487, 2.5258],
+       device='cuda:3'), covar=tensor([0.0344, 0.2944, 0.0636, 0.0090, 0.2402, 0.0507, 0.0745, 0.1895],
+       device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0309, 0.0180, 0.0101, 0.0219, 0.0132, 0.0160, 0.0195],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 19:54:26,588 INFO [train.py:892] (3/4) Epoch 5, batch 1500, loss[loss=0.3625, simple_loss=0.3876, pruned_loss=0.1686, over 19624.00 frames. ], tot_loss[loss=0.2972, simple_loss=0.3354, pruned_loss=0.1295, over 3944813.55 frames. ], batch size: 367, lr: 2.84e-02, grad_scale: 16.0
+2023-03-27 19:54:27,700 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:54:40,783 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8927.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:55:00,856 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.5365, 1.4889, 1.5761, 1.4848, 1.0589, 1.3532, 1.3574, 1.6897],
+       device='cuda:3'), covar=tensor([0.0180, 0.0302, 0.0278, 0.0242, 0.0375, 0.0366, 0.0434, 0.0259],
+       device='cuda:3'), in_proj_covar=tensor([0.0031, 0.0034, 0.0033, 0.0027, 0.0033, 0.0034, 0.0041, 0.0031],
+       device='cuda:3'), out_proj_covar=tensor([6.6004e-05, 6.9415e-05, 6.7448e-05, 5.6941e-05, 6.9575e-05, 7.0937e-05,
+        8.3338e-05, 6.6333e-05], device='cuda:3')
+2023-03-27 19:55:52,504 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9481, 2.8310, 4.0748, 3.1370, 3.5422, 4.2261, 2.4747, 2.5625],
+       device='cuda:3'), covar=tensor([0.0459, 0.2739, 0.0290, 0.0491, 0.1048, 0.0241, 0.1026, 0.1453],
+       device='cuda:3'), in_proj_covar=tensor([0.0221, 0.0277, 0.0163, 0.0157, 0.0252, 0.0141, 0.0178, 0.0186],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 19:56:10,556 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:56:13,846 INFO [train.py:892] (3/4) Epoch 5, batch 1550, loss[loss=0.3131, simple_loss=0.3515, pruned_loss=0.1373, over 19831.00 frames. ], tot_loss[loss=0.2997, simple_loss=0.3375, pruned_loss=0.131, over 3943702.84 frames. ], batch size: 76, lr: 2.83e-02, grad_scale: 16.0
+2023-03-27 19:56:45,276 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:56:50,796 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:57:26,090 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.607e+02 5.861e+02 6.933e+02 8.857e+02 2.450e+03, threshold=1.387e+03, percent-clipped=2.0
+2023-03-27 19:57:42,836 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:57:57,841 INFO [train.py:892] (3/4) Epoch 5, batch 1600, loss[loss=0.2676, simple_loss=0.3153, pruned_loss=0.1099, over 19872.00 frames. ], tot_loss[loss=0.2979, simple_loss=0.3367, pruned_loss=0.1295, over 3944584.56 frames. ], batch size: 108, lr: 2.83e-02, grad_scale: 16.0
+2023-03-27 19:57:59,237 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0
+2023-03-27 19:58:52,903 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 19:59:40,270 INFO [train.py:892] (3/4) Epoch 5, batch 1650, loss[loss=0.3038, simple_loss=0.3275, pruned_loss=0.1401, over 19871.00 frames. ], tot_loss[loss=0.2977, simple_loss=0.3365, pruned_loss=0.1295, over 3945864.69 frames. ], batch size: 158, lr: 2.82e-02, grad_scale: 16.0
+2023-03-27 19:59:48,673 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:00:51,667 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.078e+02 6.548e+02 7.778e+02 9.936e+02 1.704e+03, threshold=1.556e+03, percent-clipped=1.0
+2023-03-27 20:01:16,370 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:01:25,585 INFO [train.py:892] (3/4) Epoch 5, batch 1700, loss[loss=0.2873, simple_loss=0.3195, pruned_loss=0.1275, over 19789.00 frames. ], tot_loss[loss=0.2936, simple_loss=0.3331, pruned_loss=0.1271, over 3948370.73 frames. ], batch size: 172, lr: 2.81e-02, grad_scale: 16.0
+2023-03-27 20:02:55,527 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9379, 4.5394, 4.7540, 4.6235, 4.9006, 3.2602, 3.9477, 3.3481],
+       device='cuda:3'), covar=tensor([0.0137, 0.0113, 0.0108, 0.0102, 0.0097, 0.0569, 0.0788, 0.0655],
+       device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0073, 0.0073, 0.0080, 0.0072, 0.0094, 0.0105, 0.0089],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 20:03:04,914 INFO [train.py:892] (3/4) Epoch 5, batch 1750, loss[loss=0.2722, simple_loss=0.3113, pruned_loss=0.1166, over 19850.00 frames. ], tot_loss[loss=0.2934, simple_loss=0.3329, pruned_loss=0.1269, over 3948738.61 frames. ], batch size: 112, lr: 2.81e-02, grad_scale: 16.0
+2023-03-27 20:03:46,386 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1053, 4.3401, 4.7289, 4.3685, 4.1777, 4.4113, 4.3485, 4.9303],
+       device='cuda:3'), covar=tensor([0.1055, 0.0281, 0.0317, 0.0250, 0.0478, 0.0320, 0.0257, 0.0195],
+       device='cuda:3'), in_proj_covar=tensor([0.0200, 0.0147, 0.0140, 0.0141, 0.0145, 0.0134, 0.0123, 0.0122],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 20:04:04,132 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6963, 4.1897, 4.2747, 4.7076, 4.4604, 4.8370, 4.7506, 4.9801],
+       device='cuda:3'), covar=tensor([0.0515, 0.0320, 0.0376, 0.0273, 0.0392, 0.0177, 0.0321, 0.0243],
+       device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0115, 0.0135, 0.0118, 0.0112, 0.0090, 0.0121, 0.0131],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 20:04:05,087 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.671e+02 6.292e+02 7.596e+02 9.035e+02 1.874e+03, threshold=1.519e+03, percent-clipped=3.0
+2023-03-27 20:04:33,322 INFO [train.py:892] (3/4) Epoch 5, batch 1800, loss[loss=0.285, simple_loss=0.3277, pruned_loss=0.1211, over 19923.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3312, pruned_loss=0.1259, over 3949745.97 frames. ], batch size: 51, lr: 2.80e-02, grad_scale: 16.0
+2023-03-27 20:05:59,358 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4599, 2.7328, 2.1501, 1.7110, 2.0954, 2.5983, 2.3211, 2.8228],
+       device='cuda:3'), covar=tensor([0.0169, 0.0261, 0.0240, 0.0572, 0.0358, 0.0188, 0.0171, 0.0125],
+       device='cuda:3'), in_proj_covar=tensor([0.0037, 0.0035, 0.0037, 0.0052, 0.0051, 0.0033, 0.0027, 0.0031],
+       device='cuda:3'), out_proj_covar=tensor([7.6648e-05, 7.5041e-05, 7.5437e-05, 1.1102e-04, 1.0757e-04, 7.2622e-05,
+        6.0043e-05, 6.6820e-05], device='cuda:3')
+2023-03-27 20:06:00,368 INFO [train.py:892] (3/4) Epoch 5, batch 1850, loss[loss=0.2733, simple_loss=0.3267, pruned_loss=0.11, over 19676.00 frames. ], tot_loss[loss=0.2906, simple_loss=0.3311, pruned_loss=0.125, over 3950321.88 frames. ], batch size: 55, lr: 2.80e-02, grad_scale: 16.0
+2023-03-27 20:06:59,473 INFO [train.py:892] (3/4) Epoch 6, batch 0, loss[loss=0.3368, simple_loss=0.3636, pruned_loss=0.155, over 19690.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3636, pruned_loss=0.155, over 19690.00 frames. ], batch size: 337, lr: 2.61e-02, grad_scale: 16.0
+2023-03-27 20:06:59,473 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-27 20:07:11,639 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4984, 4.2688, 4.2868, 4.1158, 4.5206, 3.2610, 3.5582, 3.5638],
+       device='cuda:3'), covar=tensor([0.0151, 0.0151, 0.0153, 0.0159, 0.0116, 0.0712, 0.0938, 0.0442],
+       device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0075, 0.0075, 0.0083, 0.0075, 0.0098, 0.0111, 0.0093],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 20:07:25,896 INFO [train.py:926] (3/4) Epoch 6, validation: loss=0.2048, simple_loss=0.2829, pruned_loss=0.06328, over 2883724.00 frames. 
+2023-03-27 20:07:25,897 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22042MB
+2023-03-27 20:07:45,222 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:08:30,069 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.239e+02 6.201e+02 7.498e+02 9.060e+02 1.792e+03, threshold=1.500e+03, percent-clipped=2.0
+2023-03-27 20:09:13,557 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5395, 4.8305, 4.8011, 4.7981, 4.5124, 4.7353, 4.2875, 4.3510],
+       device='cuda:3'), covar=tensor([0.0369, 0.0318, 0.0632, 0.0404, 0.0603, 0.0609, 0.0527, 0.0935],
+       device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0133, 0.0191, 0.0144, 0.0141, 0.0125, 0.0162, 0.0199],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 20:09:14,656 INFO [train.py:892] (3/4) Epoch 6, batch 50, loss[loss=0.2734, simple_loss=0.3097, pruned_loss=0.1185, over 19822.00 frames. ], tot_loss[loss=0.2971, simple_loss=0.332, pruned_loss=0.1311, over 890093.58 frames. ], batch size: 187, lr: 2.60e-02, grad_scale: 16.0
+2023-03-27 20:09:48,467 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9342.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:10:46,088 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9369.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:10:59,021 INFO [train.py:892] (3/4) Epoch 6, batch 100, loss[loss=0.2964, simple_loss=0.3361, pruned_loss=0.1283, over 19808.00 frames. ], tot_loss[loss=0.2889, simple_loss=0.3282, pruned_loss=0.1248, over 1568268.60 frames. ], batch size: 96, lr: 2.60e-02, grad_scale: 16.0
+2023-03-27 20:11:09,604 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5244, 3.5899, 3.9203, 4.8542, 2.8932, 3.4268, 3.1304, 2.5272],
+       device='cuda:3'), covar=tensor([0.0268, 0.2774, 0.0600, 0.0079, 0.2124, 0.0553, 0.0869, 0.2013],
+       device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0311, 0.0181, 0.0100, 0.0216, 0.0135, 0.0164, 0.0193],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 20:11:39,367 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9554, 3.7098, 4.8292, 4.0234, 4.1714, 4.7137, 4.7598, 4.4761],
+       device='cuda:3'), covar=tensor([0.0130, 0.0398, 0.0089, 0.1047, 0.0123, 0.0143, 0.0108, 0.0105],
+       device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0066, 0.0056, 0.0131, 0.0051, 0.0058, 0.0058, 0.0050],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0001, 0.0002, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-27 20:12:00,062 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.134e+02 5.626e+02 6.778e+02 8.697e+02 1.693e+03, threshold=1.356e+03, percent-clipped=2.0
+2023-03-27 20:12:16,596 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4358, 2.7740, 3.2833, 3.4295, 3.9177, 3.8375, 4.3788, 4.6764],
+       device='cuda:3'), covar=tensor([0.0330, 0.1412, 0.0925, 0.1371, 0.0991, 0.0821, 0.0222, 0.0237],
+       device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0175, 0.0170, 0.0195, 0.0185, 0.0173, 0.0113, 0.0117],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-27 20:12:20,607 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:12:24,349 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:12:43,778 INFO [train.py:892] (3/4) Epoch 6, batch 150, loss[loss=0.3097, simple_loss=0.3434, pruned_loss=0.1381, over 19750.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3294, pruned_loss=0.1243, over 2096149.98 frames. ], batch size: 250, lr: 2.59e-02, grad_scale: 16.0
+2023-03-27 20:14:08,397 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:14:32,757 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:14:33,666 INFO [train.py:892] (3/4) Epoch 6, batch 200, loss[loss=0.2478, simple_loss=0.2972, pruned_loss=0.09916, over 19861.00 frames. ], tot_loss[loss=0.2853, simple_loss=0.3266, pruned_loss=0.122, over 2506616.20 frames. ], batch size: 99, lr: 2.59e-02, grad_scale: 16.0
+2023-03-27 20:15:34,371 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.804e+02 5.747e+02 7.100e+02 9.043e+02 1.475e+03, threshold=1.420e+03, percent-clipped=5.0
+2023-03-27 20:16:16,355 INFO [train.py:892] (3/4) Epoch 6, batch 250, loss[loss=0.378, simple_loss=0.3918, pruned_loss=0.1821, over 19613.00 frames. ], tot_loss[loss=0.2872, simple_loss=0.328, pruned_loss=0.1232, over 2825071.28 frames. ], batch size: 367, lr: 2.58e-02, grad_scale: 16.0
+2023-03-27 20:18:00,749 INFO [train.py:892] (3/4) Epoch 6, batch 300, loss[loss=0.2409, simple_loss=0.2935, pruned_loss=0.09412, over 19530.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3267, pruned_loss=0.1218, over 3073667.34 frames. ], batch size: 46, lr: 2.58e-02, grad_scale: 16.0
+2023-03-27 20:18:19,784 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:18:21,695 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:19:02,336 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.913e+02 5.609e+02 7.067e+02 9.111e+02 1.524e+03, threshold=1.413e+03, percent-clipped=2.0
+2023-03-27 20:19:49,282 INFO [train.py:892] (3/4) Epoch 6, batch 350, loss[loss=0.2191, simple_loss=0.2758, pruned_loss=0.08117, over 19863.00 frames. ], tot_loss[loss=0.2877, simple_loss=0.3288, pruned_loss=0.1233, over 3267809.39 frames. ], batch size: 46, lr: 2.57e-02, grad_scale: 16.0
+2023-03-27 20:20:00,201 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9631.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:20:22,995 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9642.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:20:29,067 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9645.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:20:57,522 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:21:18,819 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:21:32,070 INFO [train.py:892] (3/4) Epoch 6, batch 400, loss[loss=0.2473, simple_loss=0.2981, pruned_loss=0.0982, over 19864.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3309, pruned_loss=0.1239, over 3415562.71 frames. ], batch size: 99, lr: 2.57e-02, grad_scale: 16.0
+2023-03-27 20:22:03,528 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9690.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:22:32,708 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.696e+02 5.768e+02 7.171e+02 9.130e+02 1.746e+03, threshold=1.434e+03, percent-clipped=4.0
+2023-03-27 20:22:58,074 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:23:02,289 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9719.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:23:16,552 INFO [train.py:892] (3/4) Epoch 6, batch 450, loss[loss=0.2775, simple_loss=0.3282, pruned_loss=0.1134, over 19892.00 frames. ], tot_loss[loss=0.2908, simple_loss=0.3319, pruned_loss=0.1248, over 3532920.25 frames. ], batch size: 63, lr: 2.56e-02, grad_scale: 16.0
+2023-03-27 20:23:54,149 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:24:12,571 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.1890, 4.7612, 4.7585, 5.3038, 4.7815, 5.5246, 5.0480, 5.4271],
+       device='cuda:3'), covar=tensor([0.0602, 0.0347, 0.0504, 0.0233, 0.0776, 0.0225, 0.0540, 0.0535],
+       device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0111, 0.0132, 0.0111, 0.0110, 0.0089, 0.0112, 0.0129],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 20:24:50,382 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:25:02,464 INFO [train.py:892] (3/4) Epoch 6, batch 500, loss[loss=0.2959, simple_loss=0.3289, pruned_loss=0.1315, over 19744.00 frames. ], tot_loss[loss=0.2875, simple_loss=0.3292, pruned_loss=0.1229, over 3625655.21 frames. ], batch size: 139, lr: 2.56e-02, grad_scale: 16.0
+2023-03-27 20:26:01,103 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:26:01,917 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.739e+02 6.410e+02 7.467e+02 9.319e+02 1.836e+03, threshold=1.493e+03, percent-clipped=5.0
+2023-03-27 20:26:44,452 INFO [train.py:892] (3/4) Epoch 6, batch 550, loss[loss=0.2335, simple_loss=0.2919, pruned_loss=0.0876, over 19907.00 frames. ], tot_loss[loss=0.289, simple_loss=0.3302, pruned_loss=0.1239, over 3696787.59 frames. ], batch size: 45, lr: 2.55e-02, grad_scale: 16.0
+2023-03-27 20:26:51,172 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.3884, 1.3424, 2.3576, 2.6373, 2.9151, 2.9689, 2.8729, 3.0986],
+       device='cuda:3'), covar=tensor([0.0613, 0.2366, 0.0625, 0.0436, 0.0345, 0.0200, 0.0251, 0.0262],
+       device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0159, 0.0105, 0.0099, 0.0082, 0.0077, 0.0074, 0.0079],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-27 20:28:21,384 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7243, 3.5674, 4.6573, 3.9484, 4.0049, 4.5177, 4.5264, 4.1757],
+       device='cuda:3'), covar=tensor([0.0090, 0.0368, 0.0064, 0.0955, 0.0093, 0.0117, 0.0088, 0.0101],
+       device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0068, 0.0057, 0.0132, 0.0052, 0.0060, 0.0059, 0.0051],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0003, 0.0001, 0.0002, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-27 20:28:30,662 INFO [train.py:892] (3/4) Epoch 6, batch 600, loss[loss=0.4902, simple_loss=0.4803, pruned_loss=0.2501, over 19402.00 frames. ], tot_loss[loss=0.2886, simple_loss=0.3299, pruned_loss=0.1236, over 3751836.89 frames. ], batch size: 431, lr: 2.54e-02, grad_scale: 16.0
+2023-03-27 20:29:32,379 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.880e+02 5.938e+02 7.019e+02 8.775e+02 1.647e+03, threshold=1.404e+03, percent-clipped=2.0
+2023-03-27 20:30:16,202 INFO [train.py:892] (3/4) Epoch 6, batch 650, loss[loss=0.2851, simple_loss=0.3342, pruned_loss=0.118, over 19765.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3271, pruned_loss=0.1214, over 3795664.47 frames. ], batch size: 217, lr: 2.54e-02, grad_scale: 16.0
+2023-03-27 20:30:45,657 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:30:51,430 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4397, 2.6624, 2.9324, 3.5637, 2.0785, 2.7621, 2.3706, 1.9938],
+       device='cuda:3'), covar=tensor([0.0380, 0.2219, 0.0852, 0.0187, 0.2503, 0.0512, 0.1083, 0.2041],
+       device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0312, 0.0186, 0.0103, 0.0216, 0.0133, 0.0169, 0.0195],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 20:32:00,347 INFO [train.py:892] (3/4) Epoch 6, batch 700, loss[loss=0.2697, simple_loss=0.3194, pruned_loss=0.11, over 19561.00 frames. ], tot_loss[loss=0.2851, simple_loss=0.3268, pruned_loss=0.1217, over 3829966.62 frames. ], batch size: 47, lr: 2.53e-02, grad_scale: 16.0
+2023-03-27 20:32:33,057 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4560, 4.6889, 5.1114, 4.6557, 4.2868, 4.8347, 4.6540, 5.2989],
+       device='cuda:3'), covar=tensor([0.1239, 0.0288, 0.0241, 0.0317, 0.0597, 0.0316, 0.0279, 0.0196],
+       device='cuda:3'), in_proj_covar=tensor([0.0211, 0.0153, 0.0149, 0.0148, 0.0152, 0.0143, 0.0132, 0.0130],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 20:33:05,036 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.264e+02 6.107e+02 7.493e+02 9.243e+02 1.709e+03, threshold=1.499e+03, percent-clipped=3.0
+2023-03-27 20:33:24,828 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10014.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:33:40,023 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5783, 2.0882, 2.5600, 2.7715, 2.9548, 2.7278, 3.5472, 3.4327],
+       device='cuda:3'), covar=tensor([0.0477, 0.1622, 0.1188, 0.1559, 0.1174, 0.1222, 0.0312, 0.0394],
+       device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0178, 0.0174, 0.0196, 0.0191, 0.0175, 0.0118, 0.0118],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-27 20:33:49,827 INFO [train.py:892] (3/4) Epoch 6, batch 750, loss[loss=0.2795, simple_loss=0.3301, pruned_loss=0.1145, over 19803.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3262, pruned_loss=0.1214, over 3856902.39 frames. ], batch size: 72, lr: 2.53e-02, grad_scale: 16.0
+2023-03-27 20:34:17,138 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10039.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:34:51,532 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7389, 2.8534, 4.4232, 2.9075, 3.5390, 4.4077, 2.5158, 2.6371],
+       device='cuda:3'), covar=tensor([0.0743, 0.3219, 0.0310, 0.0770, 0.1387, 0.0271, 0.1198, 0.1712],
+       device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0290, 0.0182, 0.0168, 0.0269, 0.0156, 0.0196, 0.0202],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 20:35:03,085 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0
+2023-03-27 20:35:04,779 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10062.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 20:35:13,497 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6714, 3.5956, 4.2886, 5.2311, 3.2416, 4.0224, 3.3869, 2.9210],
+       device='cuda:3'), covar=tensor([0.0331, 0.3055, 0.0617, 0.0063, 0.2060, 0.0440, 0.0818, 0.1818],
+       device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0316, 0.0188, 0.0104, 0.0216, 0.0134, 0.0168, 0.0196],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 20:35:23,382 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:35:34,028 INFO [train.py:892] (3/4) Epoch 6, batch 800, loss[loss=0.2509, simple_loss=0.2961, pruned_loss=0.1028, over 19841.00 frames. ], tot_loss[loss=0.2839, simple_loss=0.3257, pruned_loss=0.121, over 3878352.32 frames. ], batch size: 115, lr: 2.52e-02, grad_scale: 16.0
+2023-03-27 20:36:21,634 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:36:25,341 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10100.0, num_to_drop=1, layers_to_drop={3}
+2023-03-27 20:36:35,380 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.822e+02 5.825e+02 7.423e+02 9.383e+02 2.108e+03, threshold=1.485e+03, percent-clipped=8.0
+2023-03-27 20:37:01,074 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:37:13,079 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10123.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 20:37:17,767 INFO [train.py:892] (3/4) Epoch 6, batch 850, loss[loss=0.2976, simple_loss=0.33, pruned_loss=0.1325, over 19739.00 frames. ], tot_loss[loss=0.2837, simple_loss=0.3254, pruned_loss=0.1209, over 3894622.94 frames. ], batch size: 63, lr: 2.52e-02, grad_scale: 16.0
+2023-03-27 20:39:04,581 INFO [train.py:892] (3/4) Epoch 6, batch 900, loss[loss=0.2601, simple_loss=0.307, pruned_loss=0.1066, over 19738.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3237, pruned_loss=0.1191, over 3908211.81 frames. ], batch size: 80, lr: 2.51e-02, grad_scale: 16.0
+2023-03-27 20:39:52,860 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.5034, 1.6380, 1.3300, 0.9265, 1.3963, 1.6845, 1.5683, 1.6003],
+       device='cuda:3'), covar=tensor([0.0228, 0.0171, 0.0220, 0.0628, 0.0505, 0.0176, 0.0145, 0.0182],
+       device='cuda:3'), in_proj_covar=tensor([0.0039, 0.0038, 0.0039, 0.0056, 0.0056, 0.0037, 0.0031, 0.0034],
+       device='cuda:3'), out_proj_covar=tensor([8.4064e-05, 8.3931e-05, 8.2891e-05, 1.2316e-04, 1.2212e-04, 8.1957e-05,
+        7.0401e-05, 7.4330e-05], device='cuda:3')
+2023-03-27 20:40:03,582 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.864e+02 5.709e+02 6.647e+02 8.458e+02 1.607e+03, threshold=1.329e+03, percent-clipped=2.0
+2023-03-27 20:40:48,893 INFO [train.py:892] (3/4) Epoch 6, batch 950, loss[loss=0.3186, simple_loss=0.3513, pruned_loss=0.143, over 19875.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3249, pruned_loss=0.1197, over 3919152.66 frames. ], batch size: 158, lr: 2.51e-02, grad_scale: 16.0
+2023-03-27 20:40:56,547 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0
+2023-03-27 20:41:18,889 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:42:33,726 INFO [train.py:892] (3/4) Epoch 6, batch 1000, loss[loss=0.3181, simple_loss=0.3422, pruned_loss=0.147, over 19759.00 frames. ], tot_loss[loss=0.2812, simple_loss=0.3238, pruned_loss=0.1193, over 3925923.29 frames. ], batch size: 233, lr: 2.50e-02, grad_scale: 16.0
+2023-03-27 20:43:00,908 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10288.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:43:22,025 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6036, 3.2381, 3.2240, 3.5550, 3.3642, 3.2961, 3.5948, 3.7426],
+       device='cuda:3'), covar=tensor([0.0550, 0.0394, 0.0462, 0.0295, 0.0474, 0.0600, 0.0366, 0.0342],
+       device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0121, 0.0140, 0.0120, 0.0114, 0.0096, 0.0121, 0.0136],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 20:43:34,992 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.462e+02 6.368e+02 7.709e+02 9.244e+02 1.500e+03, threshold=1.542e+03, percent-clipped=2.0
+2023-03-27 20:43:55,009 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:44:20,095 INFO [train.py:892] (3/4) Epoch 6, batch 1050, loss[loss=0.2615, simple_loss=0.3136, pruned_loss=0.1047, over 19842.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.324, pruned_loss=0.1189, over 3931850.28 frames. ], batch size: 43, lr: 2.50e-02, grad_scale: 16.0
+2023-03-27 20:44:51,682 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9781, 2.0366, 3.3403, 3.3799, 3.7291, 4.1960, 4.1080, 4.2267],
+       device='cuda:3'), covar=tensor([0.0698, 0.2428, 0.0612, 0.0466, 0.0276, 0.0120, 0.0188, 0.0297],
+       device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0164, 0.0110, 0.0103, 0.0085, 0.0082, 0.0077, 0.0081],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-27 20:45:36,759 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:46:05,043 INFO [train.py:892] (3/4) Epoch 6, batch 1100, loss[loss=0.279, simple_loss=0.3185, pruned_loss=0.1198, over 19795.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3237, pruned_loss=0.1184, over 3937209.19 frames. ], batch size: 185, lr: 2.49e-02, grad_scale: 16.0
+2023-03-27 20:46:47,224 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10395.0, num_to_drop=1, layers_to_drop={2}
+2023-03-27 20:46:55,652 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:47:09,515 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.844e+02 6.468e+02 7.471e+02 8.887e+02 1.453e+03, threshold=1.494e+03, percent-clipped=0.0
+2023-03-27 20:47:16,170 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:47:35,391 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10418.0, num_to_drop=1, layers_to_drop={2}
+2023-03-27 20:47:51,061 INFO [train.py:892] (3/4) Epoch 6, batch 1150, loss[loss=0.3081, simple_loss=0.3353, pruned_loss=0.1404, over 19765.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3258, pruned_loss=0.1198, over 3938369.32 frames. ], batch size: 226, lr: 2.49e-02, grad_scale: 16.0
+2023-03-27 20:48:36,940 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:48:51,808 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:49:16,967 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4954, 4.7825, 5.1454, 4.7829, 4.3749, 4.9000, 4.8117, 5.3934],
+       device='cuda:3'), covar=tensor([0.1362, 0.0313, 0.0329, 0.0325, 0.0535, 0.0339, 0.0290, 0.0271],
+       device='cuda:3'), in_proj_covar=tensor([0.0210, 0.0153, 0.0146, 0.0147, 0.0151, 0.0143, 0.0131, 0.0133],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 20:49:24,775 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:49:35,675 INFO [train.py:892] (3/4) Epoch 6, batch 1200, loss[loss=0.2972, simple_loss=0.3429, pruned_loss=0.1258, over 19802.00 frames. ], tot_loss[loss=0.283, simple_loss=0.3261, pruned_loss=0.12, over 3941045.65 frames. ], batch size: 45, lr: 2.49e-02, grad_scale: 8.0
+2023-03-27 20:50:07,350 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.1718, 2.0027, 2.1694, 1.9959, 1.7267, 1.9108, 1.8307, 2.1477],
+       device='cuda:3'), covar=tensor([0.0221, 0.0309, 0.0275, 0.0285, 0.0359, 0.0383, 0.0414, 0.0498],
+       device='cuda:3'), in_proj_covar=tensor([0.0032, 0.0034, 0.0034, 0.0028, 0.0036, 0.0035, 0.0043, 0.0033],
+       device='cuda:3'), out_proj_covar=tensor([7.0019e-05, 7.3088e-05, 7.3518e-05, 5.9920e-05, 7.7968e-05, 7.6567e-05,
+        9.2290e-05, 7.0363e-05], device='cuda:3')
+2023-03-27 20:50:41,661 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.428e+02 5.589e+02 6.626e+02 8.703e+02 1.747e+03, threshold=1.325e+03, percent-clipped=3.0
+2023-03-27 20:50:58,502 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10515.0, num_to_drop=1, layers_to_drop={3}
+2023-03-27 20:51:21,748 INFO [train.py:892] (3/4) Epoch 6, batch 1250, loss[loss=0.2568, simple_loss=0.2961, pruned_loss=0.1088, over 19777.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3253, pruned_loss=0.119, over 3942521.95 frames. ], batch size: 108, lr: 2.48e-02, grad_scale: 8.0
+2023-03-27 20:53:08,389 INFO [train.py:892] (3/4) Epoch 6, batch 1300, loss[loss=0.3238, simple_loss=0.3588, pruned_loss=0.1444, over 19744.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.324, pruned_loss=0.1178, over 3945005.75 frames. ], batch size: 253, lr: 2.48e-02, grad_scale: 8.0
+2023-03-27 20:54:11,550 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.477e+02 5.973e+02 6.933e+02 8.304e+02 2.000e+03, threshold=1.387e+03, percent-clipped=2.0
+2023-03-27 20:54:16,432 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6381, 2.6414, 3.9316, 2.6782, 3.3818, 4.0371, 2.2553, 2.2546],
+       device='cuda:3'), covar=tensor([0.0616, 0.3077, 0.0328, 0.0731, 0.1158, 0.0373, 0.1262, 0.1827],
+       device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0290, 0.0187, 0.0168, 0.0272, 0.0166, 0.0202, 0.0206],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 20:54:41,194 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.35 vs. limit=5.0
+2023-03-27 20:54:51,948 INFO [train.py:892] (3/4) Epoch 6, batch 1350, loss[loss=0.2617, simple_loss=0.3072, pruned_loss=0.1081, over 19810.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3228, pruned_loss=0.1174, over 3946002.70 frames. ], batch size: 74, lr: 2.47e-02, grad_scale: 8.0
+2023-03-27 20:56:39,330 INFO [train.py:892] (3/4) Epoch 6, batch 1400, loss[loss=0.2776, simple_loss=0.3344, pruned_loss=0.1104, over 19535.00 frames. ], tot_loss[loss=0.2803, simple_loss=0.3245, pruned_loss=0.1181, over 3945029.35 frames. ], batch size: 54, lr: 2.47e-02, grad_scale: 8.0
+2023-03-27 20:56:42,341 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4152, 4.8578, 5.2293, 4.8543, 4.2715, 4.9214, 5.0020, 5.4061],
+       device='cuda:3'), covar=tensor([0.1290, 0.0285, 0.0365, 0.0331, 0.0542, 0.0404, 0.0248, 0.0300],
+       device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0153, 0.0147, 0.0150, 0.0149, 0.0143, 0.0131, 0.0132],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 20:57:13,700 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3704, 3.3529, 4.8135, 3.6656, 4.2025, 4.7660, 2.6232, 2.7236],
+       device='cuda:3'), covar=tensor([0.0499, 0.2626, 0.0268, 0.0534, 0.0934, 0.0267, 0.1187, 0.1612],
+       device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0297, 0.0190, 0.0170, 0.0275, 0.0168, 0.0206, 0.0210],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 20:57:21,081 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10695.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:57:45,909 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9244, 2.3942, 2.9776, 2.6685, 2.6009, 3.2790, 1.7982, 2.1196],
+       device='cuda:3'), covar=tensor([0.0507, 0.1797, 0.0405, 0.0428, 0.0998, 0.0319, 0.1114, 0.1387],
+       device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0295, 0.0188, 0.0169, 0.0274, 0.0167, 0.0205, 0.0209],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 20:57:46,680 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.801e+02 6.022e+02 7.479e+02 9.663e+02 1.752e+03, threshold=1.496e+03, percent-clipped=3.0
+2023-03-27 20:57:57,849 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.93 vs. limit=2.0
+2023-03-27 20:58:11,914 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10718.0, num_to_drop=1, layers_to_drop={2}
+2023-03-27 20:58:26,791 INFO [train.py:892] (3/4) Epoch 6, batch 1450, loss[loss=0.254, simple_loss=0.2994, pruned_loss=0.1043, over 19782.00 frames. ], tot_loss[loss=0.2797, simple_loss=0.3239, pruned_loss=0.1177, over 3946107.63 frames. ], batch size: 198, lr: 2.46e-02, grad_scale: 8.0
+2023-03-27 20:59:03,305 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:59:06,619 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:59:45,265 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.19 vs. limit=5.0
+2023-03-27 20:59:50,199 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 20:59:52,114 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10766.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 21:00:09,430 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:00:12,517 INFO [train.py:892] (3/4) Epoch 6, batch 1500, loss[loss=0.2492, simple_loss=0.3098, pruned_loss=0.09429, over 19802.00 frames. ], tot_loss[loss=0.2802, simple_loss=0.3243, pruned_loss=0.1181, over 3947621.27 frames. ], batch size: 51, lr: 2.46e-02, grad_scale: 8.0
+2023-03-27 21:00:38,467 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0
+2023-03-27 21:01:14,018 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:01:16,849 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.160e+02 5.698e+02 7.148e+02 8.817e+02 1.909e+03, threshold=1.430e+03, percent-clipped=1.0
+2023-03-27 21:01:25,562 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10810.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 21:01:29,600 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0
+2023-03-27 21:01:57,112 INFO [train.py:892] (3/4) Epoch 6, batch 1550, loss[loss=0.2725, simple_loss=0.3213, pruned_loss=0.1119, over 19662.00 frames. ], tot_loss[loss=0.2809, simple_loss=0.3252, pruned_loss=0.1183, over 3947194.42 frames. ], batch size: 43, lr: 2.45e-02, grad_scale: 8.0
+2023-03-27 21:02:16,581 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:02:22,125 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0
+2023-03-27 21:03:41,428 INFO [train.py:892] (3/4) Epoch 6, batch 1600, loss[loss=0.2705, simple_loss=0.3168, pruned_loss=0.1121, over 19779.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3253, pruned_loss=0.1187, over 3947389.09 frames. ], batch size: 69, lr: 2.45e-02, grad_scale: 8.0
+2023-03-27 21:04:01,491 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10885.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 21:04:44,771 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.321e+02 5.550e+02 6.735e+02 8.191e+02 1.432e+03, threshold=1.347e+03, percent-clipped=1.0
+2023-03-27 21:05:25,954 INFO [train.py:892] (3/4) Epoch 6, batch 1650, loss[loss=0.2461, simple_loss=0.2955, pruned_loss=0.09834, over 19785.00 frames. ], tot_loss[loss=0.2783, simple_loss=0.3226, pruned_loss=0.117, over 3948924.61 frames. ], batch size: 182, lr: 2.44e-02, grad_scale: 8.0
+2023-03-27 21:06:09,500 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10946.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 21:07:03,929 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.0994, 2.3684, 1.6607, 1.5058, 1.8046, 2.4860, 2.2945, 2.2602],
+       device='cuda:3'), covar=tensor([0.0229, 0.0357, 0.0247, 0.0613, 0.0505, 0.0268, 0.0177, 0.0225],
+       device='cuda:3'), in_proj_covar=tensor([0.0041, 0.0040, 0.0041, 0.0057, 0.0057, 0.0038, 0.0033, 0.0035],
+       device='cuda:3'), out_proj_covar=tensor([8.9539e-05, 8.8801e-05, 8.6612e-05, 1.2805e-04, 1.2515e-04, 8.7274e-05,
+        7.4580e-05, 7.7974e-05], device='cuda:3')
+2023-03-27 21:07:11,129 INFO [train.py:892] (3/4) Epoch 6, batch 1700, loss[loss=0.2774, simple_loss=0.3169, pruned_loss=0.119, over 19773.00 frames. ], tot_loss[loss=0.2788, simple_loss=0.3232, pruned_loss=0.1172, over 3950170.25 frames. ], batch size: 241, lr: 2.44e-02, grad_scale: 8.0
+2023-03-27 21:08:17,554 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.223e+02 5.843e+02 7.049e+02 9.611e+02 2.273e+03, threshold=1.410e+03, percent-clipped=6.0
+2023-03-27 21:08:18,705 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.85 vs. limit=5.0
+2023-03-27 21:08:22,093 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0034, 3.0668, 3.9011, 3.4522, 3.6884, 3.8576, 3.8475, 3.8991],
+       device='cuda:3'), covar=tensor([0.0164, 0.0456, 0.0130, 0.0840, 0.0133, 0.0187, 0.0154, 0.0112],
+       device='cuda:3'), in_proj_covar=tensor([0.0062, 0.0071, 0.0058, 0.0130, 0.0052, 0.0062, 0.0060, 0.0052],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0004, 0.0001, 0.0002, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-27 21:08:36,789 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2572, 2.1390, 2.5176, 2.5562, 2.7376, 2.6159, 3.1600, 3.2647],
+       device='cuda:3'), covar=tensor([0.0516, 0.1473, 0.1169, 0.1399, 0.1227, 0.1105, 0.0348, 0.0405],
+       device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0182, 0.0186, 0.0201, 0.0206, 0.0188, 0.0125, 0.0127],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:08:51,964 INFO [train.py:892] (3/4) Epoch 6, batch 1750, loss[loss=0.2352, simple_loss=0.2897, pruned_loss=0.09032, over 19883.00 frames. ], tot_loss[loss=0.2796, simple_loss=0.3237, pruned_loss=0.1177, over 3949439.08 frames. ], batch size: 92, lr: 2.43e-02, grad_scale: 8.0
+2023-03-27 21:09:21,119 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.05 vs. limit=5.0
+2023-03-27 21:09:24,386 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.41 vs. limit=5.0
+2023-03-27 21:10:03,030 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:10:21,303 INFO [train.py:892] (3/4) Epoch 6, batch 1800, loss[loss=0.2834, simple_loss=0.329, pruned_loss=0.1189, over 19761.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3216, pruned_loss=0.1162, over 3950557.10 frames. ], batch size: 213, lr: 2.43e-02, grad_scale: 8.0
+2023-03-27 21:11:02,753 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11100.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:11:04,340 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4029, 5.6671, 5.9062, 5.7221, 5.5435, 5.1565, 5.4918, 5.5845],
+       device='cuda:3'), covar=tensor([0.1106, 0.0723, 0.0895, 0.0574, 0.0561, 0.0833, 0.1818, 0.1747],
+       device='cuda:3'), in_proj_covar=tensor([0.0220, 0.0192, 0.0262, 0.0203, 0.0193, 0.0194, 0.0247, 0.0295],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-27 21:11:13,680 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.857e+02 5.952e+02 7.096e+02 9.277e+02 1.982e+03, threshold=1.419e+03, percent-clipped=4.0
+2023-03-27 21:11:15,959 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0594, 3.1223, 1.2831, 3.9749, 3.3136, 3.8322, 3.9634, 3.0362],
+       device='cuda:3'), covar=tensor([0.0611, 0.0469, 0.1786, 0.0374, 0.0556, 0.0403, 0.0600, 0.0684],
+       device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0093, 0.0119, 0.0098, 0.0085, 0.0077, 0.0089, 0.0102],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:11:19,009 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11110.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 21:11:23,885 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:11:44,930 INFO [train.py:892] (3/4) Epoch 6, batch 1850, loss[loss=0.32, simple_loss=0.3752, pruned_loss=0.1324, over 19697.00 frames. ], tot_loss[loss=0.276, simple_loss=0.3224, pruned_loss=0.1148, over 3950031.99 frames. ], batch size: 56, lr: 2.42e-02, grad_scale: 8.0
+2023-03-27 21:12:41,536 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:12:42,781 INFO [train.py:892] (3/4) Epoch 7, batch 0, loss[loss=0.3503, simple_loss=0.3749, pruned_loss=0.1629, over 19689.00 frames. ], tot_loss[loss=0.3503, simple_loss=0.3749, pruned_loss=0.1629, over 19689.00 frames. ], batch size: 315, lr: 2.27e-02, grad_scale: 8.0
+2023-03-27 21:12:42,781 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-27 21:12:56,425 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6037, 4.3115, 4.1745, 4.7124, 4.4907, 4.7333, 4.6140, 4.8643],
+       device='cuda:3'), covar=tensor([0.0419, 0.0306, 0.0406, 0.0207, 0.0456, 0.0175, 0.0321, 0.0234],
+       device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0124, 0.0147, 0.0126, 0.0119, 0.0100, 0.0122, 0.0137],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:13:01,929 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8991, 2.7169, 4.5471, 3.2181, 3.9764, 4.5741, 2.2483, 2.1316],
+       device='cuda:3'), covar=tensor([0.0663, 0.3120, 0.0292, 0.0610, 0.1053, 0.0311, 0.1438, 0.2411],
+       device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0300, 0.0198, 0.0175, 0.0282, 0.0174, 0.0209, 0.0215],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:13:04,674 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5336, 3.0929, 3.1418, 3.4603, 3.6947, 3.9193, 4.4744, 4.8028],
+       device='cuda:3'), covar=tensor([0.0403, 0.1299, 0.1220, 0.1663, 0.1264, 0.1024, 0.0278, 0.0216],
+       device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0184, 0.0189, 0.0205, 0.0209, 0.0192, 0.0126, 0.0129],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:13:07,383 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5852, 2.6426, 3.7159, 2.9343, 3.3560, 3.7977, 3.5983, 3.7823],
+       device='cuda:3'), covar=tensor([0.0085, 0.0473, 0.0077, 0.1033, 0.0114, 0.0125, 0.0118, 0.0044],
+       device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0076, 0.0062, 0.0139, 0.0056, 0.0067, 0.0063, 0.0055],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0004, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:13:10,592 INFO [train.py:926] (3/4) Epoch 7, validation: loss=0.1961, simple_loss=0.2755, pruned_loss=0.05831, over 2883724.00 frames. 
+2023-03-27 21:13:10,593 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22042MB
+2023-03-27 21:13:59,131 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8831, 3.0125, 3.3328, 4.0323, 2.5438, 3.1449, 2.4796, 2.1898],
+       device='cuda:3'), covar=tensor([0.0277, 0.2695, 0.0811, 0.0142, 0.2218, 0.0424, 0.1037, 0.1997],
+       device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0324, 0.0197, 0.0110, 0.0226, 0.0145, 0.0175, 0.0203],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:14:09,090 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:14:42,575 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:14:52,020 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0
+2023-03-27 21:14:56,459 INFO [train.py:892] (3/4) Epoch 7, batch 50, loss[loss=0.2571, simple_loss=0.3016, pruned_loss=0.1063, over 19785.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3153, pruned_loss=0.1143, over 891083.57 frames. ], batch size: 172, lr: 2.27e-02, grad_scale: 8.0
+2023-03-27 21:15:51,181 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.005e+02 5.596e+02 6.713e+02 8.168e+02 2.476e+03, threshold=1.343e+03, percent-clipped=3.0
+2023-03-27 21:16:43,167 INFO [train.py:892] (3/4) Epoch 7, batch 100, loss[loss=0.2489, simple_loss=0.2982, pruned_loss=0.09981, over 19741.00 frames. ], tot_loss[loss=0.2667, simple_loss=0.3134, pruned_loss=0.11, over 1570241.51 frames. ], batch size: 95, lr: 2.26e-02, grad_scale: 8.0
+2023-03-27 21:16:52,188 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:17:05,925 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11241.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 21:17:40,426 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8231, 1.8960, 2.8966, 3.2160, 3.4129, 3.7026, 3.7237, 3.7393],
+       device='cuda:3'), covar=tensor([0.0629, 0.2135, 0.0627, 0.0374, 0.0299, 0.0170, 0.0213, 0.0270],
+       device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0160, 0.0110, 0.0102, 0.0084, 0.0083, 0.0079, 0.0080],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-27 21:18:30,179 INFO [train.py:892] (3/4) Epoch 7, batch 150, loss[loss=0.2354, simple_loss=0.2769, pruned_loss=0.09693, over 19877.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3121, pruned_loss=0.1088, over 2098924.33 frames. ], batch size: 136, lr: 2.26e-02, grad_scale: 8.0
+2023-03-27 21:19:26,066 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.645e+02 5.370e+02 6.417e+02 7.584e+02 2.097e+03, threshold=1.283e+03, percent-clipped=1.0
+2023-03-27 21:20:18,359 INFO [train.py:892] (3/4) Epoch 7, batch 200, loss[loss=0.2412, simple_loss=0.2849, pruned_loss=0.09877, over 19744.00 frames. ], tot_loss[loss=0.2671, simple_loss=0.3143, pruned_loss=0.1099, over 2507871.95 frames. ], batch size: 139, lr: 2.26e-02, grad_scale: 8.0
+2023-03-27 21:20:38,945 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-27 21:22:03,988 INFO [train.py:892] (3/4) Epoch 7, batch 250, loss[loss=0.2523, simple_loss=0.2962, pruned_loss=0.1042, over 19851.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3146, pruned_loss=0.1101, over 2827959.64 frames. ], batch size: 165, lr: 2.25e-02, grad_scale: 8.0
+2023-03-27 21:22:34,882 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2456, 4.1728, 2.6390, 4.8275, 5.1486, 2.2062, 4.1434, 3.9429],
+       device='cuda:3'), covar=tensor([0.0562, 0.0776, 0.2521, 0.0389, 0.0141, 0.3047, 0.0786, 0.0511],
+       device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0180, 0.0195, 0.0137, 0.0097, 0.0194, 0.0196, 0.0129],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-27 21:22:44,460 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:22:57,029 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.370e+02 5.501e+02 6.755e+02 8.647e+02 2.452e+03, threshold=1.351e+03, percent-clipped=4.0
+2023-03-27 21:23:45,386 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:23:46,580 INFO [train.py:892] (3/4) Epoch 7, batch 300, loss[loss=0.2203, simple_loss=0.2793, pruned_loss=0.08071, over 19809.00 frames. ], tot_loss[loss=0.27, simple_loss=0.3166, pruned_loss=0.1116, over 3076563.62 frames. ], batch size: 65, lr: 2.25e-02, grad_scale: 8.0
+2023-03-27 21:24:25,384 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:25:01,679 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6519, 3.7215, 4.1305, 5.0593, 2.7694, 3.5876, 2.8723, 2.4975],
+       device='cuda:3'), covar=tensor([0.0327, 0.2616, 0.0664, 0.0101, 0.2564, 0.0596, 0.1117, 0.2125],
+       device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0316, 0.0194, 0.0110, 0.0223, 0.0145, 0.0173, 0.0199],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:25:28,302 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:25:34,581 INFO [train.py:892] (3/4) Epoch 7, batch 350, loss[loss=0.2562, simple_loss=0.3117, pruned_loss=0.1003, over 19702.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3159, pruned_loss=0.1113, over 3269645.21 frames. ], batch size: 101, lr: 2.24e-02, grad_scale: 8.0
+2023-03-27 21:25:35,928 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-27 21:26:28,804 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4133, 5.5537, 5.9289, 5.7190, 5.5039, 5.2966, 5.5926, 5.4538],
+       device='cuda:3'), covar=tensor([0.1275, 0.0905, 0.0842, 0.0952, 0.0776, 0.0753, 0.2002, 0.2519],
+       device='cuda:3'), in_proj_covar=tensor([0.0215, 0.0192, 0.0262, 0.0208, 0.0194, 0.0189, 0.0249, 0.0289],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-27 21:26:29,987 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.848e+02 5.697e+02 6.987e+02 8.639e+02 1.327e+03, threshold=1.397e+03, percent-clipped=1.0
+2023-03-27 21:27:18,065 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:27:19,347 INFO [train.py:892] (3/4) Epoch 7, batch 400, loss[loss=0.2662, simple_loss=0.3236, pruned_loss=0.1044, over 19579.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3158, pruned_loss=0.111, over 3421351.74 frames. ], batch size: 53, lr: 2.24e-02, grad_scale: 8.0
+2023-03-27 21:27:22,521 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:27:40,993 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11541.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 21:27:58,096 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6170, 2.7141, 1.4010, 3.5065, 2.9966, 3.3702, 3.3537, 2.8401],
+       device='cuda:3'), covar=tensor([0.0684, 0.0591, 0.1618, 0.0348, 0.0548, 0.0341, 0.0648, 0.0609],
+       device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0091, 0.0115, 0.0096, 0.0083, 0.0077, 0.0090, 0.0101],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:29:06,950 INFO [train.py:892] (3/4) Epoch 7, batch 450, loss[loss=0.2779, simple_loss=0.3245, pruned_loss=0.1157, over 19766.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3166, pruned_loss=0.1113, over 3539108.54 frames. ], batch size: 163, lr: 2.23e-02, grad_scale: 8.0
+2023-03-27 21:29:23,362 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11589.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 21:29:31,250 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:29:31,847 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.78 vs. limit=2.0
+2023-03-27 21:30:00,683 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.550e+02 5.488e+02 6.753e+02 8.459e+02 1.411e+03, threshold=1.351e+03, percent-clipped=1.0
+2023-03-27 21:30:52,139 INFO [train.py:892] (3/4) Epoch 7, batch 500, loss[loss=0.2669, simple_loss=0.3061, pruned_loss=0.1139, over 19745.00 frames. ], tot_loss[loss=0.2692, simple_loss=0.3161, pruned_loss=0.1112, over 3629540.36 frames. ], batch size: 221, lr: 2.23e-02, grad_scale: 8.0
+2023-03-27 21:30:59,307 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6487, 2.6709, 3.7603, 3.2485, 3.5803, 3.5955, 3.4439, 3.4957],
+       device='cuda:3'), covar=tensor([0.0163, 0.0537, 0.0091, 0.0814, 0.0102, 0.0188, 0.0120, 0.0122],
+       device='cuda:3'), in_proj_covar=tensor([0.0063, 0.0073, 0.0060, 0.0130, 0.0053, 0.0063, 0.0059, 0.0053],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0004, 0.0001, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:31:01,051 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:32:34,253 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7686, 5.2442, 5.2092, 5.2480, 4.9087, 5.1342, 4.6158, 4.7818],
+       device='cuda:3'), covar=tensor([0.0357, 0.0305, 0.0550, 0.0375, 0.0508, 0.0545, 0.0614, 0.0923],
+       device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0149, 0.0199, 0.0158, 0.0153, 0.0145, 0.0174, 0.0210],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-27 21:32:36,847 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-27 21:32:37,354 INFO [train.py:892] (3/4) Epoch 7, batch 550, loss[loss=0.2531, simple_loss=0.3101, pruned_loss=0.09809, over 19736.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3147, pruned_loss=0.1101, over 3701001.97 frames. ], batch size: 80, lr: 2.23e-02, grad_scale: 8.0
+2023-03-27 21:33:09,717 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:33:15,280 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:33:19,007 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2088, 4.3864, 4.7201, 4.4558, 4.1399, 4.6194, 4.3316, 4.9252],
+       device='cuda:3'), covar=tensor([0.1203, 0.0330, 0.0325, 0.0293, 0.0660, 0.0344, 0.0355, 0.0229],
+       device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0170, 0.0158, 0.0161, 0.0164, 0.0155, 0.0148, 0.0143],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:33:31,006 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.036e+02 6.098e+02 7.200e+02 8.839e+02 1.494e+03, threshold=1.440e+03, percent-clipped=2.0
+2023-03-27 21:33:36,311 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.65 vs. limit=5.0
+2023-03-27 21:34:22,373 INFO [train.py:892] (3/4) Epoch 7, batch 600, loss[loss=0.2415, simple_loss=0.2883, pruned_loss=0.09737, over 19807.00 frames. ], tot_loss[loss=0.269, simple_loss=0.3158, pruned_loss=0.111, over 3756066.64 frames. ], batch size: 114, lr: 2.22e-02, grad_scale: 8.0
+2023-03-27 21:35:24,286 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:35:28,692 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-27 21:36:06,682 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6726, 3.8402, 2.3101, 4.1233, 4.1886, 1.8234, 3.4005, 3.3565],
+       device='cuda:3'), covar=tensor([0.0569, 0.0700, 0.2400, 0.0410, 0.0187, 0.3042, 0.0932, 0.0488],
+       device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0183, 0.0197, 0.0143, 0.0101, 0.0195, 0.0204, 0.0133],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:36:07,569 INFO [train.py:892] (3/4) Epoch 7, batch 650, loss[loss=0.2539, simple_loss=0.2947, pruned_loss=0.1066, over 19743.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3147, pruned_loss=0.1105, over 3800526.58 frames. ], batch size: 134, lr: 2.22e-02, grad_scale: 8.0
+2023-03-27 21:37:02,909 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.631e+02 5.681e+02 6.679e+02 8.355e+02 2.061e+03, threshold=1.336e+03, percent-clipped=4.0
+2023-03-27 21:37:23,927 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:37:52,507 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:37:53,592 INFO [train.py:892] (3/4) Epoch 7, batch 700, loss[loss=0.218, simple_loss=0.2784, pruned_loss=0.07887, over 19763.00 frames. ], tot_loss[loss=0.2669, simple_loss=0.3141, pruned_loss=0.1099, over 3834168.49 frames. ], batch size: 102, lr: 2.21e-02, grad_scale: 8.0
+2023-03-27 21:37:56,115 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6887, 4.1983, 4.2021, 4.7135, 4.3223, 4.7456, 4.7675, 4.9310],
+       device='cuda:3'), covar=tensor([0.0571, 0.0336, 0.0437, 0.0273, 0.0570, 0.0233, 0.0328, 0.0281],
+       device='cuda:3'), in_proj_covar=tensor([0.0111, 0.0121, 0.0141, 0.0121, 0.0117, 0.0099, 0.0117, 0.0135],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:38:25,401 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:39:08,894 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1064, 2.6915, 4.1359, 3.4803, 3.7389, 3.8799, 3.8762, 3.5963],
+       device='cuda:3'), covar=tensor([0.0109, 0.0597, 0.0088, 0.0820, 0.0119, 0.0197, 0.0132, 0.0134],
+       device='cuda:3'), in_proj_covar=tensor([0.0064, 0.0074, 0.0059, 0.0131, 0.0052, 0.0063, 0.0060, 0.0053],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0004, 0.0001, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:39:32,291 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:39:34,207 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:39:39,317 INFO [train.py:892] (3/4) Epoch 7, batch 750, loss[loss=0.2459, simple_loss=0.302, pruned_loss=0.0949, over 19756.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3136, pruned_loss=0.1094, over 3860245.16 frames. ], batch size: 110, lr: 2.21e-02, grad_scale: 8.0
+2023-03-27 21:39:53,457 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:40:34,152 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.808e+02 5.639e+02 6.770e+02 8.403e+02 1.981e+03, threshold=1.354e+03, percent-clipped=4.0
+2023-03-27 21:40:35,210 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:41:00,404 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1067, 4.6802, 4.8289, 4.6180, 4.1441, 4.6265, 4.5986, 5.0928],
+       device='cuda:3'), covar=tensor([0.1468, 0.0350, 0.0477, 0.0339, 0.0676, 0.0498, 0.0371, 0.0319],
+       device='cuda:3'), in_proj_covar=tensor([0.0221, 0.0165, 0.0155, 0.0159, 0.0162, 0.0154, 0.0145, 0.0141],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:41:02,293 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:41:22,936 INFO [train.py:892] (3/4) Epoch 7, batch 800, loss[loss=0.2524, simple_loss=0.3044, pruned_loss=0.1001, over 19667.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3129, pruned_loss=0.1086, over 3880395.76 frames. ], batch size: 43, lr: 2.21e-02, grad_scale: 8.0
+2023-03-27 21:41:59,571 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:43:08,356 INFO [train.py:892] (3/4) Epoch 7, batch 850, loss[loss=0.2789, simple_loss=0.3153, pruned_loss=0.1213, over 19780.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3164, pruned_loss=0.111, over 3895075.42 frames. ], batch size: 163, lr: 2.20e-02, grad_scale: 8.0
+2023-03-27 21:43:09,359 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:43:14,955 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:43:29,709 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:43:30,001 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6761, 3.8179, 2.2871, 4.2373, 4.2862, 1.7294, 3.5335, 3.2014],
+       device='cuda:3'), covar=tensor([0.0653, 0.0790, 0.2706, 0.0496, 0.0199, 0.3357, 0.0973, 0.0672],
+       device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0185, 0.0201, 0.0145, 0.0101, 0.0196, 0.0204, 0.0135],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:43:33,783 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0057, 4.2233, 4.4813, 4.2338, 3.9605, 4.3181, 4.1889, 4.6568],
+       device='cuda:3'), covar=tensor([0.1128, 0.0289, 0.0307, 0.0260, 0.0626, 0.0331, 0.0337, 0.0208],
+       device='cuda:3'), in_proj_covar=tensor([0.0217, 0.0161, 0.0153, 0.0157, 0.0158, 0.0150, 0.0143, 0.0138],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:44:04,438 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.79 vs. limit=5.0
+2023-03-27 21:44:07,073 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.301e+02 6.093e+02 7.463e+02 9.130e+02 1.696e+03, threshold=1.493e+03, percent-clipped=3.0
+2023-03-27 21:44:12,012 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12009.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:44:56,860 INFO [train.py:892] (3/4) Epoch 7, batch 900, loss[loss=0.2387, simple_loss=0.2887, pruned_loss=0.09429, over 19736.00 frames. ], tot_loss[loss=0.2681, simple_loss=0.3151, pruned_loss=0.1106, over 3908147.98 frames. ], batch size: 106, lr: 2.20e-02, grad_scale: 8.0
+2023-03-27 21:45:08,071 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0
+2023-03-27 21:45:27,378 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:45:48,828 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:46:11,090 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3358, 3.4078, 3.7050, 4.7897, 2.7579, 3.3178, 2.8792, 2.4742],
+       device='cuda:3'), covar=tensor([0.0371, 0.2871, 0.0844, 0.0120, 0.2267, 0.0630, 0.1049, 0.1924],
+       device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0322, 0.0202, 0.0117, 0.0227, 0.0147, 0.0176, 0.0201],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:46:32,778 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:46:34,626 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6357, 5.0302, 4.9230, 4.9339, 4.6085, 4.8626, 4.3426, 4.4930],
+       device='cuda:3'), covar=tensor([0.0376, 0.0315, 0.0586, 0.0402, 0.0615, 0.0600, 0.0610, 0.0817],
+       device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0152, 0.0203, 0.0164, 0.0158, 0.0148, 0.0179, 0.0216],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-27 21:46:44,680 INFO [train.py:892] (3/4) Epoch 7, batch 950, loss[loss=0.2412, simple_loss=0.2984, pruned_loss=0.09197, over 19741.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3151, pruned_loss=0.1099, over 3917639.86 frames. ], batch size: 44, lr: 2.19e-02, grad_scale: 8.0
+2023-03-27 21:47:10,875 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4264, 2.7981, 2.9803, 3.6417, 2.1360, 2.8161, 2.3150, 2.0541],
+       device='cuda:3'), covar=tensor([0.0437, 0.2276, 0.0837, 0.0150, 0.2328, 0.0587, 0.1146, 0.1840],
+       device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0321, 0.0203, 0.0116, 0.0226, 0.0147, 0.0176, 0.0200],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:47:39,038 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.928e+02 5.727e+02 6.871e+02 8.158e+02 1.418e+03, threshold=1.374e+03, percent-clipped=0.0
+2023-03-27 21:48:29,673 INFO [train.py:892] (3/4) Epoch 7, batch 1000, loss[loss=0.2817, simple_loss=0.329, pruned_loss=0.1172, over 19643.00 frames. ], tot_loss[loss=0.2678, simple_loss=0.3157, pruned_loss=0.11, over 3921805.05 frames. ], batch size: 72, lr: 2.19e-02, grad_scale: 8.0
+2023-03-27 21:48:43,746 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8433, 4.0257, 4.3390, 4.0652, 3.9310, 4.3030, 4.0089, 4.4891],
+       device='cuda:3'), covar=tensor([0.1179, 0.0296, 0.0312, 0.0305, 0.0687, 0.0328, 0.0326, 0.0235],
+       device='cuda:3'), in_proj_covar=tensor([0.0217, 0.0162, 0.0154, 0.0158, 0.0156, 0.0150, 0.0142, 0.0138],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:48:43,807 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12137.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:48:46,315 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.85 vs. limit=5.0
+2023-03-27 21:49:32,318 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4155, 3.3424, 2.1036, 3.5860, 3.7253, 1.6188, 3.0330, 2.9713],
+       device='cuda:3'), covar=tensor([0.0535, 0.0774, 0.2513, 0.0435, 0.0219, 0.3087, 0.1052, 0.0580],
+       device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0183, 0.0196, 0.0144, 0.0100, 0.0195, 0.0200, 0.0133],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:49:48,345 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7765, 2.6814, 3.7172, 3.2643, 3.4816, 3.6550, 3.6142, 3.5096],
+       device='cuda:3'), covar=tensor([0.0124, 0.0510, 0.0098, 0.0812, 0.0101, 0.0202, 0.0127, 0.0128],
+       device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0075, 0.0061, 0.0134, 0.0053, 0.0065, 0.0060, 0.0053],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0004, 0.0001, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:49:54,023 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0871, 3.3997, 3.7187, 3.4224, 3.4091, 3.7301, 3.4156, 3.7966],
+       device='cuda:3'), covar=tensor([0.1518, 0.0418, 0.0468, 0.0416, 0.0909, 0.0452, 0.0418, 0.0394],
+       device='cuda:3'), in_proj_covar=tensor([0.0217, 0.0163, 0.0154, 0.0159, 0.0158, 0.0152, 0.0142, 0.0139],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:49:55,673 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12172.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:50:13,703 INFO [train.py:892] (3/4) Epoch 7, batch 1050, loss[loss=0.3008, simple_loss=0.3387, pruned_loss=0.1314, over 19686.00 frames. ], tot_loss[loss=0.2704, simple_loss=0.3175, pruned_loss=0.1117, over 3926927.84 frames. ], batch size: 265, lr: 2.19e-02, grad_scale: 8.0
+2023-03-27 21:50:24,040 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.6745, 2.0733, 1.5520, 1.2128, 1.7203, 2.0941, 1.9489, 2.0551],
+       device='cuda:3'), covar=tensor([0.0235, 0.0228, 0.0202, 0.0530, 0.0433, 0.0164, 0.0140, 0.0121],
+       device='cuda:3'), in_proj_covar=tensor([0.0044, 0.0043, 0.0043, 0.0060, 0.0060, 0.0040, 0.0034, 0.0037],
+       device='cuda:3'), out_proj_covar=tensor([9.9315e-05, 9.7003e-05, 9.5455e-05, 1.3863e-04, 1.3565e-04, 9.2777e-05,
+        7.9053e-05, 8.4268e-05], device='cuda:3')
+2023-03-27 21:50:31,373 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12188.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:50:59,954 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:51:09,330 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.917e+02 5.376e+02 7.198e+02 8.445e+02 1.495e+03, threshold=1.440e+03, percent-clipped=1.0
+2023-03-27 21:52:00,175 INFO [train.py:892] (3/4) Epoch 7, batch 1100, loss[loss=0.2443, simple_loss=0.298, pruned_loss=0.09526, over 19693.00 frames. ], tot_loss[loss=0.2703, simple_loss=0.3173, pruned_loss=0.1117, over 3932069.22 frames. ], batch size: 46, lr: 2.18e-02, grad_scale: 8.0
+2023-03-27 21:52:12,293 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:53:35,987 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:53:45,135 INFO [train.py:892] (3/4) Epoch 7, batch 1150, loss[loss=0.2585, simple_loss=0.3121, pruned_loss=0.1025, over 19890.00 frames. ], tot_loss[loss=0.2687, simple_loss=0.3162, pruned_loss=0.1106, over 3936564.43 frames. ], batch size: 47, lr: 2.18e-02, grad_scale: 8.0
+2023-03-27 21:54:06,742 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:54:14,139 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7198, 3.7695, 3.9248, 5.0977, 2.8139, 3.7328, 2.9828, 2.8658],
+       device='cuda:3'), covar=tensor([0.0308, 0.2520, 0.0782, 0.0107, 0.2307, 0.0530, 0.1116, 0.1831],
+       device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0321, 0.0204, 0.0116, 0.0225, 0.0146, 0.0176, 0.0202],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 21:54:34,571 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:54:41,305 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.471e+02 5.763e+02 6.995e+02 9.054e+02 1.717e+03, threshold=1.399e+03, percent-clipped=2.0
+2023-03-27 21:55:31,483 INFO [train.py:892] (3/4) Epoch 7, batch 1200, loss[loss=0.2614, simple_loss=0.3073, pruned_loss=0.1077, over 19800.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3163, pruned_loss=0.1102, over 3939814.25 frames. ], batch size: 126, lr: 2.18e-02, grad_scale: 8.0
+2023-03-27 21:55:48,753 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12339.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:55:50,626 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12340.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:56:21,417 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:57:17,150 INFO [train.py:892] (3/4) Epoch 7, batch 1250, loss[loss=0.2883, simple_loss=0.3322, pruned_loss=0.1222, over 19810.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3157, pruned_loss=0.1095, over 3942547.59 frames. ], batch size: 67, lr: 2.17e-02, grad_scale: 8.0
+2023-03-27 21:57:49,695 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0
+2023-03-27 21:58:02,725 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:58:09,907 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.119e+02 6.346e+02 7.494e+02 8.891e+02 1.193e+03, threshold=1.499e+03, percent-clipped=0.0
+2023-03-27 21:59:01,224 INFO [train.py:892] (3/4) Epoch 7, batch 1300, loss[loss=0.258, simple_loss=0.2977, pruned_loss=0.1092, over 19795.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3158, pruned_loss=0.1098, over 3945001.23 frames. ], batch size: 162, lr: 2.17e-02, grad_scale: 8.0
+2023-03-27 21:59:03,852 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:59:33,306 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 21:59:53,456 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0
+2023-03-27 22:00:17,325 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6441, 2.0429, 2.6315, 3.2883, 3.5476, 3.6972, 3.7148, 3.8531],
+       device='cuda:3'), covar=tensor([0.0880, 0.2069, 0.0936, 0.0429, 0.0308, 0.0145, 0.0254, 0.0264],
+       device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0164, 0.0125, 0.0111, 0.0090, 0.0087, 0.0083, 0.0085],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002],
+       device='cuda:3')
+2023-03-27 22:00:27,468 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:00:45,672 INFO [train.py:892] (3/4) Epoch 7, batch 1350, loss[loss=0.3037, simple_loss=0.3414, pruned_loss=0.133, over 19757.00 frames. ], tot_loss[loss=0.2677, simple_loss=0.3153, pruned_loss=0.11, over 3947137.31 frames. ], batch size: 256, lr: 2.16e-02, grad_scale: 16.0
+2023-03-27 22:01:28,955 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:01:39,456 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.642e+02 6.155e+02 7.029e+02 9.046e+02 1.832e+03, threshold=1.406e+03, percent-clipped=3.0
+2023-03-27 22:01:40,615 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:01:44,542 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4169, 3.7911, 4.0103, 4.4025, 4.0744, 4.3808, 4.4383, 4.6330],
+       device='cuda:3'), covar=tensor([0.0552, 0.0353, 0.0464, 0.0270, 0.0514, 0.0286, 0.0371, 0.0240],
+       device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0131, 0.0152, 0.0131, 0.0125, 0.0108, 0.0124, 0.0138],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 22:02:07,087 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:02:29,803 INFO [train.py:892] (3/4) Epoch 7, batch 1400, loss[loss=0.2423, simple_loss=0.2903, pruned_loss=0.0972, over 19827.00 frames. ], tot_loss[loss=0.2666, simple_loss=0.3142, pruned_loss=0.1095, over 3949872.75 frames. ], batch size: 166, lr: 2.16e-02, grad_scale: 16.0
+2023-03-27 22:03:08,822 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:04:04,115 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12576.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:04:14,054 INFO [train.py:892] (3/4) Epoch 7, batch 1450, loss[loss=0.2775, simple_loss=0.3413, pruned_loss=0.1069, over 19536.00 frames. ], tot_loss[loss=0.2645, simple_loss=0.3131, pruned_loss=0.1079, over 3950491.71 frames. ], batch size: 54, lr: 2.16e-02, grad_scale: 16.0
+2023-03-27 22:05:00,836 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9783, 2.6765, 3.0356, 3.1465, 3.2343, 3.2501, 3.8177, 3.8786],
+       device='cuda:3'), covar=tensor([0.0431, 0.1339, 0.1169, 0.1483, 0.1564, 0.1024, 0.0341, 0.0442],
+       device='cuda:3'), in_proj_covar=tensor([0.0173, 0.0194, 0.0200, 0.0213, 0.0226, 0.0202, 0.0139, 0.0140],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 22:05:02,583 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:05:07,594 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.016e+02 5.644e+02 6.580e+02 8.199e+02 1.502e+03, threshold=1.316e+03, percent-clipped=1.0
+2023-03-27 22:05:43,403 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:05:43,792 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2032, 3.1029, 4.2789, 3.3471, 3.7792, 4.4488, 2.3066, 2.5907],
+       device='cuda:3'), covar=tensor([0.0444, 0.2330, 0.0294, 0.0500, 0.0923, 0.0290, 0.1329, 0.1770],
+       device='cuda:3'), in_proj_covar=tensor([0.0258, 0.0304, 0.0209, 0.0180, 0.0284, 0.0189, 0.0224, 0.0221],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 22:05:57,453 INFO [train.py:892] (3/4) Epoch 7, batch 1500, loss[loss=0.2577, simple_loss=0.3009, pruned_loss=0.1072, over 19800.00 frames. ], tot_loss[loss=0.2649, simple_loss=0.3133, pruned_loss=0.1083, over 3950354.29 frames. ], batch size: 150, lr: 2.15e-02, grad_scale: 16.0
+2023-03-27 22:06:16,816 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:06:28,510 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.12 vs. limit=5.0
+2023-03-27 22:06:41,614 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12652.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:07:41,420 INFO [train.py:892] (3/4) Epoch 7, batch 1550, loss[loss=0.2809, simple_loss=0.3282, pruned_loss=0.1168, over 19655.00 frames. ], tot_loss[loss=0.265, simple_loss=0.3135, pruned_loss=0.1083, over 3950206.70 frames. ], batch size: 57, lr: 2.15e-02, grad_scale: 16.0
+2023-03-27 22:07:56,844 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12688.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:08:35,757 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.712e+02 5.812e+02 6.854e+02 8.280e+02 1.889e+03, threshold=1.371e+03, percent-clipped=7.0
+2023-03-27 22:08:37,067 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0
+2023-03-27 22:09:26,215 INFO [train.py:892] (3/4) Epoch 7, batch 1600, loss[loss=0.2512, simple_loss=0.2936, pruned_loss=0.1044, over 19880.00 frames. ], tot_loss[loss=0.2642, simple_loss=0.3128, pruned_loss=0.1078, over 3950070.72 frames. ], batch size: 134, lr: 2.15e-02, grad_scale: 16.0
+2023-03-27 22:09:28,842 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:10:32,953 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:11:06,774 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12780.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:11:08,043 INFO [train.py:892] (3/4) Epoch 7, batch 1650, loss[loss=0.2675, simple_loss=0.3255, pruned_loss=0.1047, over 19603.00 frames. ], tot_loss[loss=0.2633, simple_loss=0.3123, pruned_loss=0.1071, over 3948954.04 frames. ], batch size: 50, lr: 2.14e-02, grad_scale: 16.0
+2023-03-27 22:11:54,387 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:12:04,855 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.717e+02 5.657e+02 6.905e+02 8.764e+02 1.774e+03, threshold=1.381e+03, percent-clipped=3.0
+2023-03-27 22:12:43,837 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12824.0, num_to_drop=1, layers_to_drop={2}
+2023-03-27 22:12:56,383 INFO [train.py:892] (3/4) Epoch 7, batch 1700, loss[loss=0.2892, simple_loss=0.3433, pruned_loss=0.1175, over 19737.00 frames. ], tot_loss[loss=0.2648, simple_loss=0.3138, pruned_loss=0.1079, over 3947439.58 frames. ], batch size: 63, lr: 2.14e-02, grad_scale: 16.0
+2023-03-27 22:13:16,328 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6182, 2.9067, 1.6932, 1.9952, 2.2516, 2.6458, 2.8619, 2.7649],
+       device='cuda:3'), covar=tensor([0.0182, 0.0311, 0.0272, 0.0490, 0.0403, 0.0260, 0.0150, 0.0218],
+       device='cuda:3'), in_proj_covar=tensor([0.0046, 0.0044, 0.0047, 0.0062, 0.0062, 0.0042, 0.0035, 0.0040],
+       device='cuda:3'), out_proj_covar=tensor([1.0511e-04, 1.0056e-04, 1.0353e-04, 1.4293e-04, 1.4078e-04, 9.9003e-05,
+        8.3152e-05, 9.1768e-05], device='cuda:3')
+2023-03-27 22:14:36,902 INFO [train.py:892] (3/4) Epoch 7, batch 1750, loss[loss=0.2589, simple_loss=0.2957, pruned_loss=0.111, over 19827.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3128, pruned_loss=0.1072, over 3948110.27 frames. ], batch size: 184, lr: 2.14e-02, grad_scale: 16.0
+2023-03-27 22:15:23,459 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.309e+02 5.556e+02 6.737e+02 8.077e+02 1.296e+03, threshold=1.347e+03, percent-clipped=0.0
+2023-03-27 22:16:05,766 INFO [train.py:892] (3/4) Epoch 7, batch 1800, loss[loss=0.2506, simple_loss=0.2916, pruned_loss=0.1048, over 19872.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.3126, pruned_loss=0.1072, over 3947732.98 frames. ], batch size: 138, lr: 2.13e-02, grad_scale: 16.0
+2023-03-27 22:16:34,203 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3961, 3.1987, 4.3671, 3.7598, 4.1148, 4.1468, 4.1985, 4.1113],
+       device='cuda:3'), covar=tensor([0.0099, 0.0451, 0.0080, 0.0844, 0.0081, 0.0162, 0.0119, 0.0101],
+       device='cuda:3'), in_proj_covar=tensor([0.0067, 0.0077, 0.0062, 0.0136, 0.0055, 0.0067, 0.0063, 0.0054],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0004, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 22:16:58,395 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4335, 4.1193, 2.4892, 4.7577, 4.9537, 2.1308, 3.9037, 3.5419],
+       device='cuda:3'), covar=tensor([0.0509, 0.0808, 0.2879, 0.0416, 0.0216, 0.3225, 0.0955, 0.0728],
+       device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0190, 0.0203, 0.0152, 0.0107, 0.0195, 0.0207, 0.0138],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 22:17:29,666 INFO [train.py:892] (3/4) Epoch 7, batch 1850, loss[loss=0.2782, simple_loss=0.3364, pruned_loss=0.1099, over 19820.00 frames. ], tot_loss[loss=0.263, simple_loss=0.3132, pruned_loss=0.1064, over 3947924.68 frames. ], batch size: 57, lr: 2.13e-02, grad_scale: 16.0
+2023-03-27 22:18:27,976 INFO [train.py:892] (3/4) Epoch 8, batch 0, loss[loss=0.2984, simple_loss=0.3342, pruned_loss=0.1313, over 19816.00 frames. ], tot_loss[loss=0.2984, simple_loss=0.3342, pruned_loss=0.1313, over 19816.00 frames. ], batch size: 288, lr: 2.00e-02, grad_scale: 16.0
+2023-03-27 22:18:27,977 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-27 22:18:55,072 INFO [train.py:926] (3/4) Epoch 8, validation: loss=0.189, simple_loss=0.2688, pruned_loss=0.05453, over 2883724.00 frames. 
+2023-03-27 22:18:55,073 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22221MB
+2023-03-27 22:19:33,628 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.3881, 1.7554, 2.1983, 2.8706, 3.1520, 3.3054, 3.3147, 3.2869],
+       device='cuda:3'), covar=tensor([0.0813, 0.2142, 0.1055, 0.0437, 0.0258, 0.0166, 0.0199, 0.0317],
+       device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0166, 0.0130, 0.0116, 0.0091, 0.0092, 0.0085, 0.0088],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 22:19:40,814 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.963e+02 5.840e+02 6.949e+02 8.045e+02 1.580e+03, threshold=1.390e+03, percent-clipped=1.0
+2023-03-27 22:20:44,103 INFO [train.py:892] (3/4) Epoch 8, batch 50, loss[loss=0.2519, simple_loss=0.3188, pruned_loss=0.0925, over 19790.00 frames. ], tot_loss[loss=0.2587, simple_loss=0.3061, pruned_loss=0.1057, over 892001.09 frames. ], batch size: 65, lr: 2.00e-02, grad_scale: 16.0
+2023-03-27 22:22:29,881 INFO [train.py:892] (3/4) Epoch 8, batch 100, loss[loss=0.354, simple_loss=0.4162, pruned_loss=0.1459, over 18735.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.3068, pruned_loss=0.1044, over 1569951.12 frames. ], batch size: 564, lr: 2.00e-02, grad_scale: 16.0
+2023-03-27 22:23:02,869 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=13102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:23:13,335 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.875e+02 5.582e+02 6.747e+02 8.374e+02 1.388e+03, threshold=1.349e+03, percent-clipped=0.0
+2023-03-27 22:23:37,498 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0718, 3.1862, 3.3502, 4.1351, 2.7215, 3.1321, 2.6947, 2.3480],
+       device='cuda:3'), covar=tensor([0.0384, 0.2295, 0.0827, 0.0148, 0.2073, 0.0592, 0.1059, 0.1895],
+       device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0324, 0.0209, 0.0124, 0.0229, 0.0152, 0.0181, 0.0204],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 22:23:39,229 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=13119.0, num_to_drop=1, layers_to_drop={3}
+2023-03-27 22:24:13,100 INFO [train.py:892] (3/4) Epoch 8, batch 150, loss[loss=0.2428, simple_loss=0.2912, pruned_loss=0.09722, over 19888.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.309, pruned_loss=0.1058, over 2097063.66 frames. ], batch size: 176, lr: 1.99e-02, grad_scale: 16.0
+2023-03-27 22:24:43,914 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=13150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:25:09,486 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7638, 2.5880, 1.4059, 3.3508, 2.9721, 3.2541, 3.3424, 2.4897],
+       device='cuda:3'), covar=tensor([0.0505, 0.0556, 0.1811, 0.0347, 0.0384, 0.0317, 0.0360, 0.0703],
+       device='cuda:3'), in_proj_covar=tensor([0.0110, 0.0100, 0.0123, 0.0105, 0.0090, 0.0085, 0.0097, 0.0110],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 22:25:58,991 INFO [train.py:892] (3/4) Epoch 8, batch 200, loss[loss=0.2173, simple_loss=0.2685, pruned_loss=0.08311, over 19764.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3086, pruned_loss=0.1045, over 2506658.59 frames. ], batch size: 125, lr: 1.99e-02, grad_scale: 16.0
+2023-03-27 22:26:43,151 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 4.035e+02 5.713e+02 7.016e+02 8.259e+02 1.478e+03, threshold=1.403e+03, percent-clipped=2.0
+2023-03-27 22:26:58,353 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2598, 2.2740, 3.2619, 2.8056, 3.1869, 3.3563, 3.2266, 3.2340],
+       device='cuda:3'), covar=tensor([0.0158, 0.0651, 0.0094, 0.0636, 0.0105, 0.0193, 0.0154, 0.0123],
+       device='cuda:3'), in_proj_covar=tensor([0.0067, 0.0079, 0.0062, 0.0137, 0.0056, 0.0068, 0.0063, 0.0055],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0004, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 22:27:43,919 INFO [train.py:892] (3/4) Epoch 8, batch 250, loss[loss=0.2397, simple_loss=0.2969, pruned_loss=0.09119, over 19616.00 frames. ], tot_loss[loss=0.26, simple_loss=0.3096, pruned_loss=0.1052, over 2824605.98 frames. ], batch size: 65, lr: 1.99e-02, grad_scale: 16.0
+2023-03-27 22:28:59,841 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0
+2023-03-27 22:29:31,209 INFO [train.py:892] (3/4) Epoch 8, batch 300, loss[loss=0.2356, simple_loss=0.2939, pruned_loss=0.08863, over 19799.00 frames. ], tot_loss[loss=0.2579, simple_loss=0.308, pruned_loss=0.1039, over 3074544.11 frames. ], batch size: 83, lr: 1.98e-02, grad_scale: 16.0
+2023-03-27 22:29:45,965 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8547, 2.8032, 4.0944, 3.1867, 3.7099, 4.0936, 2.1772, 2.3020],
+       device='cuda:3'), covar=tensor([0.0557, 0.2437, 0.0309, 0.0504, 0.0870, 0.0405, 0.1300, 0.1948],
+       device='cuda:3'), in_proj_covar=tensor([0.0264, 0.0308, 0.0213, 0.0184, 0.0284, 0.0196, 0.0229, 0.0224],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 22:30:13,179 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.006e+02 5.556e+02 6.827e+02 8.454e+02 1.465e+03, threshold=1.365e+03, percent-clipped=2.0
+2023-03-27 22:31:14,179 INFO [train.py:892] (3/4) Epoch 8, batch 350, loss[loss=0.2737, simple_loss=0.3219, pruned_loss=0.1128, over 19750.00 frames. ], tot_loss[loss=0.2578, simple_loss=0.3081, pruned_loss=0.1037, over 3268807.78 frames. ], batch size: 250, lr: 1.98e-02, grad_scale: 16.0
+2023-03-27 22:32:21,895 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.08 vs. limit=5.0
+2023-03-27 22:32:58,889 INFO [train.py:892] (3/4) Epoch 8, batch 400, loss[loss=0.25, simple_loss=0.3135, pruned_loss=0.0932, over 19789.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3091, pruned_loss=0.1043, over 3417068.91 frames. ], batch size: 53, lr: 1.98e-02, grad_scale: 16.0
+2023-03-27 22:33:27,280 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.79 vs. limit=5.0
+2023-03-27 22:33:41,531 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.399e+02 5.749e+02 7.183e+02 8.844e+02 1.624e+03, threshold=1.437e+03, percent-clipped=4.0
+2023-03-27 22:34:07,780 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=13419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:34:44,925 INFO [train.py:892] (3/4) Epoch 8, batch 450, loss[loss=0.2474, simple_loss=0.2904, pruned_loss=0.1022, over 19766.00 frames. ], tot_loss[loss=0.2605, simple_loss=0.3105, pruned_loss=0.1052, over 3534942.34 frames. ], batch size: 130, lr: 1.97e-02, grad_scale: 16.0
+2023-03-27 22:34:46,490 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0
+2023-03-27 22:35:12,882 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-27 22:35:49,876 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=13467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:36:27,723 INFO [train.py:892] (3/4) Epoch 8, batch 500, loss[loss=0.2267, simple_loss=0.2769, pruned_loss=0.08824, over 19830.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3081, pruned_loss=0.1039, over 3626760.06 frames. ], batch size: 146, lr: 1.97e-02, grad_scale: 16.0
+2023-03-27 22:36:46,710 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6559, 2.4854, 2.5480, 2.4727, 2.0911, 2.1257, 2.2427, 2.8210],
+       device='cuda:3'), covar=tensor([0.0220, 0.0297, 0.0335, 0.0308, 0.0379, 0.0341, 0.0369, 0.0354],
+       device='cuda:3'), in_proj_covar=tensor([0.0037, 0.0038, 0.0040, 0.0034, 0.0042, 0.0040, 0.0054, 0.0037],
+       device='cuda:3'), out_proj_covar=tensor([8.1830e-05, 8.3639e-05, 8.8358e-05, 7.6569e-05, 9.2664e-05, 8.9906e-05,
+        1.1587e-04, 8.2671e-05], device='cuda:3')
+2023-03-27 22:37:11,215 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.922e+02 5.157e+02 6.466e+02 8.211e+02 1.475e+03, threshold=1.293e+03, percent-clipped=2.0
+2023-03-27 22:38:04,046 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1422, 3.8682, 3.9705, 3.7542, 4.0887, 3.0505, 3.4127, 2.2811],
+       device='cuda:3'), covar=tensor([0.0159, 0.0168, 0.0131, 0.0156, 0.0118, 0.0676, 0.0701, 0.1087],
+       device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0096, 0.0084, 0.0096, 0.0086, 0.0107, 0.0120, 0.0103],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-27 22:38:12,608 INFO [train.py:892] (3/4) Epoch 8, batch 550, loss[loss=0.2934, simple_loss=0.3343, pruned_loss=0.1263, over 19749.00 frames. ], tot_loss[loss=0.2573, simple_loss=0.3075, pruned_loss=0.1036, over 3698160.89 frames. ], batch size: 273, lr: 1.97e-02, grad_scale: 16.0
+2023-03-27 22:39:58,921 INFO [train.py:892] (3/4) Epoch 8, batch 600, loss[loss=0.2987, simple_loss=0.3307, pruned_loss=0.1334, over 19780.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3099, pruned_loss=0.1051, over 3751883.90 frames. ], batch size: 211, lr: 1.97e-02, grad_scale: 16.0
+2023-03-27 22:40:41,839 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.580e+02 5.730e+02 6.817e+02 8.110e+02 1.556e+03, threshold=1.363e+03, percent-clipped=3.0
+2023-03-27 22:41:42,658 INFO [train.py:892] (3/4) Epoch 8, batch 650, loss[loss=0.2259, simple_loss=0.2805, pruned_loss=0.0857, over 19735.00 frames. ], tot_loss[loss=0.2599, simple_loss=0.31, pruned_loss=0.1049, over 3795810.21 frames. ], batch size: 77, lr: 1.96e-02, grad_scale: 16.0
+2023-03-27 22:43:26,145 INFO [train.py:892] (3/4) Epoch 8, batch 700, loss[loss=0.2387, simple_loss=0.2937, pruned_loss=0.09185, over 19573.00 frames. ], tot_loss[loss=0.2601, simple_loss=0.3104, pruned_loss=0.1049, over 3828867.84 frames. ], batch size: 60, lr: 1.96e-02, grad_scale: 16.0
+2023-03-27 22:44:11,780 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.892e+02 5.950e+02 6.978e+02 8.185e+02 2.283e+03, threshold=1.396e+03, percent-clipped=2.0
+2023-03-27 22:44:50,175 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.2058, 2.5912, 2.6915, 2.5086, 1.7213, 2.1570, 2.2059, 2.4794],
+       device='cuda:3'), covar=tensor([0.0290, 0.0227, 0.0329, 0.0234, 0.0360, 0.0453, 0.0452, 0.0491],
+       device='cuda:3'), in_proj_covar=tensor([0.0039, 0.0038, 0.0041, 0.0033, 0.0043, 0.0041, 0.0053, 0.0037],
+       device='cuda:3'), out_proj_covar=tensor([8.5847e-05, 8.4457e-05, 8.9474e-05, 7.5071e-05, 9.4255e-05, 9.1906e-05,
+        1.1561e-04, 8.4853e-05], device='cuda:3')
+2023-03-27 22:45:13,750 INFO [train.py:892] (3/4) Epoch 8, batch 750, loss[loss=0.2394, simple_loss=0.2971, pruned_loss=0.09083, over 19713.00 frames. ], tot_loss[loss=0.258, simple_loss=0.3086, pruned_loss=0.1037, over 3856861.93 frames. ], batch size: 81, lr: 1.96e-02, grad_scale: 16.0
+2023-03-27 22:45:16,671 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=13737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:45:41,782 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0
+2023-03-27 22:46:57,384 INFO [train.py:892] (3/4) Epoch 8, batch 800, loss[loss=0.2336, simple_loss=0.3002, pruned_loss=0.08352, over 19765.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3093, pruned_loss=0.1035, over 3877722.46 frames. ], batch size: 49, lr: 1.95e-02, grad_scale: 16.0
+2023-03-27 22:47:22,095 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=13798.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 22:47:40,988 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.356e+02 5.653e+02 6.887e+02 8.342e+02 1.511e+03, threshold=1.377e+03, percent-clipped=1.0
+2023-03-27 22:48:41,446 INFO [train.py:892] (3/4) Epoch 8, batch 850, loss[loss=0.3908, simple_loss=0.4499, pruned_loss=0.1658, over 18058.00 frames. ], tot_loss[loss=0.2576, simple_loss=0.3088, pruned_loss=0.1032, over 3892141.18 frames. ], batch size: 633, lr: 1.95e-02, grad_scale: 16.0
+2023-03-27 22:48:50,142 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5944, 4.8099, 5.1774, 4.9000, 4.2658, 5.0247, 4.9092, 5.4317],
+       device='cuda:3'), covar=tensor([0.1108, 0.0286, 0.0334, 0.0249, 0.0513, 0.0273, 0.0282, 0.0237],
+       device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0175, 0.0166, 0.0169, 0.0166, 0.0163, 0.0159, 0.0150],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 22:50:02,570 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7659, 2.9824, 3.3150, 3.7054, 2.3401, 3.1566, 2.3499, 2.2502],
+       device='cuda:3'), covar=tensor([0.0367, 0.2227, 0.0814, 0.0220, 0.2373, 0.0518, 0.1201, 0.1898],
+       device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0326, 0.0213, 0.0125, 0.0228, 0.0153, 0.0184, 0.0202],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 22:50:25,540 INFO [train.py:892] (3/4) Epoch 8, batch 900, loss[loss=0.2658, simple_loss=0.3106, pruned_loss=0.1105, over 19780.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.3091, pruned_loss=0.1041, over 3904717.83 frames. ], batch size: 247, lr: 1.95e-02, grad_scale: 16.0
+2023-03-27 22:51:07,623 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.950e+02 5.711e+02 6.914e+02 8.522e+02 2.139e+03, threshold=1.383e+03, percent-clipped=3.0
+2023-03-27 22:52:07,734 INFO [train.py:892] (3/4) Epoch 8, batch 950, loss[loss=0.2185, simple_loss=0.278, pruned_loss=0.07954, over 19687.00 frames. ], tot_loss[loss=0.2582, simple_loss=0.3088, pruned_loss=0.1038, over 3915451.25 frames. ], batch size: 75, lr: 1.94e-02, grad_scale: 16.0
+2023-03-27 22:52:29,793 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8050, 2.7204, 4.0790, 2.9990, 3.6703, 4.0208, 2.2490, 2.2983],
+       device='cuda:3'), covar=tensor([0.0545, 0.2516, 0.0307, 0.0581, 0.0923, 0.0422, 0.1268, 0.1858],
+       device='cuda:3'), in_proj_covar=tensor([0.0263, 0.0307, 0.0217, 0.0187, 0.0285, 0.0202, 0.0231, 0.0226],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 22:53:53,096 INFO [train.py:892] (3/4) Epoch 8, batch 1000, loss[loss=0.2351, simple_loss=0.2828, pruned_loss=0.0937, over 19841.00 frames. ], tot_loss[loss=0.2588, simple_loss=0.3092, pruned_loss=0.1042, over 3923149.95 frames. ], batch size: 124, lr: 1.94e-02, grad_scale: 16.0
+2023-03-27 22:54:42,904 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.512e+02 5.428e+02 6.416e+02 7.650e+02 1.405e+03, threshold=1.283e+03, percent-clipped=1.0
+2023-03-27 22:55:44,625 INFO [train.py:892] (3/4) Epoch 8, batch 1050, loss[loss=0.2497, simple_loss=0.2976, pruned_loss=0.1009, over 19795.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.309, pruned_loss=0.104, over 3928667.24 frames. ], batch size: 65, lr: 1.94e-02, grad_scale: 16.0
+2023-03-27 22:56:16,556 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4166, 1.6015, 2.0499, 2.7498, 3.0646, 3.2484, 3.2227, 3.2028],
+       device='cuda:3'), covar=tensor([0.0881, 0.2224, 0.1208, 0.0530, 0.0339, 0.0184, 0.0248, 0.0292],
+       device='cuda:3'), in_proj_covar=tensor([0.0122, 0.0163, 0.0134, 0.0114, 0.0094, 0.0088, 0.0085, 0.0086],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 22:56:38,496 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5034, 3.6244, 2.2537, 3.9972, 3.9994, 1.6796, 3.2558, 3.1998],
+       device='cuda:3'), covar=tensor([0.0638, 0.0859, 0.2341, 0.0459, 0.0221, 0.2999, 0.0966, 0.0575],
+       device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0192, 0.0203, 0.0161, 0.0116, 0.0194, 0.0207, 0.0139],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 22:57:00,420 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:57:29,683 INFO [train.py:892] (3/4) Epoch 8, batch 1100, loss[loss=0.2465, simple_loss=0.2952, pruned_loss=0.09887, over 19758.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.3068, pruned_loss=0.102, over 3933013.06 frames. ], batch size: 213, lr: 1.93e-02, grad_scale: 16.0
+2023-03-27 22:57:35,032 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.89 vs. limit=5.0
+2023-03-27 22:57:43,915 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14093.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 22:58:04,940 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:58:13,330 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.085e+02 5.555e+02 6.523e+02 7.495e+02 1.242e+03, threshold=1.305e+03, percent-clipped=0.0
+2023-03-27 22:58:40,401 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14120.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 22:59:06,737 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 22:59:11,552 INFO [train.py:892] (3/4) Epoch 8, batch 1150, loss[loss=0.285, simple_loss=0.3268, pruned_loss=0.1216, over 19633.00 frames. ], tot_loss[loss=0.2558, simple_loss=0.3064, pruned_loss=0.1026, over 3937119.30 frames. ], batch size: 299, lr: 1.93e-02, grad_scale: 16.0
+2023-03-27 23:00:12,727 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:00:49,231 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14181.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 23:00:56,913 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5681, 3.6059, 3.8648, 3.6954, 3.8815, 3.2878, 3.4754, 3.4445],
+       device='cuda:3'), covar=tensor([0.1375, 0.1320, 0.1163, 0.1083, 0.1031, 0.1364, 0.2269, 0.2507],
+       device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0217, 0.0275, 0.0214, 0.0213, 0.0207, 0.0266, 0.0301],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-27 23:00:57,102 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.0329, 2.0265, 1.2053, 2.2276, 2.0183, 2.0861, 2.1917, 1.7237],
+       device='cuda:3'), covar=tensor([0.0629, 0.0637, 0.1114, 0.0398, 0.0512, 0.0343, 0.0379, 0.0812],
+       device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0102, 0.0122, 0.0106, 0.0092, 0.0085, 0.0099, 0.0112],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 23:00:58,061 INFO [train.py:892] (3/4) Epoch 8, batch 1200, loss[loss=0.4155, simple_loss=0.4461, pruned_loss=0.1924, over 19236.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3068, pruned_loss=0.1028, over 3940601.42 frames. ], batch size: 483, lr: 1.93e-02, grad_scale: 16.0
+2023-03-27 23:01:02,909 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0
+2023-03-27 23:01:14,733 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5009, 3.2021, 3.2389, 3.5872, 3.3360, 3.3579, 3.6056, 3.7741],
+       device='cuda:3'), covar=tensor([0.0683, 0.0415, 0.0565, 0.0318, 0.0558, 0.0544, 0.0412, 0.0329],
+       device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0132, 0.0158, 0.0133, 0.0129, 0.0108, 0.0124, 0.0145],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 23:01:41,588 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.316e+02 5.724e+02 6.994e+02 8.657e+02 1.468e+03, threshold=1.399e+03, percent-clipped=2.0
+2023-03-27 23:02:04,624 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4808, 1.9149, 2.4276, 2.9570, 3.2499, 3.5219, 3.5328, 3.5580],
+       device='cuda:3'), covar=tensor([0.0886, 0.1992, 0.1016, 0.0515, 0.0417, 0.0174, 0.0197, 0.0282],
+       device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0165, 0.0137, 0.0117, 0.0098, 0.0092, 0.0087, 0.0088],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 23:02:44,260 INFO [train.py:892] (3/4) Epoch 8, batch 1250, loss[loss=0.2191, simple_loss=0.2886, pruned_loss=0.07481, over 19766.00 frames. ], tot_loss[loss=0.2551, simple_loss=0.3062, pruned_loss=0.102, over 3941991.72 frames. ], batch size: 49, lr: 1.92e-02, grad_scale: 16.0
+2023-03-27 23:04:27,624 INFO [train.py:892] (3/4) Epoch 8, batch 1300, loss[loss=0.3477, simple_loss=0.3777, pruned_loss=0.1589, over 19623.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.3051, pruned_loss=0.1014, over 3942999.47 frames. ], batch size: 367, lr: 1.92e-02, grad_scale: 16.0
+2023-03-27 23:05:14,650 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.149e+02 5.379e+02 6.577e+02 8.584e+02 1.412e+03, threshold=1.315e+03, percent-clipped=1.0
+2023-03-27 23:06:13,952 INFO [train.py:892] (3/4) Epoch 8, batch 1350, loss[loss=0.2409, simple_loss=0.2922, pruned_loss=0.09479, over 19803.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.306, pruned_loss=0.1013, over 3944373.30 frames. ], batch size: 68, lr: 1.92e-02, grad_scale: 16.0
+2023-03-27 23:07:58,323 INFO [train.py:892] (3/4) Epoch 8, batch 1400, loss[loss=0.4277, simple_loss=0.4443, pruned_loss=0.2055, over 19406.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.3073, pruned_loss=0.1027, over 3944297.60 frames. ], batch size: 431, lr: 1.92e-02, grad_scale: 16.0
+2023-03-27 23:08:13,168 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:08:34,448 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-27 23:08:40,808 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.031e+02 5.328e+02 6.639e+02 8.222e+02 1.941e+03, threshold=1.328e+03, percent-clipped=1.0
+2023-03-27 23:08:56,019 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14413.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:09:24,649 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:09:41,896 INFO [train.py:892] (3/4) Epoch 8, batch 1450, loss[loss=0.2324, simple_loss=0.2896, pruned_loss=0.08754, over 19868.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.306, pruned_loss=0.1019, over 3946114.56 frames. ], batch size: 48, lr: 1.91e-02, grad_scale: 16.0
+2023-03-27 23:09:52,306 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:10:29,147 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:10:59,965 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:11:03,488 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14476.0, num_to_drop=1, layers_to_drop={3}
+2023-03-27 23:11:24,885 INFO [train.py:892] (3/4) Epoch 8, batch 1500, loss[loss=0.2612, simple_loss=0.3077, pruned_loss=0.1073, over 19800.00 frames. ], tot_loss[loss=0.2547, simple_loss=0.3056, pruned_loss=0.1019, over 3946634.10 frames. ], batch size: 200, lr: 1.91e-02, grad_scale: 32.0
+2023-03-27 23:11:26,218 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0
+2023-03-27 23:11:27,469 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14487.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:12:08,354 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.288e+02 5.447e+02 6.433e+02 7.944e+02 1.401e+03, threshold=1.287e+03, percent-clipped=2.0
+2023-03-27 23:13:06,135 INFO [train.py:892] (3/4) Epoch 8, batch 1550, loss[loss=0.2244, simple_loss=0.2788, pruned_loss=0.08504, over 19854.00 frames. ], tot_loss[loss=0.2566, simple_loss=0.3072, pruned_loss=0.103, over 3945783.75 frames. ], batch size: 104, lr: 1.91e-02, grad_scale: 16.0
+2023-03-27 23:13:12,924 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.65 vs. limit=2.0
+2023-03-27 23:13:33,875 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14548.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:14:54,504 INFO [train.py:892] (3/4) Epoch 8, batch 1600, loss[loss=0.2525, simple_loss=0.301, pruned_loss=0.102, over 19779.00 frames. ], tot_loss[loss=0.2557, simple_loss=0.3068, pruned_loss=0.1023, over 3944460.92 frames. ], batch size: 108, lr: 1.90e-02, grad_scale: 16.0
+2023-03-27 23:15:07,494 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4447, 4.6041, 5.0517, 4.5996, 4.1711, 4.8126, 4.6399, 5.1114],
+       device='cuda:3'), covar=tensor([0.1127, 0.0328, 0.0299, 0.0306, 0.0690, 0.0333, 0.0334, 0.0288],
+       device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0177, 0.0167, 0.0170, 0.0169, 0.0169, 0.0163, 0.0152],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 23:15:42,282 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.460e+02 5.480e+02 6.486e+02 8.210e+02 1.658e+03, threshold=1.297e+03, percent-clipped=3.0
+2023-03-27 23:16:31,218 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4628, 5.9109, 5.8946, 5.8685, 5.6069, 5.9247, 5.2480, 5.2727],
+       device='cuda:3'), covar=tensor([0.0402, 0.0401, 0.0510, 0.0345, 0.0507, 0.0512, 0.0605, 0.0923],
+       device='cuda:3'), in_proj_covar=tensor([0.0177, 0.0167, 0.0213, 0.0174, 0.0169, 0.0161, 0.0190, 0.0230],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-27 23:16:47,976 INFO [train.py:892] (3/4) Epoch 8, batch 1650, loss[loss=0.2283, simple_loss=0.2887, pruned_loss=0.08394, over 19782.00 frames. ], tot_loss[loss=0.2537, simple_loss=0.3051, pruned_loss=0.1011, over 3944825.40 frames. ], batch size: 87, lr: 1.90e-02, grad_scale: 16.0
+2023-03-27 23:17:26,349 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4697, 3.5192, 2.0471, 3.6802, 3.7982, 1.6078, 3.0802, 2.9126],
+       device='cuda:3'), covar=tensor([0.0614, 0.0718, 0.2477, 0.0469, 0.0261, 0.2987, 0.0978, 0.0606],
+       device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0190, 0.0198, 0.0160, 0.0118, 0.0189, 0.0202, 0.0139],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 23:17:32,444 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.94 vs. limit=5.0
+2023-03-27 23:18:06,371 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0
+2023-03-27 23:18:31,361 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8256, 3.1093, 3.4294, 2.2651, 3.4400, 2.7876, 2.5969, 3.7940],
+       device='cuda:3'), covar=tensor([0.0714, 0.0272, 0.0732, 0.0783, 0.0502, 0.0306, 0.0449, 0.0230],
+       device='cuda:3'), in_proj_covar=tensor([0.0050, 0.0051, 0.0052, 0.0078, 0.0050, 0.0044, 0.0045, 0.0040],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-27 23:18:38,577 INFO [train.py:892] (3/4) Epoch 8, batch 1700, loss[loss=0.2133, simple_loss=0.2805, pruned_loss=0.07301, over 19674.00 frames. ], tot_loss[loss=0.2512, simple_loss=0.3033, pruned_loss=0.09961, over 3945488.01 frames. ], batch size: 49, lr: 1.90e-02, grad_scale: 16.0
+2023-03-27 23:18:50,241 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3738, 3.3654, 1.8802, 4.2644, 3.6840, 4.1145, 4.2648, 3.2865],
+       device='cuda:3'), covar=tensor([0.0518, 0.0437, 0.1599, 0.0406, 0.0448, 0.0373, 0.0690, 0.0673],
+       device='cuda:3'), in_proj_covar=tensor([0.0111, 0.0105, 0.0124, 0.0110, 0.0095, 0.0088, 0.0100, 0.0114],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-27 23:19:29,027 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.414e+02 5.107e+02 6.150e+02 7.747e+02 1.488e+03, threshold=1.230e+03, percent-clipped=2.0
+2023-03-27 23:20:10,967 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:20:25,100 INFO [train.py:892] (3/4) Epoch 8, batch 1750, loss[loss=0.2177, simple_loss=0.2673, pruned_loss=0.08404, over 19836.00 frames. ], tot_loss[loss=0.2504, simple_loss=0.3025, pruned_loss=0.09914, over 3947723.19 frames. ], batch size: 171, lr: 1.90e-02, grad_scale: 16.0
+2023-03-27 23:20:32,959 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14740.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:21:11,609 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:21:29,723 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:21:41,290 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:21:41,472 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14776.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 23:22:00,369 INFO [train.py:892] (3/4) Epoch 8, batch 1800, loss[loss=0.235, simple_loss=0.2859, pruned_loss=0.09202, over 19772.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.3042, pruned_loss=0.1007, over 3946655.71 frames. ], batch size: 130, lr: 1.89e-02, grad_scale: 16.0
+2023-03-27 23:22:09,917 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5725, 3.0029, 3.2988, 3.5293, 3.7243, 3.9590, 4.4294, 4.7668],
+       device='cuda:3'), covar=tensor([0.0440, 0.1469, 0.1184, 0.1609, 0.1312, 0.1144, 0.0361, 0.0326],
+       device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0204, 0.0211, 0.0223, 0.0235, 0.0211, 0.0148, 0.0155],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 23:22:28,396 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:22:39,826 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14807.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:22:41,057 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.298e+02 5.336e+02 6.514e+02 7.931e+02 1.246e+03, threshold=1.303e+03, percent-clipped=1.0
+2023-03-27 23:22:55,672 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.97 vs. limit=5.0
+2023-03-27 23:23:10,738 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14824.0, num_to_drop=1, layers_to_drop={0}
+2023-03-27 23:23:30,836 INFO [train.py:892] (3/4) Epoch 8, batch 1850, loss[loss=0.2798, simple_loss=0.3424, pruned_loss=0.1086, over 19568.00 frames. ], tot_loss[loss=0.2545, simple_loss=0.3072, pruned_loss=0.1009, over 3944200.74 frames. ], batch size: 53, lr: 1.89e-02, grad_scale: 16.0
+2023-03-27 23:23:35,883 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5986, 3.0755, 2.3650, 1.8996, 2.3377, 3.1461, 3.0784, 2.7877],
+       device='cuda:3'), covar=tensor([0.0140, 0.0180, 0.0193, 0.0499, 0.0288, 0.0133, 0.0110, 0.0172],
+       device='cuda:3'), in_proj_covar=tensor([0.0050, 0.0046, 0.0052, 0.0067, 0.0066, 0.0045, 0.0040, 0.0044],
+       device='cuda:3'), out_proj_covar=tensor([1.1411e-04, 1.0709e-04, 1.1822e-04, 1.5820e-04, 1.5365e-04, 1.0609e-04,
+        9.6557e-05, 1.0116e-04], device='cuda:3')
+2023-03-27 23:24:36,969 INFO [train.py:892] (3/4) Epoch 9, batch 0, loss[loss=0.2443, simple_loss=0.2913, pruned_loss=0.09868, over 19749.00 frames. ], tot_loss[loss=0.2443, simple_loss=0.2913, pruned_loss=0.09868, over 19749.00 frames. ], batch size: 188, lr: 1.79e-02, grad_scale: 16.0
+2023-03-27 23:24:36,970 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-27 23:25:11,156 INFO [train.py:926] (3/4) Epoch 9, validation: loss=0.1843, simple_loss=0.2646, pruned_loss=0.05198, over 2883724.00 frames. 
+2023-03-27 23:25:11,157 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22385MB
+2023-03-27 23:25:16,197 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:26:58,511 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4324, 4.7366, 4.7385, 4.6936, 4.4534, 4.6765, 4.1884, 4.2810],
+       device='cuda:3'), covar=tensor([0.0411, 0.0416, 0.0641, 0.0409, 0.0617, 0.0607, 0.0598, 0.0880],
+       device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0175, 0.0218, 0.0178, 0.0170, 0.0161, 0.0195, 0.0232],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-27 23:27:06,404 INFO [train.py:892] (3/4) Epoch 9, batch 50, loss[loss=0.219, simple_loss=0.2884, pruned_loss=0.07487, over 19847.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.2957, pruned_loss=0.09594, over 891116.17 frames. ], batch size: 49, lr: 1.79e-02, grad_scale: 16.0
+2023-03-27 23:27:44,396 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.418e+02 5.041e+02 5.916e+02 7.576e+02 1.365e+03, threshold=1.183e+03, percent-clipped=1.0
+2023-03-27 23:28:08,598 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.8338, 1.9264, 2.1049, 1.9742, 1.7683, 1.8114, 1.6697, 1.9800],
+       device='cuda:3'), covar=tensor([0.0211, 0.0205, 0.0177, 0.0181, 0.0276, 0.0259, 0.0427, 0.0161],
+       device='cuda:3'), in_proj_covar=tensor([0.0039, 0.0040, 0.0041, 0.0035, 0.0043, 0.0042, 0.0056, 0.0040],
+       device='cuda:3'), out_proj_covar=tensor([8.7018e-05, 8.8610e-05, 9.1700e-05, 7.8807e-05, 9.5409e-05, 9.4205e-05,
+        1.2189e-04, 9.1023e-05], device='cuda:3')
+2023-03-27 23:28:50,443 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0
+2023-03-27 23:29:00,088 INFO [train.py:892] (3/4) Epoch 9, batch 100, loss[loss=0.2498, simple_loss=0.2985, pruned_loss=0.1005, over 19800.00 frames. ], tot_loss[loss=0.2421, simple_loss=0.2956, pruned_loss=0.09427, over 1570416.75 frames. ], batch size: 148, lr: 1.78e-02, grad_scale: 16.0
+2023-03-27 23:29:41,386 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0
+2023-03-27 23:29:46,674 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3672, 3.9704, 3.9682, 4.3132, 4.0630, 4.3812, 4.5431, 4.6475],
+       device='cuda:3'), covar=tensor([0.0617, 0.0345, 0.0474, 0.0288, 0.0565, 0.0283, 0.0358, 0.0256],
+       device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0136, 0.0161, 0.0136, 0.0133, 0.0114, 0.0129, 0.0151],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 23:30:48,423 INFO [train.py:892] (3/4) Epoch 9, batch 150, loss[loss=0.2221, simple_loss=0.2756, pruned_loss=0.08432, over 19841.00 frames. ], tot_loss[loss=0.2438, simple_loss=0.2966, pruned_loss=0.09556, over 2099030.11 frames. ], batch size: 115, lr: 1.78e-02, grad_scale: 16.0
+2023-03-27 23:31:28,936 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.328e+02 4.885e+02 5.781e+02 7.549e+02 1.487e+03, threshold=1.156e+03, percent-clipped=3.0
+2023-03-27 23:31:33,979 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1728, 2.9647, 4.1703, 3.5568, 3.9068, 4.0759, 4.0629, 3.7982],
+       device='cuda:3'), covar=tensor([0.0142, 0.0537, 0.0081, 0.0813, 0.0107, 0.0169, 0.0120, 0.0102],
+       device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0079, 0.0063, 0.0136, 0.0058, 0.0071, 0.0068, 0.0057],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0004, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 23:31:39,094 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.82 vs. limit=5.0
+2023-03-27 23:32:45,019 INFO [train.py:892] (3/4) Epoch 9, batch 200, loss[loss=0.256, simple_loss=0.3243, pruned_loss=0.09384, over 19530.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.298, pruned_loss=0.09615, over 2510629.78 frames. ], batch size: 54, lr: 1.78e-02, grad_scale: 16.0
+2023-03-27 23:33:47,981 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:34:38,148 INFO [train.py:892] (3/4) Epoch 9, batch 250, loss[loss=0.2545, simple_loss=0.3169, pruned_loss=0.0961, over 19619.00 frames. ], tot_loss[loss=0.2462, simple_loss=0.299, pruned_loss=0.0967, over 2830527.23 frames. ], batch size: 52, lr: 1.78e-02, grad_scale: 16.0
+2023-03-27 23:34:42,936 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8123, 1.9552, 2.5867, 3.0822, 3.4473, 3.6047, 3.5574, 3.5560],
+       device='cuda:3'), covar=tensor([0.0726, 0.2126, 0.0995, 0.0499, 0.0399, 0.0176, 0.0278, 0.0376],
+       device='cuda:3'), in_proj_covar=tensor([0.0126, 0.0166, 0.0140, 0.0117, 0.0101, 0.0093, 0.0088, 0.0090],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 23:34:48,918 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:35:13,648 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.451e+02 5.417e+02 6.236e+02 8.058e+02 1.427e+03, threshold=1.247e+03, percent-clipped=5.0
+2023-03-27 23:35:33,447 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.7736, 1.7513, 2.1446, 1.9077, 1.8032, 1.9387, 1.7475, 2.1685],
+       device='cuda:3'), covar=tensor([0.0211, 0.0264, 0.0185, 0.0197, 0.0256, 0.0209, 0.0378, 0.0147],
+       device='cuda:3'), in_proj_covar=tensor([0.0041, 0.0041, 0.0042, 0.0036, 0.0044, 0.0043, 0.0057, 0.0040],
+       device='cuda:3'), out_proj_covar=tensor([9.0622e-05, 9.0960e-05, 9.3036e-05, 8.1312e-05, 9.8177e-05, 9.4917e-05,
+        1.2471e-04, 9.1709e-05], device='cuda:3')
+2023-03-27 23:35:35,310 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:36:26,565 INFO [train.py:892] (3/4) Epoch 9, batch 300, loss[loss=0.2192, simple_loss=0.2775, pruned_loss=0.08042, over 19677.00 frames. ], tot_loss[loss=0.2464, simple_loss=0.2997, pruned_loss=0.09653, over 3075797.08 frames. ], batch size: 73, lr: 1.77e-02, grad_scale: 16.0
+2023-03-27 23:36:33,466 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:36:47,029 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1341, 3.0739, 4.5292, 3.3610, 3.9535, 4.3729, 2.3989, 2.6307],
+       device='cuda:3'), covar=tensor([0.0594, 0.2935, 0.0310, 0.0632, 0.1070, 0.0443, 0.1421, 0.1949],
+       device='cuda:3'), in_proj_covar=tensor([0.0276, 0.0314, 0.0231, 0.0195, 0.0300, 0.0217, 0.0246, 0.0240],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 23:37:32,037 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-27 23:38:17,837 INFO [train.py:892] (3/4) Epoch 9, batch 350, loss[loss=0.2358, simple_loss=0.2804, pruned_loss=0.09555, over 19870.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.2988, pruned_loss=0.09615, over 3271372.84 frames. ], batch size: 158, lr: 1.77e-02, grad_scale: 16.0
+2023-03-27 23:38:18,758 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:38:27,444 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:38:52,909 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.594e+02 5.622e+02 6.774e+02 8.741e+02 1.605e+03, threshold=1.355e+03, percent-clipped=2.0
+2023-03-27 23:40:06,032 INFO [train.py:892] (3/4) Epoch 9, batch 400, loss[loss=0.2413, simple_loss=0.3091, pruned_loss=0.08675, over 19887.00 frames. ], tot_loss[loss=0.2444, simple_loss=0.2978, pruned_loss=0.09546, over 3422822.13 frames. ], batch size: 92, lr: 1.77e-02, grad_scale: 16.0
+2023-03-27 23:40:40,988 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15256.0, num_to_drop=1, layers_to_drop={3}
+2023-03-27 23:41:58,944 INFO [train.py:892] (3/4) Epoch 9, batch 450, loss[loss=0.2346, simple_loss=0.2817, pruned_loss=0.09371, over 19871.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.2982, pruned_loss=0.09536, over 3539038.66 frames. ], batch size: 136, lr: 1.76e-02, grad_scale: 16.0
+2023-03-27 23:42:37,315 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.510e+02 5.167e+02 6.128e+02 7.422e+02 1.689e+03, threshold=1.226e+03, percent-clipped=1.0
+2023-03-27 23:42:45,007 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5427, 2.0332, 2.4474, 3.0323, 3.4323, 3.7903, 3.6188, 3.7500],
+       device='cuda:3'), covar=tensor([0.0999, 0.1800, 0.1158, 0.0514, 0.0352, 0.0167, 0.0259, 0.0283],
+       device='cuda:3'), in_proj_covar=tensor([0.0129, 0.0166, 0.0142, 0.0119, 0.0101, 0.0094, 0.0090, 0.0091],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 23:43:00,145 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9180, 3.2244, 3.2684, 4.1000, 2.6685, 3.2733, 2.6304, 2.3171],
+       device='cuda:3'), covar=tensor([0.0394, 0.2187, 0.0843, 0.0171, 0.1919, 0.0469, 0.1077, 0.1817],
+       device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0324, 0.0211, 0.0135, 0.0229, 0.0158, 0.0188, 0.0201],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0002, 0.0001, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 23:43:54,454 INFO [train.py:892] (3/4) Epoch 9, batch 500, loss[loss=0.2318, simple_loss=0.2919, pruned_loss=0.08588, over 19781.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.2976, pruned_loss=0.09482, over 3631255.06 frames. ], batch size: 91, lr: 1.76e-02, grad_scale: 16.0
+2023-03-27 23:45:12,634 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.1524, 5.3105, 5.6447, 5.3805, 5.2322, 5.0225, 5.2242, 5.0731],
+       device='cuda:3'), covar=tensor([0.1170, 0.0967, 0.0806, 0.0893, 0.0720, 0.0764, 0.1607, 0.1932],
+       device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0221, 0.0277, 0.0212, 0.0209, 0.0204, 0.0274, 0.0305],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-27 23:45:44,794 INFO [train.py:892] (3/4) Epoch 9, batch 550, loss[loss=0.2268, simple_loss=0.2874, pruned_loss=0.0831, over 19756.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.2984, pruned_loss=0.09498, over 3702685.90 frames. ], batch size: 110, lr: 1.76e-02, grad_scale: 16.0
+2023-03-27 23:45:56,353 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:46:26,494 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.786e+02 5.578e+02 6.357e+02 7.832e+02 1.667e+03, threshold=1.271e+03, percent-clipped=4.0
+2023-03-27 23:47:38,378 INFO [train.py:892] (3/4) Epoch 9, batch 600, loss[loss=0.2124, simple_loss=0.2745, pruned_loss=0.07512, over 19743.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.2984, pruned_loss=0.09486, over 3757368.58 frames. ], batch size: 99, lr: 1.76e-02, grad_scale: 16.0
+2023-03-27 23:47:47,256 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-27 23:48:49,598 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-27 23:49:30,488 INFO [train.py:892] (3/4) Epoch 9, batch 650, loss[loss=0.2514, simple_loss=0.3036, pruned_loss=0.09961, over 19669.00 frames. ], tot_loss[loss=0.2427, simple_loss=0.2969, pruned_loss=0.09422, over 3800642.27 frames. ], batch size: 73, lr: 1.75e-02, grad_scale: 16.0
+2023-03-27 23:50:08,034 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.542e+02 5.095e+02 6.402e+02 7.692e+02 1.403e+03, threshold=1.280e+03, percent-clipped=2.0
+2023-03-27 23:50:34,675 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0
+2023-03-27 23:50:53,437 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0
+2023-03-27 23:51:23,098 INFO [train.py:892] (3/4) Epoch 9, batch 700, loss[loss=0.2489, simple_loss=0.2937, pruned_loss=0.1021, over 19768.00 frames. ], tot_loss[loss=0.244, simple_loss=0.2983, pruned_loss=0.0949, over 3835167.22 frames. ], batch size: 163, lr: 1.75e-02, grad_scale: 16.0
+2023-03-27 23:51:45,034 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15551.0, num_to_drop=1, layers_to_drop={1}
+2023-03-27 23:52:44,694 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.3566, 2.7831, 1.9461, 1.7605, 2.3099, 2.7126, 2.4864, 2.7034],
+       device='cuda:3'), covar=tensor([0.0246, 0.0232, 0.0243, 0.0569, 0.0314, 0.0222, 0.0198, 0.0225],
+       device='cuda:3'), in_proj_covar=tensor([0.0053, 0.0049, 0.0056, 0.0069, 0.0069, 0.0047, 0.0042, 0.0045],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-27 23:53:00,237 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3588, 2.8799, 4.3791, 3.7179, 3.8929, 4.3018, 4.1937, 3.9919],
+       device='cuda:3'), covar=tensor([0.0134, 0.0594, 0.0070, 0.0898, 0.0124, 0.0177, 0.0137, 0.0092],
+       device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0081, 0.0064, 0.0136, 0.0058, 0.0070, 0.0068, 0.0057],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0004, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-27 23:53:15,945 INFO [train.py:892] (3/4) Epoch 9, batch 750, loss[loss=0.2278, simple_loss=0.2815, pruned_loss=0.08698, over 19374.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.2968, pruned_loss=0.09395, over 3861410.51 frames. ], batch size: 40, lr: 1.75e-02, grad_scale: 16.0
+2023-03-27 23:53:54,126 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.826e+02 5.119e+02 6.405e+02 7.720e+02 1.158e+03, threshold=1.281e+03, percent-clipped=0.0
+2023-03-27 23:54:31,881 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-27 23:55:10,138 INFO [train.py:892] (3/4) Epoch 9, batch 800, loss[loss=0.2772, simple_loss=0.3219, pruned_loss=0.1162, over 19778.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.2965, pruned_loss=0.09369, over 3881610.85 frames. ], batch size: 247, lr: 1.75e-02, grad_scale: 16.0
+2023-03-27 23:55:39,563 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.5843, 5.9456, 5.9064, 5.8261, 5.6724, 5.9500, 5.1634, 5.3040],
+       device='cuda:3'), covar=tensor([0.0338, 0.0323, 0.0477, 0.0350, 0.0463, 0.0470, 0.0576, 0.0803],
+       device='cuda:3'), in_proj_covar=tensor([0.0179, 0.0174, 0.0221, 0.0180, 0.0175, 0.0163, 0.0194, 0.0232],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-27 23:56:12,871 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9614, 3.0572, 3.3399, 2.5272, 3.3460, 2.7552, 2.7191, 3.4546],
+       device='cuda:3'), covar=tensor([0.0718, 0.0315, 0.0595, 0.0656, 0.0263, 0.0305, 0.0419, 0.0277],
+       device='cuda:3'), in_proj_covar=tensor([0.0055, 0.0055, 0.0057, 0.0085, 0.0054, 0.0050, 0.0048, 0.0044],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-27 23:57:05,124 INFO [train.py:892] (3/4) Epoch 9, batch 850, loss[loss=0.2486, simple_loss=0.2998, pruned_loss=0.09869, over 19837.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.2971, pruned_loss=0.09366, over 3895892.62 frames. ], batch size: 190, lr: 1.74e-02, grad_scale: 16.0
+2023-03-27 23:57:44,966 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.938e+02 5.306e+02 6.456e+02 8.498e+02 2.233e+03, threshold=1.291e+03, percent-clipped=2.0
+2023-03-27 23:58:58,984 INFO [train.py:892] (3/4) Epoch 9, batch 900, loss[loss=0.2321, simple_loss=0.2866, pruned_loss=0.08876, over 19817.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.2976, pruned_loss=0.09434, over 3907907.37 frames. ], batch size: 133, lr: 1.74e-02, grad_scale: 16.0
+2023-03-27 23:59:25,546 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3925, 3.4728, 2.1166, 3.7396, 3.7789, 1.6864, 3.1529, 2.9968],
+       device='cuda:3'), covar=tensor([0.0634, 0.0803, 0.2471, 0.0540, 0.0298, 0.2880, 0.0941, 0.0605],
+       device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0197, 0.0199, 0.0170, 0.0122, 0.0191, 0.0202, 0.0144],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 00:00:51,415 INFO [train.py:892] (3/4) Epoch 9, batch 950, loss[loss=0.2254, simple_loss=0.2779, pruned_loss=0.08644, over 19772.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.2978, pruned_loss=0.09417, over 3917703.58 frames. ], batch size: 87, lr: 1.74e-02, grad_scale: 16.0
+2023-03-28 00:01:31,054 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.223e+02 4.952e+02 6.049e+02 8.217e+02 1.611e+03, threshold=1.210e+03, percent-clipped=3.0
+2023-03-28 00:02:44,549 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7065, 3.8551, 2.1970, 4.1080, 4.1797, 1.6895, 3.4331, 3.2958],
+       device='cuda:3'), covar=tensor([0.0610, 0.0783, 0.2501, 0.0593, 0.0306, 0.3241, 0.1076, 0.0643],
+       device='cuda:3'), in_proj_covar=tensor([0.0172, 0.0198, 0.0198, 0.0171, 0.0124, 0.0193, 0.0204, 0.0143],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 00:02:47,464 INFO [train.py:892] (3/4) Epoch 9, batch 1000, loss[loss=0.2561, simple_loss=0.3114, pruned_loss=0.1004, over 19750.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.2995, pruned_loss=0.0954, over 3922884.10 frames. ], batch size: 250, lr: 1.74e-02, grad_scale: 16.0
+2023-03-28 00:02:54,467 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3563, 3.2937, 4.7816, 3.4865, 3.9373, 4.4145, 2.4007, 2.6787],
+       device='cuda:3'), covar=tensor([0.0502, 0.2473, 0.0270, 0.0541, 0.1147, 0.0409, 0.1452, 0.1754],
+       device='cuda:3'), in_proj_covar=tensor([0.0277, 0.0318, 0.0237, 0.0199, 0.0306, 0.0222, 0.0252, 0.0241],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 00:03:10,926 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:04:24,349 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:04:41,513 INFO [train.py:892] (3/4) Epoch 9, batch 1050, loss[loss=0.2209, simple_loss=0.2718, pruned_loss=0.08502, over 19883.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.2997, pruned_loss=0.09556, over 3929228.27 frames. ], batch size: 47, lr: 1.73e-02, grad_scale: 16.0
+2023-03-28 00:04:58,486 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:04:58,996 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0
+2023-03-28 00:05:13,537 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0
+2023-03-28 00:05:18,403 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.416e+02 5.707e+02 6.582e+02 7.521e+02 1.130e+03, threshold=1.316e+03, percent-clipped=0.0
+2023-03-28 00:06:20,775 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9846, 2.9501, 1.4284, 3.8064, 3.3243, 3.6576, 3.8030, 2.7415],
+       device='cuda:3'), covar=tensor([0.0523, 0.0519, 0.1790, 0.0375, 0.0447, 0.0294, 0.0365, 0.0798],
+       device='cuda:3'), in_proj_covar=tensor([0.0112, 0.0107, 0.0126, 0.0112, 0.0096, 0.0090, 0.0105, 0.0114],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 00:06:34,643 INFO [train.py:892] (3/4) Epoch 9, batch 1100, loss[loss=0.3228, simple_loss=0.3898, pruned_loss=0.1279, over 18782.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.2998, pruned_loss=0.09556, over 3934446.24 frames. ], batch size: 564, lr: 1.73e-02, grad_scale: 16.0
+2023-03-28 00:06:42,232 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:08:26,890 INFO [train.py:892] (3/4) Epoch 9, batch 1150, loss[loss=0.2197, simple_loss=0.2826, pruned_loss=0.07847, over 19876.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.2981, pruned_loss=0.0945, over 3938215.13 frames. ], batch size: 99, lr: 1.73e-02, grad_scale: 16.0
+2023-03-28 00:08:54,470 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:09:09,129 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.185e+02 5.244e+02 6.208e+02 7.678e+02 1.349e+03, threshold=1.242e+03, percent-clipped=1.0
+2023-03-28 00:10:03,798 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2387, 3.5359, 3.7614, 4.3447, 2.8873, 3.3844, 2.7559, 2.6430],
+       device='cuda:3'), covar=tensor([0.0444, 0.2145, 0.0754, 0.0203, 0.2134, 0.0694, 0.1119, 0.1740],
+       device='cuda:3'), in_proj_covar=tensor([0.0180, 0.0325, 0.0215, 0.0139, 0.0230, 0.0161, 0.0188, 0.0205],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 00:10:22,656 INFO [train.py:892] (3/4) Epoch 9, batch 1200, loss[loss=0.3756, simple_loss=0.4087, pruned_loss=0.1712, over 19412.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.2983, pruned_loss=0.09473, over 3941309.35 frames. ], batch size: 412, lr: 1.73e-02, grad_scale: 16.0
+2023-03-28 00:11:12,463 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:12:03,652 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16085.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:12:11,651 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9886, 3.1635, 3.3215, 4.0088, 2.7753, 3.2789, 2.7576, 2.3212],
+       device='cuda:3'), covar=tensor([0.0399, 0.2522, 0.0876, 0.0173, 0.2045, 0.0566, 0.0973, 0.1888],
+       device='cuda:3'), in_proj_covar=tensor([0.0181, 0.0328, 0.0216, 0.0139, 0.0231, 0.0163, 0.0190, 0.0206],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 00:12:15,732 INFO [train.py:892] (3/4) Epoch 9, batch 1250, loss[loss=0.2669, simple_loss=0.3136, pruned_loss=0.1101, over 19801.00 frames. ], tot_loss[loss=0.2424, simple_loss=0.2968, pruned_loss=0.09401, over 3944295.74 frames. ], batch size: 211, lr: 1.72e-02, grad_scale: 16.0
+2023-03-28 00:12:49,933 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.594e+02 5.905e+02 6.787e+02 8.144e+02 1.550e+03, threshold=1.357e+03, percent-clipped=6.0
+2023-03-28 00:14:05,695 INFO [train.py:892] (3/4) Epoch 9, batch 1300, loss[loss=0.2527, simple_loss=0.3023, pruned_loss=0.1015, over 19774.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.2978, pruned_loss=0.09466, over 3945130.75 frames. ], batch size: 130, lr: 1.72e-02, grad_scale: 16.0
+2023-03-28 00:14:17,849 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:14:25,355 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:15:42,842 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.8519, 1.8808, 2.0292, 1.8817, 1.6221, 1.8252, 1.7122, 1.8634],
+       device='cuda:3'), covar=tensor([0.0291, 0.0210, 0.0206, 0.0206, 0.0347, 0.0371, 0.0377, 0.0234],
+       device='cuda:3'), in_proj_covar=tensor([0.0042, 0.0043, 0.0045, 0.0036, 0.0047, 0.0043, 0.0057, 0.0041],
+       device='cuda:3'), out_proj_covar=tensor([9.4518e-05, 9.5849e-05, 9.8585e-05, 8.1042e-05, 1.0552e-04, 9.6496e-05,
+        1.2585e-04, 9.4678e-05], device='cuda:3')
+2023-03-28 00:15:57,803 INFO [train.py:892] (3/4) Epoch 9, batch 1350, loss[loss=0.2464, simple_loss=0.29, pruned_loss=0.1014, over 19840.00 frames. ], tot_loss[loss=0.2433, simple_loss=0.2977, pruned_loss=0.09444, over 3945950.48 frames. ], batch size: 160, lr: 1.72e-02, grad_scale: 16.0
+2023-03-28 00:16:35,999 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.606e+02 5.207e+02 6.312e+02 7.771e+02 1.214e+03, threshold=1.262e+03, percent-clipped=0.0
+2023-03-28 00:16:41,298 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:17:20,310 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1331, 2.4945, 3.3028, 2.7936, 2.7796, 3.2579, 1.8693, 2.0922],
+       device='cuda:3'), covar=tensor([0.0724, 0.2088, 0.0470, 0.0544, 0.1072, 0.0583, 0.1502, 0.1814],
+       device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0323, 0.0245, 0.0202, 0.0313, 0.0232, 0.0258, 0.0248],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 00:17:50,391 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16239.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:17:53,149 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0
+2023-03-28 00:17:53,557 INFO [train.py:892] (3/4) Epoch 9, batch 1400, loss[loss=0.2364, simple_loss=0.2805, pruned_loss=0.09613, over 19821.00 frames. ], tot_loss[loss=0.2421, simple_loss=0.2962, pruned_loss=0.09396, over 3948513.05 frames. ], batch size: 167, lr: 1.72e-02, grad_scale: 16.0
+2023-03-28 00:18:22,862 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0
+2023-03-28 00:18:48,296 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4834, 3.4683, 5.0514, 3.6333, 4.2896, 4.5848, 2.5076, 2.7822],
+       device='cuda:3'), covar=tensor([0.0540, 0.2113, 0.0281, 0.0563, 0.0989, 0.0387, 0.1414, 0.1795],
+       device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0323, 0.0245, 0.0203, 0.0312, 0.0232, 0.0259, 0.0249],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 00:19:40,168 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8526, 4.4192, 4.4040, 4.9380, 4.5612, 5.1313, 4.9723, 5.1079],
+       device='cuda:3'), covar=tensor([0.0637, 0.0346, 0.0536, 0.0272, 0.0602, 0.0238, 0.0353, 0.0280],
+       device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0141, 0.0165, 0.0137, 0.0136, 0.0118, 0.0127, 0.0156],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 00:19:46,048 INFO [train.py:892] (3/4) Epoch 9, batch 1450, loss[loss=0.2241, simple_loss=0.2885, pruned_loss=0.07987, over 19697.00 frames. ], tot_loss[loss=0.2415, simple_loss=0.2963, pruned_loss=0.09335, over 3948291.14 frames. ], batch size: 75, lr: 1.71e-02, grad_scale: 16.0
+2023-03-28 00:19:51,816 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.32 vs. limit=5.0
+2023-03-28 00:20:26,644 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.895e+02 4.802e+02 5.731e+02 6.907e+02 1.143e+03, threshold=1.146e+03, percent-clipped=0.0
+2023-03-28 00:21:16,284 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.4644, 1.6765, 1.4200, 0.9517, 1.4638, 1.6431, 1.5357, 1.6441],
+       device='cuda:3'), covar=tensor([0.0239, 0.0179, 0.0222, 0.0476, 0.0416, 0.0170, 0.0145, 0.0136],
+       device='cuda:3'), in_proj_covar=tensor([0.0055, 0.0052, 0.0057, 0.0070, 0.0070, 0.0048, 0.0042, 0.0045],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-28 00:21:40,554 INFO [train.py:892] (3/4) Epoch 9, batch 1500, loss[loss=0.3015, simple_loss=0.3452, pruned_loss=0.1289, over 19723.00 frames. ], tot_loss[loss=0.2428, simple_loss=0.2973, pruned_loss=0.09412, over 3947423.58 frames. ], batch size: 305, lr: 1.71e-02, grad_scale: 16.0
+2023-03-28 00:22:17,147 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16357.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:22:38,524 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-28 00:22:45,777 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16371.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:23:27,934 INFO [train.py:892] (3/4) Epoch 9, batch 1550, loss[loss=0.2271, simple_loss=0.2884, pruned_loss=0.08284, over 19644.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.2991, pruned_loss=0.09467, over 3945573.32 frames. ], batch size: 72, lr: 1.71e-02, grad_scale: 16.0
+2023-03-28 00:23:50,472 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2856, 2.9789, 3.0403, 3.3682, 3.1364, 3.0601, 3.4596, 3.5715],
+       device='cuda:3'), covar=tensor([0.0709, 0.0486, 0.0569, 0.0356, 0.0615, 0.0686, 0.0373, 0.0346],
+       device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0140, 0.0165, 0.0138, 0.0136, 0.0117, 0.0127, 0.0154],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 00:23:50,616 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.5059, 1.7765, 1.4382, 0.9352, 1.5768, 1.6466, 1.5949, 1.7252],
+       device='cuda:3'), covar=tensor([0.0237, 0.0166, 0.0206, 0.0455, 0.0326, 0.0163, 0.0147, 0.0133],
+       device='cuda:3'), in_proj_covar=tensor([0.0055, 0.0052, 0.0057, 0.0070, 0.0070, 0.0049, 0.0043, 0.0046],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-28 00:24:05,932 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.814e+02 5.273e+02 6.072e+02 7.301e+02 1.702e+03, threshold=1.214e+03, percent-clipped=5.0
+2023-03-28 00:25:00,865 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16432.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:25:19,722 INFO [train.py:892] (3/4) Epoch 9, batch 1600, loss[loss=0.2399, simple_loss=0.3032, pruned_loss=0.08826, over 19829.00 frames. ], tot_loss[loss=0.2446, simple_loss=0.2991, pruned_loss=0.09504, over 3946852.56 frames. ], batch size: 57, lr: 1.71e-02, grad_scale: 16.0
+2023-03-28 00:25:20,721 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16441.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:27:01,099 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6607, 2.6119, 3.1875, 2.4098, 3.1213, 2.5412, 2.9738, 2.9636],
+       device='cuda:3'), covar=tensor([0.0517, 0.0401, 0.0336, 0.0560, 0.0219, 0.0340, 0.0246, 0.0280],
+       device='cuda:3'), in_proj_covar=tensor([0.0053, 0.0055, 0.0056, 0.0084, 0.0054, 0.0050, 0.0048, 0.0044],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-28 00:27:15,093 INFO [train.py:892] (3/4) Epoch 9, batch 1650, loss[loss=0.2291, simple_loss=0.2807, pruned_loss=0.08872, over 19838.00 frames. ], tot_loss[loss=0.2442, simple_loss=0.2995, pruned_loss=0.09443, over 3943728.67 frames. ], batch size: 161, lr: 1.71e-02, grad_scale: 16.0
+2023-03-28 00:27:27,479 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0
+2023-03-28 00:27:46,425 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:27:52,669 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.351e+02 5.128e+02 6.235e+02 7.643e+02 1.695e+03, threshold=1.247e+03, percent-clipped=3.0
+2023-03-28 00:29:04,288 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16539.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:29:07,572 INFO [train.py:892] (3/4) Epoch 9, batch 1700, loss[loss=0.2559, simple_loss=0.3082, pruned_loss=0.1018, over 19873.00 frames. ], tot_loss[loss=0.2441, simple_loss=0.2996, pruned_loss=0.09427, over 3943912.68 frames. ], batch size: 64, lr: 1.70e-02, grad_scale: 32.0
+2023-03-28 00:30:46,937 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16587.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:30:53,431 INFO [train.py:892] (3/4) Epoch 9, batch 1750, loss[loss=0.2787, simple_loss=0.3288, pruned_loss=0.1143, over 19771.00 frames. ], tot_loss[loss=0.2435, simple_loss=0.2993, pruned_loss=0.0939, over 3945690.90 frames. ], batch size: 66, lr: 1.70e-02, grad_scale: 32.0
+2023-03-28 00:31:28,592 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.990e+02 5.355e+02 6.431e+02 7.521e+02 1.438e+03, threshold=1.286e+03, percent-clipped=3.0
+2023-03-28 00:32:29,083 INFO [train.py:892] (3/4) Epoch 9, batch 1800, loss[loss=0.2284, simple_loss=0.2785, pruned_loss=0.08917, over 19868.00 frames. ], tot_loss[loss=0.2436, simple_loss=0.2992, pruned_loss=0.09403, over 3945078.96 frames. ], batch size: 157, lr: 1.70e-02, grad_scale: 32.0
+2023-03-28 00:32:57,700 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:33:01,947 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0
+2023-03-28 00:33:58,384 INFO [train.py:892] (3/4) Epoch 9, batch 1850, loss[loss=0.2516, simple_loss=0.3216, pruned_loss=0.09083, over 19838.00 frames. ], tot_loss[loss=0.2417, simple_loss=0.2984, pruned_loss=0.09251, over 3945690.66 frames. ], batch size: 58, lr: 1.70e-02, grad_scale: 32.0
+2023-03-28 00:35:08,025 INFO [train.py:892] (3/4) Epoch 10, batch 0, loss[loss=0.2246, simple_loss=0.2727, pruned_loss=0.08826, over 19843.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.2727, pruned_loss=0.08826, over 19843.00 frames. ], batch size: 180, lr: 1.61e-02, grad_scale: 32.0
+2023-03-28 00:35:08,025 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-28 00:35:42,737 INFO [train.py:926] (3/4) Epoch 10, validation: loss=0.1801, simple_loss=0.2601, pruned_loss=0.05003, over 2883724.00 frames. 
+2023-03-28 00:35:42,738 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-28 00:36:04,276 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16705.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:36:11,170 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.977e+02 5.093e+02 5.971e+02 7.550e+02 1.362e+03, threshold=1.194e+03, percent-clipped=1.0
+2023-03-28 00:36:55,601 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:37:31,376 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:37:43,762 INFO [train.py:892] (3/4) Epoch 10, batch 50, loss[loss=0.2409, simple_loss=0.2938, pruned_loss=0.09401, over 19852.00 frames. ], tot_loss[loss=0.242, simple_loss=0.2963, pruned_loss=0.09386, over 887805.59 frames. ], batch size: 142, lr: 1.61e-02, grad_scale: 32.0
+2023-03-28 00:37:56,141 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.2009, 2.6082, 2.8213, 2.3533, 2.2356, 2.1767, 2.2907, 2.6758],
+       device='cuda:3'), covar=tensor([0.0190, 0.0194, 0.0211, 0.0173, 0.0240, 0.0325, 0.0348, 0.0206],
+       device='cuda:3'), in_proj_covar=tensor([0.0044, 0.0044, 0.0046, 0.0038, 0.0049, 0.0047, 0.0060, 0.0042],
+       device='cuda:3'), out_proj_covar=tensor([9.7619e-05, 9.8847e-05, 1.0129e-04, 8.6581e-05, 1.0982e-04, 1.0449e-04,
+        1.3131e-04, 9.6376e-05], device='cuda:3')
+2023-03-28 00:39:23,326 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16789.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:39:38,183 INFO [train.py:892] (3/4) Epoch 10, batch 100, loss[loss=0.2698, simple_loss=0.3274, pruned_loss=0.1061, over 19526.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.2943, pruned_loss=0.09207, over 1567005.66 frames. ], batch size: 54, lr: 1.61e-02, grad_scale: 32.0
+2023-03-28 00:39:59,355 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16805.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:40:04,282 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.771e+02 5.348e+02 6.393e+02 7.632e+02 1.172e+03, threshold=1.279e+03, percent-clipped=0.0
+2023-03-28 00:41:04,044 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0550, 3.2995, 1.9007, 4.1599, 3.7160, 3.9829, 4.2475, 3.0505],
+       device='cuda:3'), covar=tensor([0.0613, 0.0547, 0.1352, 0.0344, 0.0432, 0.0306, 0.0259, 0.0780],
+       device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0109, 0.0124, 0.0112, 0.0098, 0.0091, 0.0105, 0.0115],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 00:41:31,411 INFO [train.py:892] (3/4) Epoch 10, batch 150, loss[loss=0.226, simple_loss=0.2869, pruned_loss=0.08258, over 19664.00 frames. ], tot_loss[loss=0.2398, simple_loss=0.2952, pruned_loss=0.09218, over 2094238.74 frames. ], batch size: 64, lr: 1.61e-02, grad_scale: 32.0
+2023-03-28 00:41:49,416 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16853.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:42:22,990 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16868.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 00:43:24,900 INFO [train.py:892] (3/4) Epoch 10, batch 200, loss[loss=0.2225, simple_loss=0.2872, pruned_loss=0.07886, over 19624.00 frames. ], tot_loss[loss=0.2404, simple_loss=0.2963, pruned_loss=0.09227, over 2504077.38 frames. ], batch size: 65, lr: 1.60e-02, grad_scale: 16.0
+2023-03-28 00:43:47,976 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:43:53,275 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.558e+02 5.040e+02 6.126e+02 7.288e+02 1.628e+03, threshold=1.225e+03, percent-clipped=1.0
+2023-03-28 00:44:41,663 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16929.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 00:45:17,957 INFO [train.py:892] (3/4) Epoch 10, batch 250, loss[loss=0.2704, simple_loss=0.3251, pruned_loss=0.1078, over 19841.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.2977, pruned_loss=0.09297, over 2823621.67 frames. ], batch size: 58, lr: 1.60e-02, grad_scale: 16.0
+2023-03-28 00:46:06,455 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16967.0, num_to_drop=1, layers_to_drop={3}
+2023-03-28 00:47:10,513 INFO [train.py:892] (3/4) Epoch 10, batch 300, loss[loss=0.2455, simple_loss=0.3115, pruned_loss=0.08981, over 19617.00 frames. ], tot_loss[loss=0.2404, simple_loss=0.2963, pruned_loss=0.09228, over 3071372.02 frames. ], batch size: 51, lr: 1.60e-02, grad_scale: 16.0
+2023-03-28 00:47:41,234 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.429e+02 5.102e+02 6.617e+02 8.376e+02 1.348e+03, threshold=1.323e+03, percent-clipped=5.0
+2023-03-28 00:48:08,800 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1036, 4.1999, 4.5235, 4.1671, 3.8940, 4.3902, 4.1230, 4.6531],
+       device='cuda:3'), covar=tensor([0.1117, 0.0347, 0.0358, 0.0348, 0.0742, 0.0390, 0.0370, 0.0287],
+       device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0181, 0.0176, 0.0180, 0.0175, 0.0179, 0.0171, 0.0163],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 00:48:23,259 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:48:45,993 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3396, 3.6491, 1.9005, 4.4945, 3.8940, 4.2095, 4.3552, 3.3509],
+       device='cuda:3'), covar=tensor([0.0549, 0.0411, 0.1416, 0.0364, 0.0376, 0.0257, 0.0384, 0.0646],
+       device='cuda:3'), in_proj_covar=tensor([0.0118, 0.0112, 0.0128, 0.0115, 0.0101, 0.0095, 0.0110, 0.0120],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 00:49:09,356 INFO [train.py:892] (3/4) Epoch 10, batch 350, loss[loss=0.2454, simple_loss=0.2938, pruned_loss=0.09849, over 19764.00 frames. ], tot_loss[loss=0.2405, simple_loss=0.2961, pruned_loss=0.0924, over 3266486.56 frames. ], batch size: 213, lr: 1.60e-02, grad_scale: 16.0
+2023-03-28 00:50:14,171 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17075.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:50:59,673 INFO [train.py:892] (3/4) Epoch 10, batch 400, loss[loss=0.2286, simple_loss=0.2792, pruned_loss=0.089, over 19803.00 frames. ], tot_loss[loss=0.241, simple_loss=0.2966, pruned_loss=0.09271, over 3418653.62 frames. ], batch size: 132, lr: 1.59e-02, grad_scale: 16.0
+2023-03-28 00:51:28,622 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.044e+02 4.864e+02 6.028e+02 7.083e+02 1.226e+03, threshold=1.206e+03, percent-clipped=0.0
+2023-03-28 00:52:51,299 INFO [train.py:892] (3/4) Epoch 10, batch 450, loss[loss=0.2259, simple_loss=0.2909, pruned_loss=0.08045, over 19724.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.296, pruned_loss=0.09221, over 3537628.59 frames. ], batch size: 101, lr: 1.59e-02, grad_scale: 16.0
+2023-03-28 00:54:31,074 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:54:47,713 INFO [train.py:892] (3/4) Epoch 10, batch 500, loss[loss=0.2349, simple_loss=0.2994, pruned_loss=0.08514, over 19683.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.2959, pruned_loss=0.09231, over 3630449.78 frames. ], batch size: 75, lr: 1.59e-02, grad_scale: 16.0
+2023-03-28 00:54:59,383 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7597, 2.1388, 2.7722, 3.2487, 3.5618, 3.7791, 3.8589, 3.9705],
+       device='cuda:3'), covar=tensor([0.0828, 0.1788, 0.1015, 0.0402, 0.0316, 0.0172, 0.0178, 0.0188],
+       device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0169, 0.0149, 0.0123, 0.0107, 0.0100, 0.0092, 0.0093],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 00:55:18,934 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.795e+02 5.158e+02 6.137e+02 7.953e+02 1.330e+03, threshold=1.227e+03, percent-clipped=1.0
+2023-03-28 00:55:50,578 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17224.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 00:56:30,754 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3964, 3.4324, 1.8216, 4.3691, 3.9142, 4.1191, 4.3687, 3.4530],
+       device='cuda:3'), covar=tensor([0.0513, 0.0462, 0.1534, 0.0398, 0.0348, 0.0348, 0.0301, 0.0562],
+       device='cuda:3'), in_proj_covar=tensor([0.0118, 0.0111, 0.0128, 0.0115, 0.0101, 0.0095, 0.0110, 0.0121],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 00:56:39,664 INFO [train.py:892] (3/4) Epoch 10, batch 550, loss[loss=0.2398, simple_loss=0.3028, pruned_loss=0.08842, over 19894.00 frames. ], tot_loss[loss=0.2392, simple_loss=0.2949, pruned_loss=0.09182, over 3702489.54 frames. ], batch size: 94, lr: 1.59e-02, grad_scale: 16.0
+2023-03-28 00:56:49,538 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 00:57:16,252 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17262.0, num_to_drop=1, layers_to_drop={3}
+2023-03-28 00:58:30,085 INFO [train.py:892] (3/4) Epoch 10, batch 600, loss[loss=0.3021, simple_loss=0.3393, pruned_loss=0.1324, over 19774.00 frames. ], tot_loss[loss=0.2389, simple_loss=0.2946, pruned_loss=0.09164, over 3757789.99 frames. ], batch size: 247, lr: 1.59e-02, grad_scale: 16.0
+2023-03-28 00:59:00,056 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.688e+02 5.075e+02 6.633e+02 7.997e+02 1.370e+03, threshold=1.327e+03, percent-clipped=3.0
+2023-03-28 00:59:16,888 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.8575, 2.2558, 1.7170, 1.3846, 1.9020, 2.0835, 2.0876, 2.0554],
+       device='cuda:3'), covar=tensor([0.0252, 0.0148, 0.0223, 0.0493, 0.0344, 0.0151, 0.0153, 0.0166],
+       device='cuda:3'), in_proj_covar=tensor([0.0056, 0.0051, 0.0059, 0.0072, 0.0071, 0.0050, 0.0043, 0.0047],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-28 01:00:24,288 INFO [train.py:892] (3/4) Epoch 10, batch 650, loss[loss=0.2213, simple_loss=0.2849, pruned_loss=0.07887, over 19769.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.2943, pruned_loss=0.09134, over 3799342.88 frames. ], batch size: 46, lr: 1.58e-02, grad_scale: 16.0
+2023-03-28 01:01:01,104 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:01:11,712 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6193, 2.7371, 3.8571, 2.9283, 3.3570, 3.6715, 2.1867, 2.3033],
+       device='cuda:3'), covar=tensor([0.0639, 0.2322, 0.0417, 0.0638, 0.1118, 0.0604, 0.1455, 0.1948],
+       device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0324, 0.0249, 0.0204, 0.0315, 0.0235, 0.0262, 0.0250],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 01:01:50,431 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17383.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 01:02:00,213 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.09 vs. limit=5.0
+2023-03-28 01:02:19,162 INFO [train.py:892] (3/4) Epoch 10, batch 700, loss[loss=0.1937, simple_loss=0.2557, pruned_loss=0.06586, over 19560.00 frames. ], tot_loss[loss=0.2385, simple_loss=0.2944, pruned_loss=0.09132, over 3832731.72 frames. ], batch size: 47, lr: 1.58e-02, grad_scale: 16.0
+2023-03-28 01:02:48,236 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.516e+02 5.181e+02 6.486e+02 7.594e+02 1.327e+03, threshold=1.297e+03, percent-clipped=1.0
+2023-03-28 01:03:20,580 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:04:08,313 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17444.0, num_to_drop=1, layers_to_drop={3}
+2023-03-28 01:04:11,117 INFO [train.py:892] (3/4) Epoch 10, batch 750, loss[loss=0.2128, simple_loss=0.2795, pruned_loss=0.07307, over 19749.00 frames. ], tot_loss[loss=0.2373, simple_loss=0.2934, pruned_loss=0.09057, over 3860163.78 frames. ], batch size: 84, lr: 1.58e-02, grad_scale: 16.0
+2023-03-28 01:04:33,343 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0
+2023-03-28 01:04:51,689 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0
+2023-03-28 01:06:02,453 INFO [train.py:892] (3/4) Epoch 10, batch 800, loss[loss=0.2566, simple_loss=0.3091, pruned_loss=0.102, over 19768.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.2947, pruned_loss=0.09125, over 3877606.82 frames. ], batch size: 88, lr: 1.58e-02, grad_scale: 16.0
+2023-03-28 01:06:22,923 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4281, 2.7614, 2.1924, 1.7363, 2.3950, 2.5251, 2.5289, 2.6364],
+       device='cuda:3'), covar=tensor([0.0226, 0.0204, 0.0206, 0.0496, 0.0300, 0.0206, 0.0157, 0.0157],
+       device='cuda:3'), in_proj_covar=tensor([0.0057, 0.0051, 0.0059, 0.0072, 0.0072, 0.0050, 0.0044, 0.0047],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-28 01:06:32,361 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.436e+02 5.370e+02 6.448e+02 7.965e+02 2.022e+03, threshold=1.290e+03, percent-clipped=3.0
+2023-03-28 01:07:07,222 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17524.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 01:07:53,917 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:07:54,967 INFO [train.py:892] (3/4) Epoch 10, batch 850, loss[loss=0.2075, simple_loss=0.2623, pruned_loss=0.0764, over 19869.00 frames. ], tot_loss[loss=0.2386, simple_loss=0.2949, pruned_loss=0.09112, over 3892944.38 frames. ], batch size: 157, lr: 1.58e-02, grad_scale: 16.0
+2023-03-28 01:08:34,154 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17562.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 01:08:53,872 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17572.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 01:09:47,744 INFO [train.py:892] (3/4) Epoch 10, batch 900, loss[loss=0.2376, simple_loss=0.2924, pruned_loss=0.09136, over 19647.00 frames. ], tot_loss[loss=0.2369, simple_loss=0.2938, pruned_loss=0.09003, over 3905134.20 frames. ], batch size: 69, lr: 1.57e-02, grad_scale: 16.0
+2023-03-28 01:10:17,490 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.166e+02 5.484e+02 6.655e+02 8.099e+02 1.732e+03, threshold=1.331e+03, percent-clipped=3.0
+2023-03-28 01:10:20,842 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:11:39,742 INFO [train.py:892] (3/4) Epoch 10, batch 950, loss[loss=0.2414, simple_loss=0.3047, pruned_loss=0.08909, over 19875.00 frames. ], tot_loss[loss=0.2366, simple_loss=0.2932, pruned_loss=0.08995, over 3915919.47 frames. ], batch size: 89, lr: 1.57e-02, grad_scale: 16.0
+2023-03-28 01:13:32,336 INFO [train.py:892] (3/4) Epoch 10, batch 1000, loss[loss=0.2261, simple_loss=0.3056, pruned_loss=0.07331, over 19526.00 frames. ], tot_loss[loss=0.2364, simple_loss=0.293, pruned_loss=0.08994, over 3923138.49 frames. ], batch size: 54, lr: 1.57e-02, grad_scale: 16.0
+2023-03-28 01:13:41,691 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:14:02,555 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.971e+02 5.177e+02 6.007e+02 7.029e+02 1.681e+03, threshold=1.201e+03, percent-clipped=2.0
+2023-03-28 01:14:21,024 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:14:47,938 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:15:09,066 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17739.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 01:15:26,013 INFO [train.py:892] (3/4) Epoch 10, batch 1050, loss[loss=0.2216, simple_loss=0.2763, pruned_loss=0.08344, over 19836.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.2931, pruned_loss=0.08951, over 3929241.06 frames. ], batch size: 146, lr: 1.57e-02, grad_scale: 16.0
+2023-03-28 01:15:57,275 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17761.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:17:06,153 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17791.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:17:16,085 INFO [train.py:892] (3/4) Epoch 10, batch 1100, loss[loss=0.2396, simple_loss=0.2963, pruned_loss=0.09143, over 19770.00 frames. ], tot_loss[loss=0.2365, simple_loss=0.2933, pruned_loss=0.08979, over 3932361.31 frames. ], batch size: 70, lr: 1.57e-02, grad_scale: 16.0
+2023-03-28 01:17:45,685 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.952e+02 5.448e+02 6.504e+02 7.690e+02 1.433e+03, threshold=1.301e+03, percent-clipped=1.0
+2023-03-28 01:18:35,236 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0
+2023-03-28 01:19:10,698 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:19:11,946 INFO [train.py:892] (3/4) Epoch 10, batch 1150, loss[loss=0.2237, simple_loss=0.2804, pruned_loss=0.0835, over 19757.00 frames. ], tot_loss[loss=0.2351, simple_loss=0.2921, pruned_loss=0.0891, over 3937034.74 frames. ], batch size: 102, lr: 1.56e-02, grad_scale: 16.0
+2023-03-28 01:19:46,754 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0980, 3.0915, 1.5014, 3.9011, 3.4596, 3.8560, 3.9975, 2.8798],
+       device='cuda:3'), covar=tensor([0.0580, 0.0469, 0.1792, 0.0477, 0.0457, 0.0362, 0.0423, 0.0739],
+       device='cuda:3'), in_proj_covar=tensor([0.0120, 0.0112, 0.0129, 0.0117, 0.0102, 0.0095, 0.0111, 0.0122],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 01:21:00,362 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:21:06,384 INFO [train.py:892] (3/4) Epoch 10, batch 1200, loss[loss=0.2246, simple_loss=0.284, pruned_loss=0.08263, over 19663.00 frames. ], tot_loss[loss=0.2355, simple_loss=0.2927, pruned_loss=0.0892, over 3941128.62 frames. ], batch size: 43, lr: 1.56e-02, grad_scale: 16.0
+2023-03-28 01:21:31,151 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1786, 4.2737, 4.6524, 4.2109, 4.0033, 4.5287, 4.3221, 4.8043],
+       device='cuda:3'), covar=tensor([0.1078, 0.0331, 0.0349, 0.0334, 0.0720, 0.0373, 0.0380, 0.0279],
+       device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0185, 0.0179, 0.0188, 0.0180, 0.0187, 0.0177, 0.0170],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 01:21:34,285 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.688e+02 5.021e+02 6.012e+02 7.162e+02 1.373e+03, threshold=1.202e+03, percent-clipped=1.0
+2023-03-28 01:22:56,644 INFO [train.py:892] (3/4) Epoch 10, batch 1250, loss[loss=0.236, simple_loss=0.2873, pruned_loss=0.09231, over 19776.00 frames. ], tot_loss[loss=0.2361, simple_loss=0.2929, pruned_loss=0.0896, over 3942481.91 frames. ], batch size: 152, lr: 1.56e-02, grad_scale: 16.0
+2023-03-28 01:23:30,761 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:24:49,111 INFO [train.py:892] (3/4) Epoch 10, batch 1300, loss[loss=0.2203, simple_loss=0.2798, pruned_loss=0.08036, over 19775.00 frames. ], tot_loss[loss=0.2384, simple_loss=0.295, pruned_loss=0.09092, over 3942025.20 frames. ], batch size: 108, lr: 1.56e-02, grad_scale: 16.0
+2023-03-28 01:25:24,407 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.596e+02 5.072e+02 5.807e+02 7.353e+02 1.557e+03, threshold=1.161e+03, percent-clipped=4.0
+2023-03-28 01:25:25,337 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8987, 5.1893, 5.2131, 5.1466, 4.8977, 5.1483, 4.6086, 4.7077],
+       device='cuda:3'), covar=tensor([0.0360, 0.0331, 0.0504, 0.0361, 0.0582, 0.0578, 0.0552, 0.0814],
+       device='cuda:3'), in_proj_covar=tensor([0.0182, 0.0183, 0.0224, 0.0186, 0.0178, 0.0168, 0.0201, 0.0238],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 01:25:44,770 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:25:50,501 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18021.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:26:07,211 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.77 vs. limit=2.0
+2023-03-28 01:26:34,172 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18039.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 01:26:47,397 INFO [train.py:892] (3/4) Epoch 10, batch 1350, loss[loss=0.2537, simple_loss=0.3081, pruned_loss=0.09962, over 19656.00 frames. ], tot_loss[loss=0.2403, simple_loss=0.2966, pruned_loss=0.09201, over 3942357.34 frames. ], batch size: 57, lr: 1.56e-02, grad_scale: 16.0
+2023-03-28 01:26:53,893 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8210, 4.7653, 4.6797, 5.1571, 5.0403, 5.5322, 4.9661, 5.2336],
+       device='cuda:3'), covar=tensor([0.1018, 0.0484, 0.0578, 0.0378, 0.0576, 0.0294, 0.0681, 0.0615],
+       device='cuda:3'), in_proj_covar=tensor([0.0123, 0.0142, 0.0166, 0.0136, 0.0138, 0.0119, 0.0128, 0.0155],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 01:27:11,970 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:27:33,202 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18066.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:28:17,476 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.1538, 5.0882, 5.6078, 5.4647, 5.4075, 4.9861, 5.2102, 5.2093],
+       device='cuda:3'), covar=tensor([0.1260, 0.1163, 0.0936, 0.0861, 0.0657, 0.0808, 0.2162, 0.1926],
+       device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0234, 0.0297, 0.0224, 0.0222, 0.0212, 0.0287, 0.0315],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 01:28:19,547 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:28:21,747 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18087.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 01:28:41,071 INFO [train.py:892] (3/4) Epoch 10, batch 1400, loss[loss=0.2663, simple_loss=0.3165, pruned_loss=0.1081, over 19645.00 frames. ], tot_loss[loss=0.2389, simple_loss=0.2957, pruned_loss=0.09108, over 3944142.92 frames. ], batch size: 299, lr: 1.55e-02, grad_scale: 16.0
+2023-03-28 01:28:42,961 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0
+2023-03-28 01:29:09,605 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.393e+02 4.780e+02 5.656e+02 7.029e+02 1.296e+03, threshold=1.131e+03, percent-clipped=2.0
+2023-03-28 01:29:43,960 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2500, 3.0121, 4.2881, 3.7519, 3.9843, 4.2826, 4.2791, 3.9889],
+       device='cuda:3'), covar=tensor([0.0143, 0.0558, 0.0090, 0.0787, 0.0100, 0.0170, 0.0108, 0.0114],
+       device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0085, 0.0068, 0.0139, 0.0061, 0.0074, 0.0070, 0.0062],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0004, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 01:30:37,309 INFO [train.py:892] (3/4) Epoch 10, batch 1450, loss[loss=0.2122, simple_loss=0.2806, pruned_loss=0.0719, over 19748.00 frames. ], tot_loss[loss=0.2374, simple_loss=0.2944, pruned_loss=0.09025, over 3945795.63 frames. ], batch size: 84, lr: 1.55e-02, grad_scale: 16.0
+2023-03-28 01:31:12,136 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3714, 2.3043, 2.6499, 2.5697, 2.8273, 2.7516, 3.1720, 3.4424],
+       device='cuda:3'), covar=tensor([0.0583, 0.1569, 0.1489, 0.1665, 0.1463, 0.1357, 0.0501, 0.0539],
+       device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0211, 0.0227, 0.0228, 0.0250, 0.0225, 0.0169, 0.0171],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 01:31:18,106 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4012, 2.5899, 3.7374, 2.8065, 3.3186, 3.4393, 2.1266, 2.1902],
+       device='cuda:3'), covar=tensor([0.0858, 0.2791, 0.0504, 0.0642, 0.1096, 0.0722, 0.1546, 0.2038],
+       device='cuda:3'), in_proj_covar=tensor([0.0293, 0.0329, 0.0257, 0.0207, 0.0317, 0.0243, 0.0270, 0.0255],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 01:31:45,138 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18174.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:31:47,422 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18175.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:32:31,628 INFO [train.py:892] (3/4) Epoch 10, batch 1500, loss[loss=0.2221, simple_loss=0.2765, pruned_loss=0.08387, over 19888.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.2935, pruned_loss=0.09, over 3947510.02 frames. ], batch size: 176, lr: 1.55e-02, grad_scale: 16.0
+2023-03-28 01:33:02,132 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0
+2023-03-28 01:33:02,761 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.786e+02 5.080e+02 6.023e+02 7.472e+02 1.411e+03, threshold=1.205e+03, percent-clipped=3.0
+2023-03-28 01:33:47,858 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.9500, 2.0878, 2.3663, 2.0135, 1.9695, 1.9630, 2.0365, 2.2125],
+       device='cuda:3'), covar=tensor([0.0259, 0.0311, 0.0230, 0.0248, 0.0334, 0.0292, 0.0365, 0.0300],
+       device='cuda:3'), in_proj_covar=tensor([0.0045, 0.0044, 0.0045, 0.0038, 0.0049, 0.0046, 0.0059, 0.0042],
+       device='cuda:3'), out_proj_covar=tensor([9.9746e-05, 9.8320e-05, 1.0026e-04, 8.6834e-05, 1.0997e-04, 1.0229e-04,
+        1.3100e-04, 9.5740e-05], device='cuda:3')
+2023-03-28 01:34:02,767 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:34:04,942 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:34:27,161 INFO [train.py:892] (3/4) Epoch 10, batch 1550, loss[loss=0.2657, simple_loss=0.319, pruned_loss=0.1062, over 19705.00 frames. ], tot_loss[loss=0.238, simple_loss=0.295, pruned_loss=0.09052, over 3946806.53 frames. ], batch size: 295, lr: 1.55e-02, grad_scale: 16.0
+2023-03-28 01:34:54,069 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.47 vs. limit=5.0
+2023-03-28 01:36:22,683 INFO [train.py:892] (3/4) Epoch 10, batch 1600, loss[loss=0.2026, simple_loss=0.2725, pruned_loss=0.06635, over 19945.00 frames. ], tot_loss[loss=0.2359, simple_loss=0.2931, pruned_loss=0.08935, over 3949326.45 frames. ], batch size: 46, lr: 1.55e-02, grad_scale: 16.0
+2023-03-28 01:36:49,440 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.277e+02 5.238e+02 6.120e+02 7.230e+02 1.238e+03, threshold=1.224e+03, percent-clipped=2.0
+2023-03-28 01:37:03,893 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:37:21,289 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2392, 2.3432, 3.1831, 3.4937, 3.9031, 4.4477, 4.3721, 4.5827],
+       device='cuda:3'), covar=tensor([0.0647, 0.1876, 0.1072, 0.0473, 0.0300, 0.0151, 0.0171, 0.0278],
+       device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0173, 0.0155, 0.0126, 0.0108, 0.0101, 0.0092, 0.0097],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 01:38:12,208 INFO [train.py:892] (3/4) Epoch 10, batch 1650, loss[loss=0.2081, simple_loss=0.2759, pruned_loss=0.07017, over 19731.00 frames. ], tot_loss[loss=0.2342, simple_loss=0.2918, pruned_loss=0.08825, over 3950310.39 frames. ], batch size: 80, lr: 1.54e-02, grad_scale: 16.0
+2023-03-28 01:38:36,358 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18356.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:38:52,781 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6219, 2.8764, 2.3790, 1.9429, 2.3513, 2.8531, 2.8158, 2.8153],
+       device='cuda:3'), covar=tensor([0.0185, 0.0176, 0.0226, 0.0448, 0.0341, 0.0169, 0.0125, 0.0136],
+       device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0054, 0.0061, 0.0073, 0.0074, 0.0051, 0.0045, 0.0048],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-28 01:39:11,553 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1097, 4.0312, 2.3389, 4.5979, 4.7503, 1.8709, 3.7642, 3.4977],
+       device='cuda:3'), covar=tensor([0.0568, 0.0759, 0.2459, 0.0484, 0.0245, 0.3001, 0.0914, 0.0662],
+       device='cuda:3'), in_proj_covar=tensor([0.0185, 0.0208, 0.0204, 0.0185, 0.0144, 0.0193, 0.0212, 0.0151],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 01:39:43,883 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:40:06,507 INFO [train.py:892] (3/4) Epoch 10, batch 1700, loss[loss=0.218, simple_loss=0.284, pruned_loss=0.07595, over 19848.00 frames. ], tot_loss[loss=0.2335, simple_loss=0.291, pruned_loss=0.08797, over 3951385.66 frames. ], batch size: 49, lr: 1.54e-02, grad_scale: 16.0
+2023-03-28 01:40:24,865 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:40:35,304 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.958e+02 5.263e+02 6.960e+02 8.795e+02 1.250e+03, threshold=1.392e+03, percent-clipped=4.0
+2023-03-28 01:41:26,419 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8197, 2.7913, 4.2048, 3.0005, 3.6441, 3.8008, 2.2293, 2.2592],
+       device='cuda:3'), covar=tensor([0.0593, 0.2537, 0.0433, 0.0607, 0.1034, 0.0589, 0.1459, 0.2009],
+       device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0326, 0.0254, 0.0210, 0.0319, 0.0245, 0.0270, 0.0253],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 01:41:30,293 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:41:54,335 INFO [train.py:892] (3/4) Epoch 10, batch 1750, loss[loss=0.2039, simple_loss=0.2774, pruned_loss=0.06526, over 19610.00 frames. ], tot_loss[loss=0.2331, simple_loss=0.2907, pruned_loss=0.08778, over 3951048.96 frames. ], batch size: 48, lr: 1.54e-02, grad_scale: 16.0
+2023-03-28 01:42:47,617 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.8707, 2.1587, 1.8047, 1.2592, 1.8952, 2.1748, 2.1053, 2.1250],
+       device='cuda:3'), covar=tensor([0.0268, 0.0207, 0.0264, 0.0571, 0.0415, 0.0204, 0.0171, 0.0183],
+       device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0054, 0.0062, 0.0074, 0.0075, 0.0052, 0.0046, 0.0049],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-28 01:43:20,470 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18490.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:43:30,625 INFO [train.py:892] (3/4) Epoch 10, batch 1800, loss[loss=0.217, simple_loss=0.2683, pruned_loss=0.08287, over 19766.00 frames. ], tot_loss[loss=0.2336, simple_loss=0.2912, pruned_loss=0.08807, over 3949729.56 frames. ], batch size: 155, lr: 1.54e-02, grad_scale: 16.0
+2023-03-28 01:43:52,043 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18508.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 01:43:52,992 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.190e+02 4.885e+02 5.842e+02 7.130e+02 1.277e+03, threshold=1.168e+03, percent-clipped=0.0
+2023-03-28 01:44:30,977 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:44:32,860 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:44:58,714 INFO [train.py:892] (3/4) Epoch 10, batch 1850, loss[loss=0.3013, simple_loss=0.3553, pruned_loss=0.1237, over 19586.00 frames. ], tot_loss[loss=0.2336, simple_loss=0.2926, pruned_loss=0.08735, over 3948251.61 frames. ], batch size: 53, lr: 1.54e-02, grad_scale: 16.0
+2023-03-28 01:46:02,464 INFO [train.py:892] (3/4) Epoch 11, batch 0, loss[loss=0.398, simple_loss=0.4144, pruned_loss=0.1908, over 19452.00 frames. ], tot_loss[loss=0.398, simple_loss=0.4144, pruned_loss=0.1908, over 19452.00 frames. ], batch size: 396, lr: 1.47e-02, grad_scale: 16.0
+2023-03-28 01:46:02,464 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-28 01:46:36,863 INFO [train.py:926] (3/4) Epoch 11, validation: loss=0.1783, simple_loss=0.2585, pruned_loss=0.04909, over 2883724.00 frames. 
+2023-03-28 01:46:36,864 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-28 01:46:37,842 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:47:18,685 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18569.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 01:48:16,508 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1362, 4.1511, 4.4763, 4.1477, 3.8650, 4.3308, 4.0979, 4.6388],
+       device='cuda:3'), covar=tensor([0.0957, 0.0310, 0.0298, 0.0304, 0.0859, 0.0414, 0.0398, 0.0267],
+       device='cuda:3'), in_proj_covar=tensor([0.0254, 0.0191, 0.0184, 0.0193, 0.0186, 0.0192, 0.0184, 0.0172],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 01:48:31,658 INFO [train.py:892] (3/4) Epoch 11, batch 50, loss[loss=0.3709, simple_loss=0.404, pruned_loss=0.1689, over 19418.00 frames. ], tot_loss[loss=0.2389, simple_loss=0.294, pruned_loss=0.09187, over 889968.84 frames. ], batch size: 412, lr: 1.46e-02, grad_scale: 16.0
+2023-03-28 01:48:47,839 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.428e+02 5.161e+02 6.219e+02 7.148e+02 1.502e+03, threshold=1.244e+03, percent-clipped=1.0
+2023-03-28 01:49:05,489 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:50:10,720 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3655, 3.5588, 3.7892, 4.5813, 2.9067, 3.5580, 3.0796, 2.7387],
+       device='cuda:3'), covar=tensor([0.0397, 0.2116, 0.0781, 0.0182, 0.2120, 0.0703, 0.0964, 0.1665],
+       device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0330, 0.0221, 0.0141, 0.0235, 0.0168, 0.0193, 0.0207],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 01:50:21,238 INFO [train.py:892] (3/4) Epoch 11, batch 100, loss[loss=0.3093, simple_loss=0.3444, pruned_loss=0.1371, over 19758.00 frames. ], tot_loss[loss=0.2351, simple_loss=0.2913, pruned_loss=0.08942, over 1568441.97 frames. ], batch size: 321, lr: 1.46e-02, grad_scale: 16.0
+2023-03-28 01:50:49,850 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18664.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:51:15,963 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:52:13,645 INFO [train.py:892] (3/4) Epoch 11, batch 150, loss[loss=0.2, simple_loss=0.2595, pruned_loss=0.07021, over 19849.00 frames. ], tot_loss[loss=0.2356, simple_loss=0.2919, pruned_loss=0.08963, over 2094600.49 frames. ], batch size: 109, lr: 1.46e-02, grad_scale: 16.0
+2023-03-28 01:52:35,350 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.696e+02 5.324e+02 6.674e+02 8.055e+02 1.622e+03, threshold=1.335e+03, percent-clipped=4.0
+2023-03-28 01:52:54,248 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-28 01:53:31,460 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:54:09,400 INFO [train.py:892] (3/4) Epoch 11, batch 200, loss[loss=0.2004, simple_loss=0.263, pruned_loss=0.06891, over 19853.00 frames. ], tot_loss[loss=0.2336, simple_loss=0.2905, pruned_loss=0.0883, over 2507610.62 frames. ], batch size: 99, lr: 1.46e-02, grad_scale: 16.0
+2023-03-28 01:54:33,677 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8587, 2.1733, 2.8816, 3.1731, 3.7752, 4.1631, 4.0609, 4.3441],
+       device='cuda:3'), covar=tensor([0.0811, 0.2097, 0.1189, 0.0532, 0.0296, 0.0132, 0.0306, 0.0213],
+       device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0171, 0.0155, 0.0126, 0.0108, 0.0102, 0.0095, 0.0098],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 01:54:59,768 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3784, 4.1404, 4.2583, 4.0057, 4.4326, 3.0953, 3.6697, 2.3259],
+       device='cuda:3'), covar=tensor([0.0188, 0.0194, 0.0127, 0.0145, 0.0104, 0.0762, 0.0692, 0.1247],
+       device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0111, 0.0097, 0.0109, 0.0096, 0.0116, 0.0130, 0.0112],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 01:55:27,093 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.16 vs. limit=5.0
+2023-03-28 01:56:02,245 INFO [train.py:892] (3/4) Epoch 11, batch 250, loss[loss=0.2254, simple_loss=0.2894, pruned_loss=0.08071, over 19620.00 frames. ], tot_loss[loss=0.2348, simple_loss=0.292, pruned_loss=0.08878, over 2827088.14 frames. ], batch size: 65, lr: 1.46e-02, grad_scale: 16.0
+2023-03-28 01:56:14,021 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9129, 2.7759, 3.2658, 2.4599, 3.3431, 2.6106, 2.7525, 3.2169],
+       device='cuda:3'), covar=tensor([0.0649, 0.0370, 0.0434, 0.0664, 0.0224, 0.0330, 0.0394, 0.0277],
+       device='cuda:3'), in_proj_covar=tensor([0.0056, 0.0058, 0.0060, 0.0090, 0.0057, 0.0053, 0.0051, 0.0047],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-28 01:56:18,958 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.168e+02 4.641e+02 5.456e+02 6.612e+02 1.459e+03, threshold=1.091e+03, percent-clipped=1.0
+2023-03-28 01:57:08,081 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18830.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:57:08,201 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.9725, 1.9459, 1.3663, 2.0665, 1.9519, 1.9776, 1.9423, 1.5647],
+       device='cuda:3'), covar=tensor([0.0755, 0.0748, 0.1138, 0.0457, 0.0662, 0.0460, 0.0466, 0.0868],
+       device='cuda:3'), in_proj_covar=tensor([0.0121, 0.0114, 0.0129, 0.0117, 0.0104, 0.0098, 0.0114, 0.0121],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 01:57:10,093 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18831.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:57:44,357 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18846.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:57:56,651 INFO [train.py:892] (3/4) Epoch 11, batch 300, loss[loss=0.2658, simple_loss=0.3243, pruned_loss=0.1037, over 19692.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.2937, pruned_loss=0.0893, over 3072698.20 frames. ], batch size: 283, lr: 1.46e-02, grad_scale: 16.0
+2023-03-28 01:58:25,301 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18864.0, num_to_drop=1, layers_to_drop={3}
+2023-03-28 01:58:56,603 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:58:58,548 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 01:59:48,637 INFO [train.py:892] (3/4) Epoch 11, batch 350, loss[loss=0.214, simple_loss=0.2833, pruned_loss=0.07232, over 19881.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.2921, pruned_loss=0.08779, over 3267240.41 frames. ], batch size: 77, lr: 1.45e-02, grad_scale: 32.0
+2023-03-28 02:00:05,180 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.253e+02 5.242e+02 6.180e+02 7.064e+02 1.171e+03, threshold=1.236e+03, percent-clipped=1.0
+2023-03-28 02:00:54,934 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18930.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:01:38,286 INFO [train.py:892] (3/4) Epoch 11, batch 400, loss[loss=0.231, simple_loss=0.2954, pruned_loss=0.08325, over 19596.00 frames. ], tot_loss[loss=0.234, simple_loss=0.2922, pruned_loss=0.08787, over 3417555.21 frames. ], batch size: 44, lr: 1.45e-02, grad_scale: 32.0
+2023-03-28 02:02:24,482 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7478, 4.8729, 5.2585, 4.8730, 4.2371, 5.0139, 4.9459, 5.4522],
+       device='cuda:3'), covar=tensor([0.0930, 0.0271, 0.0358, 0.0314, 0.0652, 0.0400, 0.0298, 0.0247],
+       device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0188, 0.0182, 0.0191, 0.0181, 0.0188, 0.0180, 0.0172],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 02:03:08,396 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:03:30,248 INFO [train.py:892] (3/4) Epoch 11, batch 450, loss[loss=0.2135, simple_loss=0.2733, pruned_loss=0.07683, over 19807.00 frames. ], tot_loss[loss=0.2338, simple_loss=0.2922, pruned_loss=0.08773, over 3533589.59 frames. ], batch size: 167, lr: 1.45e-02, grad_scale: 32.0
+2023-03-28 02:03:47,202 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.137e+02 4.905e+02 5.923e+02 7.329e+02 1.154e+03, threshold=1.185e+03, percent-clipped=0.0
+2023-03-28 02:04:35,126 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:05:18,622 INFO [train.py:892] (3/4) Epoch 11, batch 500, loss[loss=0.2121, simple_loss=0.2628, pruned_loss=0.08064, over 19811.00 frames. ], tot_loss[loss=0.2324, simple_loss=0.2906, pruned_loss=0.08715, over 3627212.55 frames. ], batch size: 148, lr: 1.45e-02, grad_scale: 32.0
+2023-03-28 02:06:40,816 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1819, 2.8536, 3.0850, 3.0953, 3.4031, 3.4025, 3.9342, 4.3265],
+       device='cuda:3'), covar=tensor([0.0550, 0.1568, 0.1458, 0.1848, 0.1566, 0.1346, 0.0437, 0.0383],
+       device='cuda:3'), in_proj_covar=tensor([0.0204, 0.0215, 0.0235, 0.0231, 0.0257, 0.0226, 0.0172, 0.0177],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 02:06:45,353 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.67 vs. limit=2.0
+2023-03-28 02:07:11,291 INFO [train.py:892] (3/4) Epoch 11, batch 550, loss[loss=0.3596, simple_loss=0.3969, pruned_loss=0.1611, over 19423.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.2903, pruned_loss=0.08668, over 3699486.21 frames. ], batch size: 412, lr: 1.45e-02, grad_scale: 32.0
+2023-03-28 02:07:27,681 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.995e+02 4.889e+02 6.159e+02 7.664e+02 1.397e+03, threshold=1.232e+03, percent-clipped=1.0
+2023-03-28 02:08:01,454 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4894, 1.9030, 2.1110, 2.8183, 3.1287, 3.2865, 3.2357, 3.3665],
+       device='cuda:3'), covar=tensor([0.0820, 0.1880, 0.1174, 0.0506, 0.0343, 0.0218, 0.0246, 0.0267],
+       device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0169, 0.0154, 0.0126, 0.0109, 0.0102, 0.0096, 0.0096],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 02:08:11,701 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7390, 5.0097, 5.0116, 4.9943, 4.7264, 4.9923, 4.3903, 4.4977],
+       device='cuda:3'), covar=tensor([0.0361, 0.0373, 0.0525, 0.0355, 0.0522, 0.0534, 0.0697, 0.0861],
+       device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0195, 0.0235, 0.0199, 0.0186, 0.0182, 0.0212, 0.0250],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 02:08:16,543 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19129.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 02:08:51,801 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19146.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:09:03,310 INFO [train.py:892] (3/4) Epoch 11, batch 600, loss[loss=0.2228, simple_loss=0.284, pruned_loss=0.0808, over 19662.00 frames. ], tot_loss[loss=0.2307, simple_loss=0.2896, pruned_loss=0.08594, over 3755082.37 frames. ], batch size: 43, lr: 1.44e-02, grad_scale: 32.0
+2023-03-28 02:09:34,472 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19164.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 02:09:40,422 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9710, 2.9689, 4.3886, 3.2312, 3.8155, 3.8743, 2.2749, 2.3332],
+       device='cuda:3'), covar=tensor([0.0699, 0.2626, 0.0342, 0.0701, 0.1216, 0.0746, 0.1724, 0.2283],
+       device='cuda:3'), in_proj_covar=tensor([0.0296, 0.0332, 0.0259, 0.0214, 0.0324, 0.0256, 0.0276, 0.0260],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 02:09:46,703 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0
+2023-03-28 02:10:33,422 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19190.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 02:10:41,243 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:10:55,609 INFO [train.py:892] (3/4) Epoch 11, batch 650, loss[loss=0.2296, simple_loss=0.2891, pruned_loss=0.08508, over 19730.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.2899, pruned_loss=0.08652, over 3796592.32 frames. ], batch size: 71, lr: 1.44e-02, grad_scale: 32.0
+2023-03-28 02:11:13,801 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.235e+02 5.164e+02 6.078e+02 7.650e+02 1.184e+03, threshold=1.216e+03, percent-clipped=0.0
+2023-03-28 02:11:21,848 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19212.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 02:12:15,460 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2951, 4.2616, 2.7386, 4.7794, 4.9895, 1.9526, 4.0916, 3.6347],
+       device='cuda:3'), covar=tensor([0.0538, 0.0721, 0.2251, 0.0518, 0.0338, 0.2866, 0.0908, 0.0597],
+       device='cuda:3'), in_proj_covar=tensor([0.0187, 0.0210, 0.0207, 0.0194, 0.0151, 0.0195, 0.0220, 0.0153],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 02:12:45,810 INFO [train.py:892] (3/4) Epoch 11, batch 700, loss[loss=0.2221, simple_loss=0.2779, pruned_loss=0.08314, over 19765.00 frames. ], tot_loss[loss=0.2316, simple_loss=0.29, pruned_loss=0.08664, over 3830172.59 frames. ], batch size: 198, lr: 1.44e-02, grad_scale: 32.0
+2023-03-28 02:14:07,902 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:14:41,781 INFO [train.py:892] (3/4) Epoch 11, batch 750, loss[loss=0.2412, simple_loss=0.3042, pruned_loss=0.08906, over 19888.00 frames. ], tot_loss[loss=0.2308, simple_loss=0.2895, pruned_loss=0.08603, over 3857475.52 frames. ], batch size: 63, lr: 1.44e-02, grad_scale: 16.0
+2023-03-28 02:15:01,324 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.572e+02 4.867e+02 5.905e+02 6.947e+02 1.334e+03, threshold=1.181e+03, percent-clipped=1.0
+2023-03-28 02:15:47,100 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:16:03,826 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0133, 2.2778, 3.1954, 3.2838, 3.8508, 4.1955, 4.1879, 4.4641],
+       device='cuda:3'), covar=tensor([0.0755, 0.1960, 0.1127, 0.0515, 0.0289, 0.0175, 0.0265, 0.0237],
+       device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0165, 0.0152, 0.0124, 0.0106, 0.0101, 0.0095, 0.0093],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 02:16:12,347 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1022, 3.3263, 3.3822, 4.1218, 2.6711, 3.3357, 2.6848, 2.6074],
+       device='cuda:3'), covar=tensor([0.0414, 0.2560, 0.0986, 0.0265, 0.2268, 0.0706, 0.1227, 0.1706],
+       device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0332, 0.0227, 0.0145, 0.0239, 0.0172, 0.0195, 0.0211],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 02:16:33,248 INFO [train.py:892] (3/4) Epoch 11, batch 800, loss[loss=0.2217, simple_loss=0.2796, pruned_loss=0.08188, over 19812.00 frames. ], tot_loss[loss=0.23, simple_loss=0.2893, pruned_loss=0.08537, over 3878486.15 frames. ], batch size: 231, lr: 1.44e-02, grad_scale: 16.0
+2023-03-28 02:17:37,520 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19378.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:18:13,077 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:18:27,071 INFO [train.py:892] (3/4) Epoch 11, batch 850, loss[loss=0.1954, simple_loss=0.2604, pruned_loss=0.06514, over 19715.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.2882, pruned_loss=0.08452, over 3895553.31 frames. ], batch size: 81, lr: 1.44e-02, grad_scale: 16.0
+2023-03-28 02:18:49,366 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.769e+02 4.779e+02 5.853e+02 6.766e+02 1.621e+03, threshold=1.171e+03, percent-clipped=3.0
+2023-03-28 02:20:21,653 INFO [train.py:892] (3/4) Epoch 11, batch 900, loss[loss=0.2113, simple_loss=0.2816, pruned_loss=0.07044, over 19770.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.2868, pruned_loss=0.08351, over 3909048.23 frames. ], batch size: 69, lr: 1.43e-02, grad_scale: 16.0
+2023-03-28 02:20:30,508 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:21:39,558 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19485.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 02:22:13,032 INFO [train.py:892] (3/4) Epoch 11, batch 950, loss[loss=0.2373, simple_loss=0.2944, pruned_loss=0.09014, over 19866.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.2898, pruned_loss=0.08541, over 3916458.26 frames. ], batch size: 154, lr: 1.43e-02, grad_scale: 16.0
+2023-03-28 02:22:30,343 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.330e+02 4.753e+02 5.819e+02 7.309e+02 1.447e+03, threshold=1.164e+03, percent-clipped=3.0
+2023-03-28 02:23:14,966 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6345, 3.8675, 4.1870, 3.8384, 3.6785, 4.1035, 3.9073, 4.2847],
+       device='cuda:3'), covar=tensor([0.1300, 0.0377, 0.0460, 0.0415, 0.1072, 0.0476, 0.0403, 0.0404],
+       device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0186, 0.0182, 0.0191, 0.0184, 0.0187, 0.0183, 0.0172],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 02:23:42,265 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-28 02:24:02,920 INFO [train.py:892] (3/4) Epoch 11, batch 1000, loss[loss=0.2519, simple_loss=0.3042, pruned_loss=0.09984, over 19787.00 frames. ], tot_loss[loss=0.2295, simple_loss=0.2888, pruned_loss=0.08511, over 3925056.15 frames. ], batch size: 213, lr: 1.43e-02, grad_scale: 16.0
+2023-03-28 02:24:31,371 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3079, 2.8505, 4.3341, 3.6912, 3.9975, 4.2565, 4.2541, 4.0561],
+       device='cuda:3'), covar=tensor([0.0145, 0.0633, 0.0090, 0.0794, 0.0114, 0.0175, 0.0113, 0.0113],
+       device='cuda:3'), in_proj_covar=tensor([0.0076, 0.0088, 0.0069, 0.0140, 0.0062, 0.0076, 0.0072, 0.0063],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0004, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 02:25:24,915 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:25:55,152 INFO [train.py:892] (3/4) Epoch 11, batch 1050, loss[loss=0.2339, simple_loss=0.2988, pruned_loss=0.08448, over 19700.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.2903, pruned_loss=0.08577, over 3929954.51 frames. ], batch size: 48, lr: 1.43e-02, grad_scale: 16.0
+2023-03-28 02:26:15,114 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.901e+02 5.121e+02 6.164e+02 7.352e+02 1.446e+03, threshold=1.233e+03, percent-clipped=2.0
+2023-03-28 02:26:35,403 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:26:50,135 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9303, 4.9601, 5.3488, 5.1183, 4.9952, 4.6509, 4.9908, 4.8827],
+       device='cuda:3'), covar=tensor([0.1175, 0.1225, 0.0874, 0.0969, 0.0807, 0.0996, 0.1968, 0.1837],
+       device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0240, 0.0299, 0.0231, 0.0228, 0.0216, 0.0285, 0.0319],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 02:27:13,355 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19634.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:27:49,149 INFO [train.py:892] (3/4) Epoch 11, batch 1100, loss[loss=0.2121, simple_loss=0.265, pruned_loss=0.07956, over 19756.00 frames. ], tot_loss[loss=0.2286, simple_loss=0.2885, pruned_loss=0.08436, over 3933648.71 frames. ], batch size: 213, lr: 1.43e-02, grad_scale: 16.0
+2023-03-28 02:28:10,374 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3271, 3.4135, 2.0573, 3.5987, 3.6881, 1.5894, 3.0074, 2.8113],
+       device='cuda:3'), covar=tensor([0.0741, 0.0803, 0.2673, 0.0644, 0.0484, 0.2903, 0.1067, 0.0700],
+       device='cuda:3'), in_proj_covar=tensor([0.0190, 0.0212, 0.0210, 0.0197, 0.0155, 0.0196, 0.0219, 0.0155],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 02:28:53,897 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:28:57,809 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.0372, 2.1432, 2.3499, 2.1503, 2.0316, 2.1862, 1.9560, 2.1858],
+       device='cuda:3'), covar=tensor([0.0236, 0.0284, 0.0231, 0.0262, 0.0281, 0.0244, 0.0447, 0.0276],
+       device='cuda:3'), in_proj_covar=tensor([0.0046, 0.0045, 0.0048, 0.0041, 0.0051, 0.0048, 0.0063, 0.0044],
+       device='cuda:3'), out_proj_covar=tensor([1.0333e-04, 1.0117e-04, 1.0628e-04, 9.1856e-05, 1.1468e-04, 1.0841e-04,
+        1.3847e-04, 9.9892e-05], device='cuda:3')
+2023-03-28 02:29:11,201 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3204, 3.6264, 3.7050, 4.6511, 2.8207, 3.2657, 2.8341, 2.4973],
+       device='cuda:3'), covar=tensor([0.0439, 0.2190, 0.0849, 0.0204, 0.2081, 0.0849, 0.1189, 0.1901],
+       device='cuda:3'), in_proj_covar=tensor([0.0196, 0.0330, 0.0227, 0.0146, 0.0238, 0.0173, 0.0197, 0.0208],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 02:29:42,838 INFO [train.py:892] (3/4) Epoch 11, batch 1150, loss[loss=0.2205, simple_loss=0.2766, pruned_loss=0.08215, over 19826.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.2878, pruned_loss=0.08417, over 3938458.47 frames. ], batch size: 147, lr: 1.43e-02, grad_scale: 16.0
+2023-03-28 02:30:05,101 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.048e+02 4.779e+02 5.602e+02 6.982e+02 1.239e+03, threshold=1.120e+03, percent-clipped=1.0
+2023-03-28 02:30:58,237 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0
+2023-03-28 02:31:36,110 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:31:37,393 INFO [train.py:892] (3/4) Epoch 11, batch 1200, loss[loss=0.2139, simple_loss=0.2747, pruned_loss=0.07658, over 19850.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.288, pruned_loss=0.08433, over 3941272.48 frames. ], batch size: 81, lr: 1.42e-02, grad_scale: 16.0
+2023-03-28 02:32:26,957 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.8849, 1.8920, 2.1388, 1.9758, 1.9426, 2.0453, 1.8072, 2.0322],
+       device='cuda:3'), covar=tensor([0.0220, 0.0263, 0.0178, 0.0238, 0.0264, 0.0205, 0.0328, 0.0277],
+       device='cuda:3'), in_proj_covar=tensor([0.0046, 0.0046, 0.0047, 0.0041, 0.0051, 0.0048, 0.0063, 0.0044],
+       device='cuda:3'), out_proj_covar=tensor([1.0289e-04, 1.0127e-04, 1.0566e-04, 9.2708e-05, 1.1487e-04, 1.0887e-04,
+        1.3909e-04, 1.0009e-04], device='cuda:3')
+2023-03-28 02:32:55,551 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19785.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 02:33:30,045 INFO [train.py:892] (3/4) Epoch 11, batch 1250, loss[loss=0.2139, simple_loss=0.2603, pruned_loss=0.08372, over 19800.00 frames. ], tot_loss[loss=0.2275, simple_loss=0.287, pruned_loss=0.08404, over 3943902.32 frames. ], batch size: 149, lr: 1.42e-02, grad_scale: 16.0
+2023-03-28 02:33:49,269 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.235e+02 4.789e+02 5.822e+02 7.202e+02 1.423e+03, threshold=1.164e+03, percent-clipped=5.0
+2023-03-28 02:34:44,025 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19833.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 02:35:24,445 INFO [train.py:892] (3/4) Epoch 11, batch 1300, loss[loss=0.2253, simple_loss=0.2839, pruned_loss=0.0834, over 19713.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.2864, pruned_loss=0.08419, over 3946568.07 frames. ], batch size: 78, lr: 1.42e-02, grad_scale: 16.0
+2023-03-28 02:37:16,640 INFO [train.py:892] (3/4) Epoch 11, batch 1350, loss[loss=0.2136, simple_loss=0.2699, pruned_loss=0.07868, over 19875.00 frames. ], tot_loss[loss=0.2279, simple_loss=0.2869, pruned_loss=0.08449, over 3946879.64 frames. ], batch size: 159, lr: 1.42e-02, grad_scale: 16.0
+2023-03-28 02:37:38,388 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.602e+02 5.196e+02 5.941e+02 7.329e+02 1.053e+03, threshold=1.188e+03, percent-clipped=0.0
+2023-03-28 02:39:04,626 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6631, 4.8363, 5.2326, 4.7957, 4.3056, 5.0410, 4.8497, 5.4231],
+       device='cuda:3'), covar=tensor([0.0917, 0.0261, 0.0364, 0.0308, 0.0640, 0.0344, 0.0299, 0.0250],
+       device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0185, 0.0180, 0.0192, 0.0180, 0.0189, 0.0181, 0.0173],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 02:39:10,158 INFO [train.py:892] (3/4) Epoch 11, batch 1400, loss[loss=0.2237, simple_loss=0.2951, pruned_loss=0.07615, over 19781.00 frames. ], tot_loss[loss=0.2277, simple_loss=0.2867, pruned_loss=0.08439, over 3947176.66 frames. ], batch size: 53, lr: 1.42e-02, grad_scale: 16.0
+2023-03-28 02:40:05,784 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:41:09,345 INFO [train.py:892] (3/4) Epoch 11, batch 1450, loss[loss=0.2422, simple_loss=0.2982, pruned_loss=0.09313, over 19761.00 frames. ], tot_loss[loss=0.226, simple_loss=0.2856, pruned_loss=0.08318, over 3948584.81 frames. ], batch size: 273, lr: 1.42e-02, grad_scale: 16.0
+2023-03-28 02:41:30,121 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.362e+02 4.773e+02 5.638e+02 7.039e+02 1.448e+03, threshold=1.128e+03, percent-clipped=1.0
+2023-03-28 02:41:54,659 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:42:38,405 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-28 02:43:00,680 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:43:01,874 INFO [train.py:892] (3/4) Epoch 11, batch 1500, loss[loss=0.1963, simple_loss=0.2636, pruned_loss=0.06454, over 19669.00 frames. ], tot_loss[loss=0.2257, simple_loss=0.2853, pruned_loss=0.08304, over 3948599.26 frames. ], batch size: 43, lr: 1.41e-02, grad_scale: 16.0
+2023-03-28 02:44:11,223 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:44:18,239 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3117, 4.1992, 4.6707, 4.5237, 4.5607, 3.8436, 4.3132, 4.2711],
+       device='cuda:3'), covar=tensor([0.1316, 0.1428, 0.1003, 0.1118, 0.0898, 0.1314, 0.2279, 0.2142],
+       device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0248, 0.0303, 0.0234, 0.0231, 0.0217, 0.0289, 0.0326],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 02:44:30,167 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:44:50,401 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:44:55,651 INFO [train.py:892] (3/4) Epoch 11, batch 1550, loss[loss=0.2259, simple_loss=0.3016, pruned_loss=0.07514, over 19855.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.286, pruned_loss=0.08311, over 3947390.48 frames. ], batch size: 56, lr: 1.41e-02, grad_scale: 16.0
+2023-03-28 02:45:16,072 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.591e+02 5.026e+02 5.881e+02 6.796e+02 1.345e+03, threshold=1.176e+03, percent-clipped=3.0
+2023-03-28 02:45:21,252 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20112.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:46:41,420 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:46:42,455 INFO [train.py:892] (3/4) Epoch 11, batch 1600, loss[loss=0.1804, simple_loss=0.2425, pruned_loss=0.05912, over 19685.00 frames. ], tot_loss[loss=0.2261, simple_loss=0.2862, pruned_loss=0.08304, over 3947752.73 frames. ], batch size: 45, lr: 1.41e-02, grad_scale: 16.0
+2023-03-28 02:47:25,042 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0
+2023-03-28 02:47:33,514 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20173.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:47:41,552 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4487, 2.3576, 2.5373, 2.4338, 2.2292, 2.2124, 2.2823, 2.5701],
+       device='cuda:3'), covar=tensor([0.0198, 0.0302, 0.0304, 0.0212, 0.0304, 0.0324, 0.0356, 0.0346],
+       device='cuda:3'), in_proj_covar=tensor([0.0046, 0.0047, 0.0048, 0.0042, 0.0052, 0.0048, 0.0065, 0.0044],
+       device='cuda:3'), out_proj_covar=tensor([1.0411e-04, 1.0548e-04, 1.0755e-04, 9.5329e-05, 1.1767e-04, 1.0846e-04,
+        1.4245e-04, 1.0099e-04], device='cuda:3')
+2023-03-28 02:48:38,428 INFO [train.py:892] (3/4) Epoch 11, batch 1650, loss[loss=0.3291, simple_loss=0.4064, pruned_loss=0.1259, over 17984.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.2863, pruned_loss=0.0835, over 3947400.80 frames. ], batch size: 633, lr: 1.41e-02, grad_scale: 16.0
+2023-03-28 02:48:45,237 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:48:56,034 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.919e+02 4.977e+02 5.709e+02 7.210e+02 1.480e+03, threshold=1.142e+03, percent-clipped=1.0
+2023-03-28 02:50:17,231 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.1215, 2.4449, 2.0443, 1.5961, 2.2032, 2.4584, 2.4587, 2.3132],
+       device='cuda:3'), covar=tensor([0.0205, 0.0206, 0.0243, 0.0560, 0.0324, 0.0184, 0.0157, 0.0193],
+       device='cuda:3'), in_proj_covar=tensor([0.0062, 0.0057, 0.0066, 0.0076, 0.0077, 0.0052, 0.0048, 0.0050],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-28 02:50:19,644 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0
+2023-03-28 02:50:27,686 INFO [train.py:892] (3/4) Epoch 11, batch 1700, loss[loss=0.2474, simple_loss=0.2984, pruned_loss=0.09818, over 19750.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.2879, pruned_loss=0.08447, over 3947607.40 frames. ], batch size: 250, lr: 1.41e-02, grad_scale: 16.0
+2023-03-28 02:50:58,810 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:51:04,937 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.7248, 1.7197, 1.9948, 1.8598, 1.7712, 1.9090, 1.7950, 1.9954],
+       device='cuda:3'), covar=tensor([0.0244, 0.0245, 0.0185, 0.0182, 0.0314, 0.0202, 0.0320, 0.0240],
+       device='cuda:3'), in_proj_covar=tensor([0.0046, 0.0047, 0.0048, 0.0041, 0.0052, 0.0048, 0.0064, 0.0044],
+       device='cuda:3'), out_proj_covar=tensor([1.0420e-04, 1.0487e-04, 1.0663e-04, 9.3162e-05, 1.1708e-04, 1.0751e-04,
+        1.4030e-04, 9.9895e-05], device='cuda:3')
+2023-03-28 02:51:19,819 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20274.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:52:12,821 INFO [train.py:892] (3/4) Epoch 11, batch 1750, loss[loss=0.2148, simple_loss=0.2769, pruned_loss=0.07632, over 19585.00 frames. ], tot_loss[loss=0.2266, simple_loss=0.2863, pruned_loss=0.08348, over 3948375.51 frames. ], batch size: 42, lr: 1.41e-02, grad_scale: 16.0
+2023-03-28 02:52:29,859 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.218e+02 5.025e+02 5.710e+02 7.275e+02 1.664e+03, threshold=1.142e+03, percent-clipped=3.0
+2023-03-28 02:52:53,449 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:53:48,075 INFO [train.py:892] (3/4) Epoch 11, batch 1800, loss[loss=0.2938, simple_loss=0.3455, pruned_loss=0.121, over 19781.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.2859, pruned_loss=0.08296, over 3949318.75 frames. ], batch size: 247, lr: 1.40e-02, grad_scale: 16.0
+2023-03-28 02:54:35,770 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:54:57,988 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:55:18,953 INFO [train.py:892] (3/4) Epoch 11, batch 1850, loss[loss=0.2281, simple_loss=0.2925, pruned_loss=0.08179, over 19579.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.2864, pruned_loss=0.0827, over 3949348.20 frames. ], batch size: 53, lr: 1.40e-02, grad_scale: 16.0
+2023-03-28 02:56:26,120 INFO [train.py:892] (3/4) Epoch 12, batch 0, loss[loss=0.2259, simple_loss=0.3016, pruned_loss=0.07511, over 19847.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.3016, pruned_loss=0.07511, over 19847.00 frames. ], batch size: 58, lr: 1.34e-02, grad_scale: 16.0
+2023-03-28 02:56:26,121 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-28 02:56:56,308 INFO [train.py:926] (3/4) Epoch 12, validation: loss=0.1761, simple_loss=0.2565, pruned_loss=0.0478, over 2883724.00 frames. 
+2023-03-28 02:56:56,309 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-28 02:57:06,912 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.354e+02 5.231e+02 6.194e+02 7.302e+02 1.843e+03, threshold=1.239e+03, percent-clipped=4.0
+2023-03-28 02:57:28,087 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:58:29,440 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:58:39,860 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:58:54,029 INFO [train.py:892] (3/4) Epoch 12, batch 50, loss[loss=0.2268, simple_loss=0.2833, pruned_loss=0.08508, over 19737.00 frames. ], tot_loss[loss=0.2287, simple_loss=0.2872, pruned_loss=0.08513, over 890038.19 frames. ], batch size: 77, lr: 1.34e-02, grad_scale: 16.0
+2023-03-28 02:59:16,031 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5649, 4.6244, 5.0979, 4.6355, 4.1840, 4.9032, 4.7522, 5.1694],
+       device='cuda:3'), covar=tensor([0.0982, 0.0320, 0.0302, 0.0300, 0.0686, 0.0373, 0.0378, 0.0292],
+       device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0185, 0.0179, 0.0188, 0.0178, 0.0189, 0.0182, 0.0171],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 02:59:19,994 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20468.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 02:59:47,784 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:00:46,540 INFO [train.py:892] (3/4) Epoch 12, batch 100, loss[loss=0.2219, simple_loss=0.2845, pruned_loss=0.0796, over 19749.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.2863, pruned_loss=0.08378, over 1568038.26 frames. ], batch size: 226, lr: 1.34e-02, grad_scale: 16.0
+2023-03-28 03:00:55,044 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.431e+02 4.892e+02 6.215e+02 7.181e+02 1.654e+03, threshold=1.243e+03, percent-clipped=3.0
+2023-03-28 03:02:21,866 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8095, 2.8381, 3.2511, 2.8270, 2.9712, 3.1769, 3.0286, 3.2998],
+       device='cuda:3'), covar=tensor([0.1370, 0.0455, 0.0454, 0.0545, 0.1153, 0.0699, 0.0469, 0.0446],
+       device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0187, 0.0181, 0.0191, 0.0179, 0.0193, 0.0185, 0.0173],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 03:02:39,175 INFO [train.py:892] (3/4) Epoch 12, batch 150, loss[loss=0.2249, simple_loss=0.2945, pruned_loss=0.07761, over 19931.00 frames. ], tot_loss[loss=0.2271, simple_loss=0.2867, pruned_loss=0.08377, over 2095434.45 frames. ], batch size: 49, lr: 1.34e-02, grad_scale: 16.0
+2023-03-28 03:02:51,614 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20560.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:03:31,324 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4289, 3.4994, 4.9063, 3.6651, 4.1948, 4.3253, 2.5108, 2.7989],
+       device='cuda:3'), covar=tensor([0.0642, 0.2436, 0.0353, 0.0676, 0.1248, 0.0714, 0.1734, 0.1998],
+       device='cuda:3'), in_proj_covar=tensor([0.0306, 0.0340, 0.0270, 0.0224, 0.0330, 0.0266, 0.0290, 0.0265],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 03:03:56,816 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:04:11,949 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0
+2023-03-28 03:04:38,605 INFO [train.py:892] (3/4) Epoch 12, batch 200, loss[loss=0.183, simple_loss=0.2463, pruned_loss=0.0599, over 19785.00 frames. ], tot_loss[loss=0.2254, simple_loss=0.2855, pruned_loss=0.08268, over 2507614.54 frames. ], batch size: 94, lr: 1.34e-02, grad_scale: 16.0
+2023-03-28 03:04:47,253 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.285e+02 4.571e+02 5.228e+02 6.546e+02 1.569e+03, threshold=1.046e+03, percent-clipped=1.0
+2023-03-28 03:06:25,778 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:06:39,392 INFO [train.py:892] (3/4) Epoch 12, batch 250, loss[loss=0.1838, simple_loss=0.2519, pruned_loss=0.05781, over 19920.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.2843, pruned_loss=0.08207, over 2827690.07 frames. ], batch size: 45, lr: 1.34e-02, grad_scale: 16.0
+2023-03-28 03:07:25,021 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:08:16,365 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7236, 3.8182, 3.9567, 4.8267, 2.9063, 3.4415, 3.1052, 2.6836],
+       device='cuda:3'), covar=tensor([0.0399, 0.2200, 0.0878, 0.0213, 0.2289, 0.0783, 0.1101, 0.2014],
+       device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0329, 0.0222, 0.0150, 0.0237, 0.0173, 0.0193, 0.0207],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 03:08:29,929 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.3637, 1.7750, 2.0267, 2.7772, 3.0091, 3.0715, 2.9800, 3.1677],
+       device='cuda:3'), covar=tensor([0.0876, 0.1851, 0.1401, 0.0487, 0.0412, 0.0255, 0.0305, 0.0286],
+       device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0171, 0.0158, 0.0127, 0.0111, 0.0104, 0.0098, 0.0095],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 03:08:29,992 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.8993, 2.0870, 1.8119, 1.2209, 1.9960, 2.0371, 2.0094, 2.0397],
+       device='cuda:3'), covar=tensor([0.0223, 0.0180, 0.0236, 0.0513, 0.0330, 0.0189, 0.0170, 0.0174],
+       device='cuda:3'), in_proj_covar=tensor([0.0062, 0.0058, 0.0066, 0.0076, 0.0077, 0.0052, 0.0048, 0.0051],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-28 03:08:30,767 INFO [train.py:892] (3/4) Epoch 12, batch 300, loss[loss=0.2402, simple_loss=0.2962, pruned_loss=0.09206, over 19789.00 frames. ], tot_loss[loss=0.2264, simple_loss=0.2857, pruned_loss=0.0835, over 3075587.16 frames. ], batch size: 247, lr: 1.33e-02, grad_scale: 16.0
+2023-03-28 03:08:41,176 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.455e+02 4.985e+02 6.229e+02 8.000e+02 1.241e+03, threshold=1.246e+03, percent-clipped=1.0
+2023-03-28 03:09:12,872 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:09:57,988 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:10:01,771 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:10:24,863 INFO [train.py:892] (3/4) Epoch 12, batch 350, loss[loss=0.2112, simple_loss=0.2692, pruned_loss=0.07658, over 19793.00 frames. ], tot_loss[loss=0.2281, simple_loss=0.2871, pruned_loss=0.08456, over 3267306.62 frames. ], batch size: 174, lr: 1.33e-02, grad_scale: 16.0
+2023-03-28 03:10:52,545 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:11:06,484 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:11:12,775 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.39 vs. limit=5.0
+2023-03-28 03:11:52,647 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:12:20,219 INFO [train.py:892] (3/4) Epoch 12, batch 400, loss[loss=0.2041, simple_loss=0.2719, pruned_loss=0.06816, over 19924.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.2878, pruned_loss=0.08488, over 3419040.02 frames. ], batch size: 51, lr: 1.33e-02, grad_scale: 16.0
+2023-03-28 03:12:28,475 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.077e+02 4.871e+02 5.546e+02 6.425e+02 9.713e+02, threshold=1.109e+03, percent-clipped=0.0
+2023-03-28 03:12:42,787 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:13:33,716 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-28 03:14:10,008 INFO [train.py:892] (3/4) Epoch 12, batch 450, loss[loss=0.191, simple_loss=0.2576, pruned_loss=0.06217, over 19718.00 frames. ], tot_loss[loss=0.2274, simple_loss=0.2871, pruned_loss=0.08383, over 3535967.01 frames. ], batch size: 104, lr: 1.33e-02, grad_scale: 16.0
+2023-03-28 03:14:21,446 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20860.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:14:26,020 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.0509, 2.1203, 2.0827, 2.0813, 1.9796, 2.1152, 2.1290, 2.2796],
+       device='cuda:3'), covar=tensor([0.0168, 0.0225, 0.0237, 0.0184, 0.0254, 0.0211, 0.0296, 0.0206],
+       device='cuda:3'), in_proj_covar=tensor([0.0047, 0.0048, 0.0048, 0.0043, 0.0053, 0.0049, 0.0064, 0.0044],
+       device='cuda:3'), out_proj_covar=tensor([1.0595e-04, 1.0668e-04, 1.0846e-04, 9.6875e-05, 1.1901e-04, 1.1003e-04,
+        1.4071e-04, 1.0102e-04], device='cuda:3')
+2023-03-28 03:14:28,114 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.2540, 2.6726, 2.1923, 1.6656, 2.4015, 2.5528, 2.6009, 2.5755],
+       device='cuda:3'), covar=tensor([0.0212, 0.0211, 0.0222, 0.0550, 0.0274, 0.0208, 0.0146, 0.0155],
+       device='cuda:3'), in_proj_covar=tensor([0.0063, 0.0059, 0.0067, 0.0077, 0.0078, 0.0052, 0.0048, 0.0051],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-28 03:15:33,971 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0
+2023-03-28 03:16:03,420 INFO [train.py:892] (3/4) Epoch 12, batch 500, loss[loss=0.2149, simple_loss=0.2732, pruned_loss=0.0783, over 19800.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.2863, pruned_loss=0.0835, over 3627834.20 frames. ], batch size: 150, lr: 1.33e-02, grad_scale: 16.0
+2023-03-28 03:16:08,893 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:16:11,897 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.335e+02 4.795e+02 5.889e+02 7.056e+02 1.371e+03, threshold=1.178e+03, percent-clipped=3.0
+2023-03-28 03:16:24,447 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0953, 5.4603, 5.7238, 5.5355, 5.3129, 5.2400, 5.3599, 5.2437],
+       device='cuda:3'), covar=tensor([0.1284, 0.0966, 0.0797, 0.0961, 0.0637, 0.0771, 0.1732, 0.1664],
+       device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0244, 0.0299, 0.0231, 0.0221, 0.0217, 0.0291, 0.0326],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 03:17:30,610 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:17:57,152 INFO [train.py:892] (3/4) Epoch 12, batch 550, loss[loss=0.1977, simple_loss=0.2627, pruned_loss=0.06635, over 19743.00 frames. ], tot_loss[loss=0.2268, simple_loss=0.2864, pruned_loss=0.08357, over 3697630.86 frames. ], batch size: 95, lr: 1.33e-02, grad_scale: 16.0
+2023-03-28 03:19:54,087 INFO [train.py:892] (3/4) Epoch 12, batch 600, loss[loss=0.2064, simple_loss=0.2632, pruned_loss=0.07481, over 19818.00 frames. ], tot_loss[loss=0.2267, simple_loss=0.2864, pruned_loss=0.08354, over 3754570.16 frames. ], batch size: 121, lr: 1.32e-02, grad_scale: 16.0
+2023-03-28 03:20:01,787 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.833e+02 4.713e+02 5.422e+02 6.814e+02 1.308e+03, threshold=1.084e+03, percent-clipped=1.0
+2023-03-28 03:20:39,858 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-28 03:21:20,865 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:21:37,405 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:21:45,735 INFO [train.py:892] (3/4) Epoch 12, batch 650, loss[loss=0.2141, simple_loss=0.2821, pruned_loss=0.07307, over 19678.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.2858, pruned_loss=0.08297, over 3797386.11 frames. ], batch size: 52, lr: 1.32e-02, grad_scale: 16.0
+2023-03-28 03:22:28,050 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:23:09,947 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:23:10,632 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-28 03:23:39,589 INFO [train.py:892] (3/4) Epoch 12, batch 700, loss[loss=0.2273, simple_loss=0.2894, pruned_loss=0.0826, over 19706.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.285, pruned_loss=0.08221, over 3833085.67 frames. ], batch size: 101, lr: 1.32e-02, grad_scale: 16.0
+2023-03-28 03:23:49,612 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.058e+02 5.405e+02 6.727e+02 8.130e+02 1.465e+03, threshold=1.345e+03, percent-clipped=5.0
+2023-03-28 03:23:56,939 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:24:07,609 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:24:18,045 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:24:56,841 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7680, 3.0948, 2.4860, 2.0661, 2.5642, 3.1906, 2.8810, 3.0570],
+       device='cuda:3'), covar=tensor([0.0195, 0.0287, 0.0243, 0.0495, 0.0354, 0.0169, 0.0172, 0.0158],
+       device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0060, 0.0068, 0.0079, 0.0080, 0.0054, 0.0050, 0.0054],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-28 03:24:58,929 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7551, 2.2144, 2.6618, 3.0911, 3.5197, 3.6615, 3.6488, 3.8171],
+       device='cuda:3'), covar=tensor([0.0803, 0.1575, 0.1067, 0.0479, 0.0346, 0.0201, 0.0236, 0.0227],
+       device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0166, 0.0154, 0.0128, 0.0107, 0.0102, 0.0095, 0.0093],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 03:25:33,698 INFO [train.py:892] (3/4) Epoch 12, batch 750, loss[loss=0.2334, simple_loss=0.2982, pruned_loss=0.08426, over 19760.00 frames. ], tot_loss[loss=0.223, simple_loss=0.2837, pruned_loss=0.08118, over 3860039.51 frames. ], batch size: 256, lr: 1.32e-02, grad_scale: 16.0
+2023-03-28 03:26:29,233 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:27:28,544 INFO [train.py:892] (3/4) Epoch 12, batch 800, loss[loss=0.2153, simple_loss=0.2688, pruned_loss=0.08087, over 19764.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.2829, pruned_loss=0.08091, over 3881315.21 frames. ], batch size: 213, lr: 1.32e-02, grad_scale: 16.0
+2023-03-28 03:27:37,191 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.248e+02 4.727e+02 5.705e+02 6.809e+02 1.528e+03, threshold=1.141e+03, percent-clipped=1.0
+2023-03-28 03:28:05,043 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0
+2023-03-28 03:28:55,979 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:29:23,791 INFO [train.py:892] (3/4) Epoch 12, batch 850, loss[loss=0.2044, simple_loss=0.2578, pruned_loss=0.07545, over 19829.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.2822, pruned_loss=0.08048, over 3897607.18 frames. ], batch size: 146, lr: 1.32e-02, grad_scale: 16.0
+2023-03-28 03:30:18,130 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0
+2023-03-28 03:30:45,257 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21292.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:31:14,739 INFO [train.py:892] (3/4) Epoch 12, batch 900, loss[loss=0.1928, simple_loss=0.2593, pruned_loss=0.0631, over 19704.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.2821, pruned_loss=0.08062, over 3909614.04 frames. ], batch size: 101, lr: 1.32e-02, grad_scale: 32.0
+2023-03-28 03:31:22,616 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.874e+02 4.590e+02 5.772e+02 7.040e+02 1.378e+03, threshold=1.154e+03, percent-clipped=2.0
+2023-03-28 03:31:27,818 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:32:07,214 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21328.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:33:10,587 INFO [train.py:892] (3/4) Epoch 12, batch 950, loss[loss=0.2007, simple_loss=0.2699, pruned_loss=0.0657, over 19735.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.2824, pruned_loss=0.08094, over 3918538.87 frames. ], batch size: 71, lr: 1.31e-02, grad_scale: 16.0
+2023-03-28 03:33:48,264 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:34:06,103 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8292, 3.7780, 2.3954, 4.0787, 4.1490, 1.8020, 3.4795, 3.2648],
+       device='cuda:3'), covar=tensor([0.0582, 0.0808, 0.2372, 0.0644, 0.0376, 0.2992, 0.1011, 0.0563],
+       device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0214, 0.0208, 0.0202, 0.0161, 0.0195, 0.0220, 0.0158],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 03:34:26,617 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:34:28,807 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7500, 3.7113, 2.3474, 4.0598, 4.1728, 1.7463, 3.3347, 3.1931],
+       device='cuda:3'), covar=tensor([0.0708, 0.0928, 0.2618, 0.0664, 0.0359, 0.3498, 0.1105, 0.0715],
+       device='cuda:3'), in_proj_covar=tensor([0.0193, 0.0215, 0.0208, 0.0203, 0.0162, 0.0195, 0.0221, 0.0158],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 03:35:02,719 INFO [train.py:892] (3/4) Epoch 12, batch 1000, loss[loss=0.2142, simple_loss=0.2873, pruned_loss=0.07054, over 19791.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.2827, pruned_loss=0.08081, over 3926640.68 frames. ], batch size: 79, lr: 1.31e-02, grad_scale: 16.0
+2023-03-28 03:35:08,965 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:35:14,100 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.713e+02 5.169e+02 6.246e+02 8.325e+02 1.505e+03, threshold=1.249e+03, percent-clipped=4.0
+2023-03-28 03:35:17,193 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.6522, 6.0452, 6.0647, 5.9486, 5.7676, 6.0464, 5.3376, 5.5029],
+       device='cuda:3'), covar=tensor([0.0330, 0.0335, 0.0521, 0.0362, 0.0543, 0.0558, 0.0662, 0.0875],
+       device='cuda:3'), in_proj_covar=tensor([0.0197, 0.0200, 0.0237, 0.0201, 0.0191, 0.0186, 0.0214, 0.0251],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 03:35:51,940 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4313, 4.4548, 4.8826, 4.4598, 4.1045, 4.7058, 4.5583, 4.9861],
+       device='cuda:3'), covar=tensor([0.0957, 0.0341, 0.0348, 0.0359, 0.0763, 0.0475, 0.0483, 0.0295],
+       device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0188, 0.0185, 0.0192, 0.0181, 0.0192, 0.0183, 0.0174],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 03:36:53,864 INFO [train.py:892] (3/4) Epoch 12, batch 1050, loss[loss=0.21, simple_loss=0.2697, pruned_loss=0.0751, over 19781.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.2826, pruned_loss=0.08076, over 3931556.27 frames. ], batch size: 131, lr: 1.31e-02, grad_scale: 16.0
+2023-03-28 03:37:06,604 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0
+2023-03-28 03:37:32,708 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21474.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:37:59,240 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0
+2023-03-28 03:38:11,987 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-28 03:38:34,357 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7220, 2.3657, 2.9259, 3.2165, 3.7894, 4.1301, 4.0814, 4.2848],
+       device='cuda:3'), covar=tensor([0.1008, 0.1897, 0.1277, 0.0607, 0.0322, 0.0177, 0.0259, 0.0322],
+       device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0170, 0.0160, 0.0133, 0.0112, 0.0106, 0.0098, 0.0096],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 03:38:43,128 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2364, 2.2151, 3.4732, 3.0314, 3.3372, 3.5264, 3.4355, 3.4433],
+       device='cuda:3'), covar=tensor([0.0241, 0.0735, 0.0092, 0.0528, 0.0102, 0.0193, 0.0127, 0.0118],
+       device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0091, 0.0072, 0.0143, 0.0066, 0.0081, 0.0074, 0.0065],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0004, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 03:38:44,179 INFO [train.py:892] (3/4) Epoch 12, batch 1100, loss[loss=0.224, simple_loss=0.2855, pruned_loss=0.08118, over 19750.00 frames. ], tot_loss[loss=0.2217, simple_loss=0.2824, pruned_loss=0.08053, over 3936394.18 frames. ], batch size: 250, lr: 1.31e-02, grad_scale: 16.0
+2023-03-28 03:38:56,799 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.523e+02 4.626e+02 5.286e+02 6.822e+02 1.206e+03, threshold=1.057e+03, percent-clipped=0.0
+2023-03-28 03:40:37,763 INFO [train.py:892] (3/4) Epoch 12, batch 1150, loss[loss=0.2017, simple_loss=0.2641, pruned_loss=0.06962, over 19801.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.2843, pruned_loss=0.08172, over 3937946.85 frames. ], batch size: 107, lr: 1.31e-02, grad_scale: 16.0
+2023-03-28 03:40:55,821 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8498, 2.2941, 2.7001, 2.5766, 2.4755, 2.8072, 1.8279, 1.9904],
+       device='cuda:3'), covar=tensor([0.0728, 0.1392, 0.0533, 0.0634, 0.1143, 0.0740, 0.1500, 0.1561],
+       device='cuda:3'), in_proj_covar=tensor([0.0306, 0.0336, 0.0271, 0.0224, 0.0329, 0.0271, 0.0289, 0.0266],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 03:41:37,311 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9415, 2.8706, 3.1553, 2.5034, 3.1525, 2.6482, 2.8827, 3.1869],
+       device='cuda:3'), covar=tensor([0.0510, 0.0396, 0.0550, 0.0717, 0.0408, 0.0392, 0.0401, 0.0263],
+       device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0063, 0.0065, 0.0094, 0.0061, 0.0057, 0.0056, 0.0051],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 03:42:02,493 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8031, 3.8761, 4.1945, 5.0470, 3.0444, 3.5191, 3.1400, 2.8516],
+       device='cuda:3'), covar=tensor([0.0361, 0.2297, 0.0646, 0.0204, 0.2019, 0.0770, 0.1140, 0.1662],
+       device='cuda:3'), in_proj_covar=tensor([0.0198, 0.0328, 0.0226, 0.0153, 0.0235, 0.0173, 0.0195, 0.0205],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 03:42:06,835 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:42:32,677 INFO [train.py:892] (3/4) Epoch 12, batch 1200, loss[loss=0.189, simple_loss=0.2619, pruned_loss=0.0581, over 19875.00 frames. ], tot_loss[loss=0.226, simple_loss=0.2859, pruned_loss=0.08309, over 3938582.80 frames. ], batch size: 92, lr: 1.31e-02, grad_scale: 16.0
+2023-03-28 03:42:43,491 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.726e+02 4.687e+02 5.841e+02 7.196e+02 1.649e+03, threshold=1.168e+03, percent-clipped=4.0
+2023-03-28 03:42:48,465 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:43:53,340 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0661, 3.9276, 3.9114, 3.6957, 4.0387, 2.9172, 3.3744, 2.1157],
+       device='cuda:3'), covar=tensor([0.0199, 0.0178, 0.0128, 0.0168, 0.0133, 0.0866, 0.0677, 0.1341],
+       device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0117, 0.0095, 0.0112, 0.0100, 0.0119, 0.0129, 0.0114],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 03:44:28,814 INFO [train.py:892] (3/4) Epoch 12, batch 1250, loss[loss=0.2065, simple_loss=0.2549, pruned_loss=0.07904, over 19877.00 frames. ], tot_loss[loss=0.223, simple_loss=0.2832, pruned_loss=0.08133, over 3941166.19 frames. ], batch size: 77, lr: 1.31e-02, grad_scale: 16.0
+2023-03-28 03:44:29,857 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:44:54,430 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:45:08,162 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:45:32,211 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21684.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:46:16,802 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-28 03:46:19,253 INFO [train.py:892] (3/4) Epoch 12, batch 1300, loss[loss=0.2486, simple_loss=0.3106, pruned_loss=0.09328, over 19700.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.2826, pruned_loss=0.08104, over 3943067.97 frames. ], batch size: 48, lr: 1.30e-02, grad_scale: 16.0
+2023-03-28 03:46:24,493 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:46:31,047 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.976e+02 4.739e+02 5.550e+02 6.963e+02 1.418e+03, threshold=1.110e+03, percent-clipped=2.0
+2023-03-28 03:48:12,974 INFO [train.py:892] (3/4) Epoch 12, batch 1350, loss[loss=0.1964, simple_loss=0.2636, pruned_loss=0.06458, over 19905.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2813, pruned_loss=0.08021, over 3944761.61 frames. ], batch size: 116, lr: 1.30e-02, grad_scale: 16.0
+2023-03-28 03:48:13,704 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:48:27,776 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:48:52,599 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:50:04,166 INFO [train.py:892] (3/4) Epoch 12, batch 1400, loss[loss=0.1973, simple_loss=0.2625, pruned_loss=0.06602, over 19860.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2817, pruned_loss=0.07992, over 3944219.62 frames. ], batch size: 106, lr: 1.30e-02, grad_scale: 16.0
+2023-03-28 03:50:16,312 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.208e+02 4.274e+02 5.050e+02 6.143e+02 1.134e+03, threshold=1.010e+03, percent-clipped=1.0
+2023-03-28 03:50:40,268 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21822.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:50:47,574 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:50:58,582 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5895, 2.4883, 2.6685, 2.6030, 2.8998, 2.9411, 3.4244, 3.7269],
+       device='cuda:3'), covar=tensor([0.0536, 0.1634, 0.1592, 0.1894, 0.1661, 0.1294, 0.0485, 0.0516],
+       device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0214, 0.0235, 0.0231, 0.0258, 0.0224, 0.0179, 0.0183],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 03:51:37,269 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0439, 5.2663, 5.6264, 5.3821, 5.4058, 5.0275, 5.2719, 5.1546],
+       device='cuda:3'), covar=tensor([0.1379, 0.1106, 0.0908, 0.1125, 0.0655, 0.0898, 0.1961, 0.2035],
+       device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0255, 0.0309, 0.0241, 0.0226, 0.0226, 0.0295, 0.0332],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 03:51:44,490 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:51:46,413 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4462, 4.9095, 5.1019, 4.7972, 5.2744, 3.5845, 4.2139, 2.7097],
+       device='cuda:3'), covar=tensor([0.0143, 0.0142, 0.0119, 0.0153, 0.0123, 0.0635, 0.0778, 0.1347],
+       device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0118, 0.0097, 0.0114, 0.0101, 0.0120, 0.0131, 0.0116],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 03:51:55,131 INFO [train.py:892] (3/4) Epoch 12, batch 1450, loss[loss=0.2356, simple_loss=0.2979, pruned_loss=0.08666, over 19839.00 frames. ], tot_loss[loss=0.2211, simple_loss=0.2821, pruned_loss=0.08009, over 3945496.72 frames. ], batch size: 58, lr: 1.30e-02, grad_scale: 16.0
+2023-03-28 03:51:58,313 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3387, 4.4490, 4.7823, 4.5398, 4.6872, 4.1751, 4.4718, 4.2980],
+       device='cuda:3'), covar=tensor([0.1307, 0.1277, 0.0858, 0.1090, 0.0764, 0.1025, 0.1803, 0.2084],
+       device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0253, 0.0307, 0.0238, 0.0225, 0.0225, 0.0292, 0.0330],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 03:52:01,327 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0
+2023-03-28 03:53:47,770 INFO [train.py:892] (3/4) Epoch 12, batch 1500, loss[loss=0.1902, simple_loss=0.2487, pruned_loss=0.06582, over 19768.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2817, pruned_loss=0.08006, over 3947526.00 frames. ], batch size: 130, lr: 1.30e-02, grad_scale: 16.0
+2023-03-28 03:53:51,308 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8620, 2.8687, 2.9527, 2.3459, 3.0098, 2.4390, 2.7938, 2.9992],
+       device='cuda:3'), covar=tensor([0.0467, 0.0330, 0.0372, 0.0657, 0.0322, 0.0356, 0.0398, 0.0234],
+       device='cuda:3'), in_proj_covar=tensor([0.0058, 0.0063, 0.0064, 0.0092, 0.0060, 0.0057, 0.0055, 0.0050],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 03:53:58,746 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.501e+02 4.832e+02 5.724e+02 6.820e+02 1.583e+03, threshold=1.145e+03, percent-clipped=5.0
+2023-03-28 03:53:59,727 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:55:31,452 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:55:42,233 INFO [train.py:892] (3/4) Epoch 12, batch 1550, loss[loss=0.2245, simple_loss=0.2897, pruned_loss=0.07962, over 19779.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2813, pruned_loss=0.07968, over 3947230.72 frames. ], batch size: 247, lr: 1.30e-02, grad_scale: 16.0
+2023-03-28 03:55:50,327 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9390, 4.5989, 4.6849, 4.4337, 4.8587, 3.2233, 3.9268, 2.3909],
+       device='cuda:3'), covar=tensor([0.0192, 0.0159, 0.0135, 0.0156, 0.0113, 0.0798, 0.0894, 0.1430],
+       device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0117, 0.0096, 0.0113, 0.0100, 0.0120, 0.0129, 0.0113],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 03:56:10,744 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:56:12,934 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:56:15,180 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:56:15,308 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2416, 3.4165, 3.5748, 4.3901, 2.6313, 3.1730, 2.8484, 2.5331],
+       device='cuda:3'), covar=tensor([0.0384, 0.2206, 0.0843, 0.0213, 0.2118, 0.0759, 0.1062, 0.1733],
+       device='cuda:3'), in_proj_covar=tensor([0.0202, 0.0335, 0.0230, 0.0157, 0.0241, 0.0176, 0.0198, 0.0207],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 03:56:49,453 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21984.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:57:43,210 INFO [train.py:892] (3/4) Epoch 12, batch 1600, loss[loss=0.1956, simple_loss=0.2752, pruned_loss=0.05796, over 19806.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2808, pruned_loss=0.07898, over 3947764.53 frames. ], batch size: 68, lr: 1.30e-02, grad_scale: 16.0
+2023-03-28 03:57:44,462 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6445, 4.5924, 2.9264, 5.1978, 5.2339, 2.1051, 4.3630, 3.8959],
+       device='cuda:3'), covar=tensor([0.0514, 0.0651, 0.2162, 0.0368, 0.0292, 0.2889, 0.0869, 0.0590],
+       device='cuda:3'), in_proj_covar=tensor([0.0195, 0.0218, 0.0209, 0.0206, 0.0167, 0.0196, 0.0223, 0.0162],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 03:57:54,043 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.885e+02 4.412e+02 5.565e+02 6.803e+02 1.208e+03, threshold=1.113e+03, percent-clipped=1.0
+2023-03-28 03:58:08,642 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:58:40,465 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22031.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:58:41,953 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22032.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 03:59:37,630 INFO [train.py:892] (3/4) Epoch 12, batch 1650, loss[loss=0.2247, simple_loss=0.2946, pruned_loss=0.07737, over 19535.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2814, pruned_loss=0.07908, over 3948907.73 frames. ], batch size: 54, lr: 1.29e-02, grad_scale: 16.0
+2023-03-28 04:00:55,065 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.44 vs. limit=5.0
+2023-03-28 04:01:33,557 INFO [train.py:892] (3/4) Epoch 12, batch 1700, loss[loss=0.2473, simple_loss=0.3046, pruned_loss=0.09499, over 19690.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.2819, pruned_loss=0.07942, over 3948714.20 frames. ], batch size: 265, lr: 1.29e-02, grad_scale: 16.0
+2023-03-28 04:01:45,047 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.089e+02 4.459e+02 5.514e+02 6.707e+02 1.132e+03, threshold=1.103e+03, percent-clipped=1.0
+2023-03-28 04:02:01,926 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:02:26,632 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:02:40,500 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.7322, 1.9608, 1.8209, 1.1356, 1.8092, 1.9558, 1.7741, 1.9272],
+       device='cuda:3'), covar=tensor([0.0255, 0.0220, 0.0230, 0.0552, 0.0355, 0.0180, 0.0182, 0.0158],
+       device='cuda:3'), in_proj_covar=tensor([0.0064, 0.0060, 0.0068, 0.0078, 0.0080, 0.0054, 0.0051, 0.0053],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-28 04:03:13,832 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0
+2023-03-28 04:03:17,835 INFO [train.py:892] (3/4) Epoch 12, batch 1750, loss[loss=0.2342, simple_loss=0.294, pruned_loss=0.08719, over 19662.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2799, pruned_loss=0.07838, over 3950547.05 frames. ], batch size: 67, lr: 1.29e-02, grad_scale: 16.0
+2023-03-28 04:03:56,102 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:04:22,210 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22191.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:04:47,345 INFO [train.py:892] (3/4) Epoch 12, batch 1800, loss[loss=0.1762, simple_loss=0.2546, pruned_loss=0.04893, over 19718.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2818, pruned_loss=0.08003, over 3949288.06 frames. ], batch size: 54, lr: 1.29e-02, grad_scale: 16.0
+2023-03-28 04:04:47,836 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:04:55,761 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.946e+02 4.603e+02 5.856e+02 6.792e+02 1.518e+03, threshold=1.171e+03, percent-clipped=5.0
+2023-03-28 04:05:03,104 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:05:41,168 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:05:41,183 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22237.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:06:03,981 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:06:11,637 INFO [train.py:892] (3/4) Epoch 12, batch 1850, loss[loss=0.1972, simple_loss=0.2624, pruned_loss=0.06601, over 19824.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2821, pruned_loss=0.07909, over 3949535.02 frames. ], batch size: 57, lr: 1.29e-02, grad_scale: 16.0
+2023-03-28 04:07:10,139 INFO [train.py:892] (3/4) Epoch 13, batch 0, loss[loss=0.2179, simple_loss=0.2789, pruned_loss=0.07848, over 19848.00 frames. ], tot_loss[loss=0.2179, simple_loss=0.2789, pruned_loss=0.07848, over 19848.00 frames. ], batch size: 85, lr: 1.24e-02, grad_scale: 16.0
+2023-03-28 04:07:10,140 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-28 04:07:38,662 INFO [train.py:926] (3/4) Epoch 13, validation: loss=0.1745, simple_loss=0.2543, pruned_loss=0.04732, over 2883724.00 frames. 
+2023-03-28 04:07:38,663 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-28 04:07:56,633 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22269.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:07:59,061 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3792, 4.6888, 4.7232, 4.6455, 4.3726, 4.6300, 4.1829, 4.1747],
+       device='cuda:3'), covar=tensor([0.0467, 0.0490, 0.0566, 0.0461, 0.0726, 0.0616, 0.0712, 0.0926],
+       device='cuda:3'), in_proj_covar=tensor([0.0202, 0.0209, 0.0244, 0.0209, 0.0194, 0.0192, 0.0219, 0.0258],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 04:08:13,855 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:09:03,355 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22298.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:09:05,476 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:09:32,908 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.888e+02 4.546e+02 5.479e+02 6.348e+02 1.428e+03, threshold=1.096e+03, percent-clipped=1.0
+2023-03-28 04:09:32,930 INFO [train.py:892] (3/4) Epoch 13, batch 50, loss[loss=0.2208, simple_loss=0.283, pruned_loss=0.07931, over 19799.00 frames. ], tot_loss[loss=0.2117, simple_loss=0.2732, pruned_loss=0.07511, over 892567.23 frames. ], batch size: 231, lr: 1.24e-02, grad_scale: 16.0
+2023-03-28 04:09:46,025 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22317.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:10:02,956 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.3755, 1.6616, 2.0494, 2.6758, 2.9932, 3.0720, 3.1145, 3.1538],
+       device='cuda:3'), covar=tensor([0.0896, 0.2027, 0.1347, 0.0559, 0.0391, 0.0260, 0.0319, 0.0346],
+       device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0166, 0.0158, 0.0129, 0.0111, 0.0105, 0.0097, 0.0094],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 04:10:08,860 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:11:26,793 INFO [train.py:892] (3/4) Epoch 13, batch 100, loss[loss=0.2045, simple_loss=0.2845, pruned_loss=0.06221, over 19902.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2764, pruned_loss=0.07579, over 1571728.95 frames. ], batch size: 50, lr: 1.24e-02, grad_scale: 16.0
+2023-03-28 04:12:24,555 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:13:22,673 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.479e+02 4.716e+02 5.729e+02 6.753e+02 1.100e+03, threshold=1.146e+03, percent-clipped=1.0
+2023-03-28 04:13:22,702 INFO [train.py:892] (3/4) Epoch 13, batch 150, loss[loss=0.211, simple_loss=0.2714, pruned_loss=0.07533, over 19591.00 frames. ], tot_loss[loss=0.2144, simple_loss=0.2764, pruned_loss=0.07613, over 2098332.16 frames. ], batch size: 44, lr: 1.23e-02, grad_scale: 16.0
+2023-03-28 04:13:38,256 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4887, 3.6857, 3.6747, 4.8264, 2.7884, 3.4013, 3.0991, 2.6553],
+       device='cuda:3'), covar=tensor([0.0427, 0.2457, 0.0940, 0.0199, 0.2232, 0.0820, 0.1005, 0.1837],
+       device='cuda:3'), in_proj_covar=tensor([0.0204, 0.0336, 0.0231, 0.0158, 0.0241, 0.0178, 0.0199, 0.0210],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 04:13:40,128 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22419.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:14:41,874 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.5302, 5.8246, 6.1493, 5.8635, 5.7598, 5.3418, 5.7748, 5.6586],
+       device='cuda:3'), covar=tensor([0.1237, 0.0934, 0.0801, 0.0983, 0.0573, 0.0851, 0.1832, 0.1800],
+       device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0252, 0.0306, 0.0235, 0.0225, 0.0223, 0.0297, 0.0330],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 04:14:42,067 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22446.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:15:13,590 INFO [train.py:892] (3/4) Epoch 13, batch 200, loss[loss=0.2041, simple_loss=0.2769, pruned_loss=0.06565, over 19891.00 frames. ], tot_loss[loss=0.215, simple_loss=0.2774, pruned_loss=0.07625, over 2509404.35 frames. ], batch size: 52, lr: 1.23e-02, grad_scale: 16.0
+2023-03-28 04:15:26,972 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:16:09,402 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:16:56,163 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:17:04,715 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.872e+02 4.579e+02 5.437e+02 6.895e+02 1.289e+03, threshold=1.087e+03, percent-clipped=2.0
+2023-03-28 04:17:04,748 INFO [train.py:892] (3/4) Epoch 13, batch 250, loss[loss=0.2015, simple_loss=0.2709, pruned_loss=0.06603, over 19900.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2781, pruned_loss=0.07652, over 2829270.31 frames. ], batch size: 87, lr: 1.23e-02, grad_scale: 16.0
+2023-03-28 04:17:28,700 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0
+2023-03-28 04:17:40,412 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3460, 2.4039, 3.5956, 3.0837, 3.4989, 3.5424, 3.3751, 3.3485],
+       device='cuda:3'), covar=tensor([0.0284, 0.0771, 0.0105, 0.0605, 0.0120, 0.0236, 0.0183, 0.0152],
+       device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0088, 0.0071, 0.0140, 0.0065, 0.0079, 0.0074, 0.0064],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0004, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 04:17:55,847 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22532.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:18:06,770 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1574, 3.3377, 3.5386, 4.2367, 2.7992, 3.2654, 2.6405, 2.4772],
+       device='cuda:3'), covar=tensor([0.0407, 0.2161, 0.0910, 0.0202, 0.1963, 0.0695, 0.1235, 0.1744],
+       device='cuda:3'), in_proj_covar=tensor([0.0202, 0.0331, 0.0228, 0.0157, 0.0239, 0.0177, 0.0199, 0.0208],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 04:18:47,726 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:19:01,076 INFO [train.py:892] (3/4) Epoch 13, batch 300, loss[loss=0.217, simple_loss=0.2774, pruned_loss=0.07832, over 19834.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2785, pruned_loss=0.0764, over 3078639.47 frames. ], batch size: 146, lr: 1.23e-02, grad_scale: 16.0
+2023-03-28 04:19:22,900 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:20:14,540 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:20:52,587 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.162e+02 4.855e+02 5.580e+02 6.636e+02 1.112e+03, threshold=1.116e+03, percent-clipped=2.0
+2023-03-28 04:20:52,609 INFO [train.py:892] (3/4) Epoch 13, batch 350, loss[loss=0.245, simple_loss=0.3059, pruned_loss=0.09209, over 19791.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2785, pruned_loss=0.0766, over 3272930.96 frames. ], batch size: 224, lr: 1.23e-02, grad_scale: 16.0
+2023-03-28 04:21:24,638 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:21:52,184 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:22:35,964 INFO [train.py:892] (3/4) Epoch 13, batch 400, loss[loss=0.1915, simple_loss=0.261, pruned_loss=0.06102, over 19955.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2811, pruned_loss=0.07795, over 3421082.53 frames. ], batch size: 53, lr: 1.23e-02, grad_scale: 16.0
+2023-03-28 04:23:05,955 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22674.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:24:00,986 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22700.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 04:24:25,240 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.862e+02 4.701e+02 5.537e+02 6.698e+02 1.146e+03, threshold=1.107e+03, percent-clipped=1.0
+2023-03-28 04:24:25,264 INFO [train.py:892] (3/4) Epoch 13, batch 450, loss[loss=0.2443, simple_loss=0.2994, pruned_loss=0.09455, over 19789.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2805, pruned_loss=0.07753, over 3540234.38 frames. ], batch size: 241, lr: 1.23e-02, grad_scale: 16.0
+2023-03-28 04:24:59,703 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6232, 3.4541, 4.9910, 3.8015, 4.2744, 4.2069, 2.5956, 2.8296],
+       device='cuda:3'), covar=tensor([0.0610, 0.2594, 0.0312, 0.0639, 0.1130, 0.0794, 0.1788, 0.2077],
+       device='cuda:3'), in_proj_covar=tensor([0.0311, 0.0344, 0.0276, 0.0227, 0.0336, 0.0276, 0.0296, 0.0272],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 04:25:36,262 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:26:18,714 INFO [train.py:892] (3/4) Epoch 13, batch 500, loss[loss=0.1924, simple_loss=0.2714, pruned_loss=0.05671, over 19927.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.2809, pruned_loss=0.07888, over 3631386.98 frames. ], batch size: 51, lr: 1.23e-02, grad_scale: 16.0
+2023-03-28 04:27:08,314 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0194, 3.0668, 3.3501, 2.5006, 3.5997, 2.6119, 2.8994, 3.5276],
+       device='cuda:3'), covar=tensor([0.0639, 0.0306, 0.0436, 0.0679, 0.0275, 0.0420, 0.0464, 0.0223],
+       device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0064, 0.0066, 0.0095, 0.0062, 0.0060, 0.0057, 0.0051],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 04:27:17,053 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22786.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:27:33,383 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:28:12,205 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.034e+02 4.675e+02 5.713e+02 7.210e+02 1.341e+03, threshold=1.143e+03, percent-clipped=1.0
+2023-03-28 04:28:12,228 INFO [train.py:892] (3/4) Epoch 13, batch 550, loss[loss=0.2657, simple_loss=0.3187, pruned_loss=0.1064, over 19798.00 frames. ], tot_loss[loss=0.2188, simple_loss=0.2807, pruned_loss=0.07843, over 3702556.07 frames. ], batch size: 288, lr: 1.22e-02, grad_scale: 16.0
+2023-03-28 04:28:43,085 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6024, 4.8246, 4.8219, 4.9239, 4.5360, 4.7952, 4.3995, 3.9933],
+       device='cuda:3'), covar=tensor([0.0766, 0.0882, 0.1084, 0.0745, 0.1121, 0.0981, 0.1442, 0.2470],
+       device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0209, 0.0244, 0.0208, 0.0200, 0.0193, 0.0221, 0.0258],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 04:28:45,162 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22825.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:29:01,429 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:29:06,127 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:29:16,936 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3942, 2.4079, 3.3106, 3.3705, 4.0767, 4.5106, 4.3453, 4.4728],
+       device='cuda:3'), covar=tensor([0.0703, 0.1859, 0.1090, 0.0555, 0.0314, 0.0151, 0.0248, 0.0306],
+       device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0166, 0.0157, 0.0129, 0.0112, 0.0104, 0.0098, 0.0096],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 04:29:54,802 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:29:55,129 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.69 vs. limit=5.0
+2023-03-28 04:30:05,685 INFO [train.py:892] (3/4) Epoch 13, batch 600, loss[loss=0.1924, simple_loss=0.2542, pruned_loss=0.06528, over 19745.00 frames. ], tot_loss[loss=0.218, simple_loss=0.2802, pruned_loss=0.07787, over 3757556.64 frames. ], batch size: 106, lr: 1.22e-02, grad_scale: 16.0
+2023-03-28 04:30:27,501 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:30:47,656 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22880.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:31:02,241 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:31:18,040 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22893.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:31:50,167 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6450, 3.6386, 2.2179, 3.9179, 3.9366, 1.6334, 3.2400, 3.0050],
+       device='cuda:3'), covar=tensor([0.0644, 0.0765, 0.2675, 0.0616, 0.0448, 0.3331, 0.1059, 0.0712],
+       device='cuda:3'), in_proj_covar=tensor([0.0199, 0.0222, 0.0213, 0.0210, 0.0173, 0.0200, 0.0222, 0.0163],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 04:31:58,587 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.284e+02 4.701e+02 5.475e+02 6.591e+02 1.011e+03, threshold=1.095e+03, percent-clipped=0.0
+2023-03-28 04:31:58,613 INFO [train.py:892] (3/4) Epoch 13, batch 650, loss[loss=0.2129, simple_loss=0.2672, pruned_loss=0.07935, over 19751.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2809, pruned_loss=0.07825, over 3798968.88 frames. ], batch size: 139, lr: 1.22e-02, grad_scale: 16.0
+2023-03-28 04:32:16,544 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:32:22,026 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4979, 5.8922, 6.1257, 5.8897, 5.6107, 5.4847, 5.6606, 5.6583],
+       device='cuda:3'), covar=tensor([0.1456, 0.1053, 0.0898, 0.1021, 0.0730, 0.0737, 0.2021, 0.1981],
+       device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0250, 0.0306, 0.0237, 0.0227, 0.0221, 0.0296, 0.0329],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 04:32:48,421 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2822, 2.5721, 3.2277, 2.9241, 3.3239, 3.3354, 4.0768, 4.3776],
+       device='cuda:3'), covar=tensor([0.0469, 0.1848, 0.1292, 0.1879, 0.1597, 0.1379, 0.0432, 0.0371],
+       device='cuda:3'), in_proj_covar=tensor([0.0208, 0.0219, 0.0238, 0.0235, 0.0263, 0.0230, 0.0183, 0.0192],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 04:33:06,209 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:33:50,897 INFO [train.py:892] (3/4) Epoch 13, batch 700, loss[loss=0.2252, simple_loss=0.2742, pruned_loss=0.0881, over 19866.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2815, pruned_loss=0.07862, over 3832518.03 frames. ], batch size: 129, lr: 1.22e-02, grad_scale: 16.0
+2023-03-28 04:35:06,105 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22995.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 04:35:40,967 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.530e+02 4.675e+02 5.708e+02 7.114e+02 1.590e+03, threshold=1.142e+03, percent-clipped=5.0
+2023-03-28 04:35:41,001 INFO [train.py:892] (3/4) Epoch 13, batch 750, loss[loss=0.1823, simple_loss=0.2497, pruned_loss=0.05743, over 19852.00 frames. ], tot_loss[loss=0.2187, simple_loss=0.2807, pruned_loss=0.07838, over 3859576.23 frames. ], batch size: 43, lr: 1.22e-02, grad_scale: 16.0
+2023-03-28 04:36:49,835 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:37:34,797 INFO [train.py:892] (3/4) Epoch 13, batch 800, loss[loss=0.2452, simple_loss=0.2974, pruned_loss=0.09646, over 19729.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2807, pruned_loss=0.07813, over 3878937.87 frames. ], batch size: 219, lr: 1.22e-02, grad_scale: 16.0
+2023-03-28 04:38:38,898 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:39:23,150 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:39:28,914 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.056e+02 4.768e+02 5.306e+02 6.241e+02 1.399e+03, threshold=1.061e+03, percent-clipped=1.0
+2023-03-28 04:39:28,942 INFO [train.py:892] (3/4) Epoch 13, batch 850, loss[loss=0.198, simple_loss=0.2695, pruned_loss=0.06327, over 19738.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.282, pruned_loss=0.0784, over 3891299.53 frames. ], batch size: 71, lr: 1.22e-02, grad_scale: 16.0
+2023-03-28 04:40:56,421 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:41:21,503 INFO [train.py:892] (3/4) Epoch 13, batch 900, loss[loss=0.2068, simple_loss=0.2736, pruned_loss=0.06997, over 19786.00 frames. ], tot_loss[loss=0.2194, simple_loss=0.2817, pruned_loss=0.07859, over 3903297.82 frames. ], batch size: 73, lr: 1.22e-02, grad_scale: 16.0
+2023-03-28 04:41:26,706 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1134, 2.4076, 3.2494, 3.2967, 3.8008, 4.2650, 4.0644, 4.3848],
+       device='cuda:3'), covar=tensor([0.0747, 0.1772, 0.1055, 0.0520, 0.0313, 0.0146, 0.0226, 0.0197],
+       device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0167, 0.0159, 0.0131, 0.0112, 0.0105, 0.0098, 0.0096],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 04:41:39,619 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:42:00,684 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4209, 4.4664, 4.8291, 4.4278, 4.1172, 4.6924, 4.4695, 4.9917],
+       device='cuda:3'), covar=tensor([0.0984, 0.0309, 0.0350, 0.0354, 0.0745, 0.0415, 0.0431, 0.0308],
+       device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0190, 0.0187, 0.0196, 0.0185, 0.0195, 0.0190, 0.0177],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 04:42:09,296 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:43:15,287 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.371e+02 4.496e+02 5.311e+02 6.592e+02 1.240e+03, threshold=1.062e+03, percent-clipped=1.0
+2023-03-28 04:43:15,320 INFO [train.py:892] (3/4) Epoch 13, batch 950, loss[loss=0.1878, simple_loss=0.2528, pruned_loss=0.06139, over 19751.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2805, pruned_loss=0.07786, over 3915383.30 frames. ], batch size: 110, lr: 1.21e-02, grad_scale: 16.0
+2023-03-28 04:43:16,342 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0684, 2.0957, 3.3221, 2.8060, 3.2821, 3.3556, 3.1550, 3.2358],
+       device='cuda:3'), covar=tensor([0.0292, 0.0856, 0.0108, 0.0532, 0.0108, 0.0204, 0.0171, 0.0133],
+       device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0091, 0.0073, 0.0145, 0.0067, 0.0081, 0.0076, 0.0066],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 04:43:28,224 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:43:53,121 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:45:02,585 INFO [train.py:892] (3/4) Epoch 13, batch 1000, loss[loss=0.1906, simple_loss=0.2588, pruned_loss=0.06119, over 19861.00 frames. ], tot_loss[loss=0.2166, simple_loss=0.2788, pruned_loss=0.07721, over 3923892.63 frames. ], batch size: 60, lr: 1.21e-02, grad_scale: 16.0
+2023-03-28 04:45:15,038 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1428, 2.6120, 3.6573, 3.6413, 3.9989, 4.4248, 4.3139, 4.6225],
+       device='cuda:3'), covar=tensor([0.0814, 0.1746, 0.0892, 0.0482, 0.0295, 0.0168, 0.0198, 0.0272],
+       device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0169, 0.0161, 0.0134, 0.0114, 0.0108, 0.0100, 0.0098],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 04:45:39,911 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23278.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:46:10,062 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:46:20,644 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:46:46,262 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0
+2023-03-28 04:46:55,457 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.267e+02 4.704e+02 5.526e+02 6.494e+02 9.984e+02, threshold=1.105e+03, percent-clipped=0.0
+2023-03-28 04:46:55,492 INFO [train.py:892] (3/4) Epoch 13, batch 1050, loss[loss=0.2265, simple_loss=0.2914, pruned_loss=0.0808, over 19829.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2776, pruned_loss=0.07629, over 3930595.42 frames. ], batch size: 101, lr: 1.21e-02, grad_scale: 16.0
+2023-03-28 04:46:57,065 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0
+2023-03-28 04:46:58,820 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0236, 2.7581, 3.1518, 2.9463, 3.3082, 3.4097, 3.7939, 4.2477],
+       device='cuda:3'), covar=tensor([0.0532, 0.1572, 0.1289, 0.1811, 0.1658, 0.1222, 0.0542, 0.0460],
+       device='cuda:3'), in_proj_covar=tensor([0.0211, 0.0219, 0.0239, 0.0235, 0.0262, 0.0229, 0.0185, 0.0195],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 04:47:19,515 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6799, 4.3656, 4.5049, 4.2617, 4.6341, 3.2530, 3.8193, 2.2316],
+       device='cuda:3'), covar=tensor([0.0172, 0.0188, 0.0126, 0.0156, 0.0119, 0.0746, 0.0798, 0.1407],
+       device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0120, 0.0099, 0.0115, 0.0101, 0.0118, 0.0130, 0.0115],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 04:47:32,758 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.47 vs. limit=5.0
+2023-03-28 04:48:08,222 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:48:49,795 INFO [train.py:892] (3/4) Epoch 13, batch 1100, loss[loss=0.2255, simple_loss=0.2881, pruned_loss=0.08143, over 19652.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2776, pruned_loss=0.07594, over 3936061.37 frames. ], batch size: 79, lr: 1.21e-02, grad_scale: 16.0
+2023-03-28 04:50:43,331 INFO [train.py:892] (3/4) Epoch 13, batch 1150, loss[loss=0.2043, simple_loss=0.2706, pruned_loss=0.069, over 19942.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2804, pruned_loss=0.07795, over 3935706.41 frames. ], batch size: 46, lr: 1.21e-02, grad_scale: 16.0
+2023-03-28 04:50:45,464 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.703e+02 4.708e+02 5.742e+02 7.091e+02 1.301e+03, threshold=1.148e+03, percent-clipped=2.0
+2023-03-28 04:51:48,325 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8184, 3.4530, 3.5601, 3.8849, 3.6777, 3.7908, 3.9394, 4.0908],
+       device='cuda:3'), covar=tensor([0.0687, 0.0429, 0.0537, 0.0304, 0.0573, 0.0463, 0.0445, 0.0301],
+       device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0149, 0.0175, 0.0144, 0.0146, 0.0129, 0.0135, 0.0167],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 04:52:12,650 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23450.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:52:19,870 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5646, 2.4583, 3.8092, 3.3737, 3.7195, 3.9012, 3.7461, 3.7346],
+       device='cuda:3'), covar=tensor([0.0254, 0.0773, 0.0105, 0.0636, 0.0113, 0.0189, 0.0157, 0.0120],
+       device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0090, 0.0073, 0.0142, 0.0066, 0.0080, 0.0074, 0.0065],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0004, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 04:52:34,815 INFO [train.py:892] (3/4) Epoch 13, batch 1200, loss[loss=0.2032, simple_loss=0.2615, pruned_loss=0.07246, over 19816.00 frames. ], tot_loss[loss=0.2183, simple_loss=0.2808, pruned_loss=0.07794, over 3939089.40 frames. ], batch size: 123, lr: 1.21e-02, grad_scale: 16.0
+2023-03-28 04:52:42,259 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:53:21,311 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:53:39,080 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23488.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:53:58,272 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23498.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:54:26,406 INFO [train.py:892] (3/4) Epoch 13, batch 1250, loss[loss=0.2019, simple_loss=0.2638, pruned_loss=0.07002, over 19773.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2803, pruned_loss=0.07801, over 3941216.88 frames. ], batch size: 169, lr: 1.21e-02, grad_scale: 16.0
+2023-03-28 04:54:28,502 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.878e+02 4.638e+02 5.638e+02 7.017e+02 1.329e+03, threshold=1.128e+03, percent-clipped=4.0
+2023-03-28 04:55:03,134 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23529.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:55:48,773 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23549.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 04:56:12,799 INFO [train.py:892] (3/4) Epoch 13, batch 1300, loss[loss=0.2325, simple_loss=0.3025, pruned_loss=0.08126, over 19773.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.282, pruned_loss=0.07833, over 3937717.31 frames. ], batch size: 87, lr: 1.21e-02, grad_scale: 16.0
+2023-03-28 04:56:39,785 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23573.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:57:08,839 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 04:58:05,883 INFO [train.py:892] (3/4) Epoch 13, batch 1350, loss[loss=0.2419, simple_loss=0.294, pruned_loss=0.09494, over 19790.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2833, pruned_loss=0.07908, over 3939695.05 frames. ], batch size: 224, lr: 1.20e-02, grad_scale: 16.0
+2023-03-28 04:58:08,085 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.191e+02 4.401e+02 5.493e+02 6.790e+02 1.400e+03, threshold=1.099e+03, percent-clipped=0.0
+2023-03-28 04:59:38,121 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9429, 4.5267, 4.6666, 4.3949, 4.8491, 3.1753, 3.9304, 2.4865],
+       device='cuda:3'), covar=tensor([0.0156, 0.0196, 0.0139, 0.0172, 0.0131, 0.0835, 0.0815, 0.1321],
+       device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0121, 0.0101, 0.0116, 0.0102, 0.0120, 0.0132, 0.0117],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 04:59:59,167 INFO [train.py:892] (3/4) Epoch 13, batch 1400, loss[loss=0.1764, simple_loss=0.2472, pruned_loss=0.05277, over 19805.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.282, pruned_loss=0.07853, over 3941961.03 frames. ], batch size: 114, lr: 1.20e-02, grad_scale: 16.0
+2023-03-28 05:01:48,003 INFO [train.py:892] (3/4) Epoch 13, batch 1450, loss[loss=0.1871, simple_loss=0.2473, pruned_loss=0.06344, over 19875.00 frames. ], tot_loss[loss=0.2177, simple_loss=0.2807, pruned_loss=0.07732, over 3944198.35 frames. ], batch size: 159, lr: 1.20e-02, grad_scale: 16.0
+2023-03-28 05:01:50,146 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.049e+02 4.743e+02 5.551e+02 6.654e+02 1.229e+03, threshold=1.110e+03, percent-clipped=4.0
+2023-03-28 05:02:20,029 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.1739, 4.7076, 4.7759, 5.2151, 4.7846, 5.4831, 5.2790, 5.4873],
+       device='cuda:3'), covar=tensor([0.0691, 0.0363, 0.0456, 0.0263, 0.0563, 0.0279, 0.0367, 0.0280],
+       device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0151, 0.0176, 0.0145, 0.0147, 0.0131, 0.0137, 0.0168],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 05:02:31,116 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.3755, 5.7532, 6.0610, 5.7947, 5.6759, 5.4921, 5.5756, 5.5357],
+       device='cuda:3'), covar=tensor([0.1283, 0.1276, 0.0819, 0.0924, 0.0668, 0.0737, 0.1976, 0.1809],
+       device='cuda:3'), in_proj_covar=tensor([0.0259, 0.0258, 0.0315, 0.0240, 0.0232, 0.0226, 0.0305, 0.0333],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 05:02:58,611 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.74 vs. limit=5.0
+2023-03-28 05:03:30,371 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-28 05:03:37,024 INFO [train.py:892] (3/4) Epoch 13, batch 1500, loss[loss=0.1978, simple_loss=0.2695, pruned_loss=0.06306, over 19720.00 frames. ], tot_loss[loss=0.2173, simple_loss=0.28, pruned_loss=0.07728, over 3946295.06 frames. ], batch size: 81, lr: 1.20e-02, grad_scale: 16.0
+2023-03-28 05:03:45,719 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:05:18,532 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.3170, 2.4368, 2.4230, 1.7042, 2.5301, 1.9652, 2.3417, 2.5404],
+       device='cuda:3'), covar=tensor([0.0422, 0.0292, 0.0386, 0.0941, 0.0292, 0.0431, 0.0358, 0.0208],
+       device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0064, 0.0065, 0.0094, 0.0061, 0.0060, 0.0057, 0.0050],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 05:05:29,853 INFO [train.py:892] (3/4) Epoch 13, batch 1550, loss[loss=0.219, simple_loss=0.2812, pruned_loss=0.07839, over 19801.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2788, pruned_loss=0.07621, over 3947658.98 frames. ], batch size: 107, lr: 1.20e-02, grad_scale: 16.0
+2023-03-28 05:05:31,560 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.081e+02 4.439e+02 5.194e+02 6.421e+02 1.590e+03, threshold=1.039e+03, percent-clipped=3.0
+2023-03-28 05:05:32,332 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23812.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:06:39,336 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23844.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 05:07:17,777 INFO [train.py:892] (3/4) Epoch 13, batch 1600, loss[loss=0.2014, simple_loss=0.2703, pruned_loss=0.06628, over 19794.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.2784, pruned_loss=0.07544, over 3948329.73 frames. ], batch size: 83, lr: 1.20e-02, grad_scale: 16.0
+2023-03-28 05:07:26,528 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0
+2023-03-28 05:07:46,770 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23873.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:08:13,321 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:09:11,700 INFO [train.py:892] (3/4) Epoch 13, batch 1650, loss[loss=0.2114, simple_loss=0.2736, pruned_loss=0.07456, over 19822.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2779, pruned_loss=0.0753, over 3947890.53 frames. ], batch size: 204, lr: 1.20e-02, grad_scale: 16.0
+2023-03-28 05:09:13,800 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.923e+02 4.526e+02 5.652e+02 6.938e+02 1.810e+03, threshold=1.130e+03, percent-clipped=2.0
+2023-03-28 05:09:27,914 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0
+2023-03-28 05:09:36,897 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23921.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:10:02,333 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:11:04,316 INFO [train.py:892] (3/4) Epoch 13, batch 1700, loss[loss=0.2211, simple_loss=0.2795, pruned_loss=0.08131, over 19793.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2784, pruned_loss=0.07567, over 3948226.59 frames. ], batch size: 224, lr: 1.20e-02, grad_scale: 16.0
+2023-03-28 05:12:43,091 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7903, 3.1151, 3.4368, 3.3996, 4.0288, 4.0811, 4.6073, 5.1972],
+       device='cuda:3'), covar=tensor([0.0415, 0.1518, 0.1355, 0.1882, 0.1272, 0.1106, 0.0449, 0.0218],
+       device='cuda:3'), in_proj_covar=tensor([0.0216, 0.0219, 0.0241, 0.0236, 0.0265, 0.0233, 0.0188, 0.0196],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 05:12:44,945 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9724, 3.6224, 3.7709, 4.0809, 3.7949, 3.9884, 4.1477, 4.2337],
+       device='cuda:3'), covar=tensor([0.0663, 0.0406, 0.0502, 0.0283, 0.0576, 0.0455, 0.0371, 0.0332],
+       device='cuda:3'), in_proj_covar=tensor([0.0131, 0.0151, 0.0176, 0.0145, 0.0148, 0.0132, 0.0137, 0.0168],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 05:12:55,563 INFO [train.py:892] (3/4) Epoch 13, batch 1750, loss[loss=0.2112, simple_loss=0.274, pruned_loss=0.07427, over 19769.00 frames. ], tot_loss[loss=0.2171, simple_loss=0.2798, pruned_loss=0.07715, over 3948064.09 frames. ], batch size: 113, lr: 1.19e-02, grad_scale: 16.0
+2023-03-28 05:12:57,430 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.884e+02 4.345e+02 5.223e+02 6.698e+02 1.111e+03, threshold=1.045e+03, percent-clipped=0.0
+2023-03-28 05:13:13,380 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0644, 2.8110, 3.0764, 2.8546, 3.3864, 3.3356, 3.9348, 4.2980],
+       device='cuda:3'), covar=tensor([0.0553, 0.1513, 0.1387, 0.1941, 0.1431, 0.1246, 0.0481, 0.0424],
+       device='cuda:3'), in_proj_covar=tensor([0.0216, 0.0219, 0.0240, 0.0236, 0.0264, 0.0233, 0.0188, 0.0196],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 05:13:46,321 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:13:46,400 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2871, 3.3687, 3.4555, 4.4195, 2.8231, 3.4354, 2.9588, 2.5147],
+       device='cuda:3'), covar=tensor([0.0495, 0.2276, 0.1055, 0.0273, 0.2300, 0.0850, 0.1203, 0.1834],
+       device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0331, 0.0227, 0.0161, 0.0241, 0.0180, 0.0201, 0.0207],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 05:14:30,474 INFO [train.py:892] (3/4) Epoch 13, batch 1800, loss[loss=0.1942, simple_loss=0.2732, pruned_loss=0.05764, over 19695.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.2788, pruned_loss=0.07627, over 3948535.06 frames. ], batch size: 48, lr: 1.19e-02, grad_scale: 16.0
+2023-03-28 05:14:52,472 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.2545, 2.4489, 2.5848, 2.3792, 2.1919, 2.4080, 2.4145, 2.4890],
+       device='cuda:3'), covar=tensor([0.0205, 0.0272, 0.0193, 0.0201, 0.0314, 0.0224, 0.0284, 0.0290],
+       device='cuda:3'), in_proj_covar=tensor([0.0050, 0.0050, 0.0052, 0.0045, 0.0056, 0.0051, 0.0069, 0.0046],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-28 05:15:37,803 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:16:01,669 INFO [train.py:892] (3/4) Epoch 13, batch 1850, loss[loss=0.2304, simple_loss=0.3052, pruned_loss=0.07784, over 19825.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2796, pruned_loss=0.07561, over 3949694.87 frames. ], batch size: 57, lr: 1.19e-02, grad_scale: 16.0
+2023-03-28 05:16:03,698 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.093e+02 4.684e+02 5.642e+02 7.102e+02 1.280e+03, threshold=1.128e+03, percent-clipped=3.0
+2023-03-28 05:17:07,729 INFO [train.py:892] (3/4) Epoch 14, batch 0, loss[loss=0.1979, simple_loss=0.2543, pruned_loss=0.07075, over 19794.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2543, pruned_loss=0.07075, over 19794.00 frames. ], batch size: 126, lr: 1.15e-02, grad_scale: 16.0
+2023-03-28 05:17:07,730 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-28 05:17:42,314 INFO [train.py:926] (3/4) Epoch 14, validation: loss=0.1725, simple_loss=0.2522, pruned_loss=0.04642, over 2883724.00 frames. 
+2023-03-28 05:17:42,315 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-28 05:18:52,128 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=24144.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 05:19:45,555 INFO [train.py:892] (3/4) Epoch 14, batch 50, loss[loss=0.2622, simple_loss=0.3157, pruned_loss=0.1043, over 19637.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2704, pruned_loss=0.07179, over 890449.35 frames. ], batch size: 330, lr: 1.15e-02, grad_scale: 16.0
+2023-03-28 05:20:44,925 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=24192.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:21:30,231 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.134e+02 4.338e+02 5.278e+02 6.303e+02 1.236e+03, threshold=1.056e+03, percent-clipped=1.0
+2023-03-28 05:21:38,380 INFO [train.py:892] (3/4) Epoch 14, batch 100, loss[loss=0.2131, simple_loss=0.2772, pruned_loss=0.0745, over 19701.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2727, pruned_loss=0.07328, over 1569155.30 frames. ], batch size: 46, lr: 1.15e-02, grad_scale: 16.0
+2023-03-28 05:21:47,255 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4426, 3.5388, 3.7392, 4.5001, 2.9985, 3.4305, 2.9698, 2.5542],
+       device='cuda:3'), covar=tensor([0.0417, 0.2452, 0.0878, 0.0272, 0.2112, 0.0802, 0.1101, 0.1842],
+       device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0333, 0.0229, 0.0163, 0.0242, 0.0181, 0.0201, 0.0209],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 05:23:30,580 INFO [train.py:892] (3/4) Epoch 14, batch 150, loss[loss=0.1869, simple_loss=0.2552, pruned_loss=0.05927, over 19750.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2736, pruned_loss=0.07378, over 2096525.99 frames. ], batch size: 44, lr: 1.15e-02, grad_scale: 16.0
+2023-03-28 05:24:41,488 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0151, 3.8411, 3.8322, 3.6297, 3.9731, 2.9487, 3.2915, 1.9445],
+       device='cuda:3'), covar=tensor([0.0223, 0.0212, 0.0167, 0.0195, 0.0168, 0.0871, 0.0727, 0.1547],
+       device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0122, 0.0102, 0.0116, 0.0104, 0.0121, 0.0130, 0.0116],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 05:25:15,402 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.679e+02 4.473e+02 5.483e+02 6.976e+02 1.116e+03, threshold=1.097e+03, percent-clipped=1.0
+2023-03-28 05:25:23,189 INFO [train.py:892] (3/4) Epoch 14, batch 200, loss[loss=0.1687, simple_loss=0.2319, pruned_loss=0.0528, over 19729.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2761, pruned_loss=0.07493, over 2506995.71 frames. ], batch size: 47, lr: 1.14e-02, grad_scale: 16.0
+2023-03-28 05:25:29,258 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.01 vs. limit=5.0
+2023-03-28 05:25:48,463 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2343, 2.4331, 3.5238, 3.0546, 3.4693, 3.5747, 3.4334, 3.3735],
+       device='cuda:3'), covar=tensor([0.0290, 0.0753, 0.0112, 0.0573, 0.0106, 0.0211, 0.0140, 0.0145],
+       device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0092, 0.0074, 0.0144, 0.0068, 0.0083, 0.0077, 0.0069],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0002, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 05:27:13,849 INFO [train.py:892] (3/4) Epoch 14, batch 250, loss[loss=0.2169, simple_loss=0.2711, pruned_loss=0.08133, over 19795.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2769, pruned_loss=0.07507, over 2826424.91 frames. ], batch size: 185, lr: 1.14e-02, grad_scale: 16.0
+2023-03-28 05:28:09,403 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0
+2023-03-28 05:28:15,411 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:28:28,171 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2992, 2.8041, 3.1480, 3.0317, 3.3904, 3.3900, 4.1803, 4.5462],
+       device='cuda:3'), covar=tensor([0.0505, 0.1725, 0.1544, 0.1979, 0.1916, 0.1494, 0.0479, 0.0454],
+       device='cuda:3'), in_proj_covar=tensor([0.0214, 0.0221, 0.0241, 0.0236, 0.0267, 0.0232, 0.0188, 0.0198],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 05:28:38,814 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4887, 2.5841, 3.7116, 2.8575, 3.2356, 3.1605, 2.0315, 2.1329],
+       device='cuda:3'), covar=tensor([0.0891, 0.2816, 0.0523, 0.0783, 0.1431, 0.1046, 0.1994, 0.2489],
+       device='cuda:3'), in_proj_covar=tensor([0.0313, 0.0346, 0.0284, 0.0229, 0.0339, 0.0285, 0.0304, 0.0278],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 05:28:44,949 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3374, 3.2031, 4.7576, 3.5144, 3.9122, 3.9027, 2.2509, 2.5451],
+       device='cuda:3'), covar=tensor([0.0648, 0.2907, 0.0366, 0.0687, 0.1400, 0.0824, 0.2090, 0.2272],
+       device='cuda:3'), in_proj_covar=tensor([0.0313, 0.0346, 0.0285, 0.0229, 0.0339, 0.0285, 0.0304, 0.0278],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 05:28:53,721 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.621e+02 4.746e+02 5.640e+02 6.808e+02 1.324e+03, threshold=1.128e+03, percent-clipped=3.0
+2023-03-28 05:29:04,966 INFO [train.py:892] (3/4) Epoch 14, batch 300, loss[loss=0.179, simple_loss=0.253, pruned_loss=0.05251, over 19492.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2776, pruned_loss=0.07482, over 3075660.01 frames. ], batch size: 43, lr: 1.14e-02, grad_scale: 16.0
+2023-03-28 05:29:55,389 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1210, 2.7392, 3.1271, 2.9359, 3.3367, 3.2965, 4.0240, 4.3614],
+       device='cuda:3'), covar=tensor([0.0452, 0.1505, 0.1334, 0.1793, 0.1420, 0.1203, 0.0423, 0.0313],
+       device='cuda:3'), in_proj_covar=tensor([0.0216, 0.0222, 0.0243, 0.0238, 0.0268, 0.0233, 0.0190, 0.0199],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 05:30:41,226 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0
+2023-03-28 05:30:56,502 INFO [train.py:892] (3/4) Epoch 14, batch 350, loss[loss=0.2069, simple_loss=0.2807, pruned_loss=0.06658, over 19814.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2772, pruned_loss=0.07445, over 3268773.72 frames. ], batch size: 40, lr: 1.14e-02, grad_scale: 16.0
+2023-03-28 05:31:06,314 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0644, 2.6874, 3.1328, 2.9502, 3.2636, 3.1939, 3.9835, 4.2868],
+       device='cuda:3'), covar=tensor([0.0548, 0.1685, 0.1420, 0.1832, 0.1740, 0.1551, 0.0434, 0.0393],
+       device='cuda:3'), in_proj_covar=tensor([0.0217, 0.0223, 0.0244, 0.0239, 0.0270, 0.0234, 0.0190, 0.0200],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 05:32:18,085 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5039, 2.0008, 2.5821, 2.9441, 3.4395, 3.6661, 3.5998, 3.7579],
+       device='cuda:3'), covar=tensor([0.0870, 0.1782, 0.1131, 0.0559, 0.0371, 0.0210, 0.0267, 0.0280],
+       device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0170, 0.0165, 0.0133, 0.0117, 0.0110, 0.0103, 0.0098],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 05:32:38,982 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.146e+02 4.463e+02 5.276e+02 6.433e+02 1.148e+03, threshold=1.055e+03, percent-clipped=1.0
+2023-03-28 05:32:42,879 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.30 vs. limit=5.0
+2023-03-28 05:32:48,483 INFO [train.py:892] (3/4) Epoch 14, batch 400, loss[loss=0.2246, simple_loss=0.2809, pruned_loss=0.08415, over 19831.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2764, pruned_loss=0.07403, over 3419919.64 frames. ], batch size: 204, lr: 1.14e-02, grad_scale: 16.0
+2023-03-28 05:34:00,610 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:34:41,058 INFO [train.py:892] (3/4) Epoch 14, batch 450, loss[loss=0.2038, simple_loss=0.2747, pruned_loss=0.06648, over 19654.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2761, pruned_loss=0.07333, over 3536764.62 frames. ], batch size: 79, lr: 1.14e-02, grad_scale: 16.0
+2023-03-28 05:34:48,617 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:36:16,280 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:36:24,998 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.188e+02 4.392e+02 5.020e+02 6.221e+02 1.818e+03, threshold=1.004e+03, percent-clipped=1.0
+2023-03-28 05:36:33,460 INFO [train.py:892] (3/4) Epoch 14, batch 500, loss[loss=0.2116, simple_loss=0.2766, pruned_loss=0.07327, over 19895.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2775, pruned_loss=0.07442, over 3627147.76 frames. ], batch size: 87, lr: 1.14e-02, grad_scale: 16.0
+2023-03-28 05:37:04,342 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:37:04,655 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-28 05:37:10,412 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0347, 3.2626, 3.5521, 2.6599, 3.5289, 2.7761, 2.9026, 3.6106],
+       device='cuda:3'), covar=tensor([0.0606, 0.0284, 0.0346, 0.0675, 0.0313, 0.0389, 0.0544, 0.0215],
+       device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0063, 0.0065, 0.0094, 0.0062, 0.0060, 0.0058, 0.0050],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 05:38:20,299 INFO [train.py:892] (3/4) Epoch 14, batch 550, loss[loss=0.2226, simple_loss=0.2859, pruned_loss=0.07969, over 19785.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.279, pruned_loss=0.0753, over 3697916.61 frames. ], batch size: 120, lr: 1.14e-02, grad_scale: 16.0
+2023-03-28 05:39:20,328 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=24693.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:39:57,876 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.918e+02 4.778e+02 5.297e+02 6.778e+02 1.181e+03, threshold=1.059e+03, percent-clipped=3.0
+2023-03-28 05:40:07,599 INFO [train.py:892] (3/4) Epoch 14, batch 600, loss[loss=0.2, simple_loss=0.2619, pruned_loss=0.06901, over 19613.00 frames. ], tot_loss[loss=0.213, simple_loss=0.2771, pruned_loss=0.07443, over 3754616.04 frames. ], batch size: 51, lr: 1.14e-02, grad_scale: 16.0
+2023-03-28 05:40:18,113 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.3263, 1.4342, 1.2479, 0.7517, 1.3711, 1.3173, 1.3227, 1.3585],
+       device='cuda:3'), covar=tensor([0.0238, 0.0186, 0.0295, 0.0473, 0.0430, 0.0179, 0.0184, 0.0188],
+       device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0066, 0.0074, 0.0081, 0.0085, 0.0058, 0.0055, 0.0058],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-28 05:41:01,101 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=24741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:41:54,327 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:41:55,319 INFO [train.py:892] (3/4) Epoch 14, batch 650, loss[loss=0.1899, simple_loss=0.2563, pruned_loss=0.06175, over 19829.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2766, pruned_loss=0.07413, over 3797152.24 frames. ], batch size: 101, lr: 1.13e-02, grad_scale: 16.0
+2023-03-28 05:43:34,887 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.281e+02 4.693e+02 5.521e+02 6.311e+02 1.002e+03, threshold=1.104e+03, percent-clipped=0.0
+2023-03-28 05:43:42,883 INFO [train.py:892] (3/4) Epoch 14, batch 700, loss[loss=0.2089, simple_loss=0.2691, pruned_loss=0.07436, over 19818.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2775, pruned_loss=0.07472, over 3830501.29 frames. ], batch size: 127, lr: 1.13e-02, grad_scale: 16.0
+2023-03-28 05:43:47,649 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3740, 4.7070, 4.7258, 4.6440, 4.3979, 4.6724, 4.1888, 4.2541],
+       device='cuda:3'), covar=tensor([0.0425, 0.0418, 0.0498, 0.0397, 0.0538, 0.0529, 0.0576, 0.0840],
+       device='cuda:3'), in_proj_covar=tensor([0.0213, 0.0217, 0.0248, 0.0213, 0.0207, 0.0198, 0.0224, 0.0259],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 05:44:08,707 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4948, 3.5574, 3.8466, 3.6322, 3.8693, 3.2678, 3.4974, 3.4378],
+       device='cuda:3'), covar=tensor([0.1528, 0.1603, 0.1292, 0.1335, 0.1020, 0.1412, 0.2241, 0.2571],
+       device='cuda:3'), in_proj_covar=tensor([0.0256, 0.0257, 0.0314, 0.0241, 0.0232, 0.0229, 0.0306, 0.0335],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 05:44:08,863 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:44:20,644 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9415, 2.9251, 3.4529, 2.2796, 3.4968, 2.8738, 2.8600, 3.4555],
+       device='cuda:3'), covar=tensor([0.0682, 0.0389, 0.0375, 0.0823, 0.0249, 0.0321, 0.0338, 0.0209],
+       device='cuda:3'), in_proj_covar=tensor([0.0062, 0.0064, 0.0065, 0.0095, 0.0062, 0.0061, 0.0059, 0.0052],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 05:44:55,397 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:45:12,580 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0
+2023-03-28 05:45:36,126 INFO [train.py:892] (3/4) Epoch 14, batch 750, loss[loss=0.1866, simple_loss=0.254, pruned_loss=0.05964, over 19644.00 frames. ], tot_loss[loss=0.2138, simple_loss=0.2778, pruned_loss=0.0749, over 3856025.69 frames. ], batch size: 69, lr: 1.13e-02, grad_scale: 16.0
+2023-03-28 05:46:58,234 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:47:10,911 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:47:18,009 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.878e+02 4.189e+02 5.333e+02 6.516e+02 1.195e+03, threshold=1.067e+03, percent-clipped=1.0
+2023-03-28 05:47:26,448 INFO [train.py:892] (3/4) Epoch 14, batch 800, loss[loss=0.2078, simple_loss=0.2757, pruned_loss=0.06992, over 19705.00 frames. ], tot_loss[loss=0.2133, simple_loss=0.2773, pruned_loss=0.07466, over 3877549.34 frames. ], batch size: 78, lr: 1.13e-02, grad_scale: 16.0
+2023-03-28 05:47:45,219 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:48:47,028 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5895, 4.5944, 5.0787, 4.5849, 4.1699, 4.8031, 4.6256, 5.1519],
+       device='cuda:3'), covar=tensor([0.0911, 0.0371, 0.0330, 0.0375, 0.0801, 0.0403, 0.0467, 0.0286],
+       device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0195, 0.0191, 0.0202, 0.0190, 0.0200, 0.0200, 0.0182],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 05:49:15,225 INFO [train.py:892] (3/4) Epoch 14, batch 850, loss[loss=0.1835, simple_loss=0.2544, pruned_loss=0.0563, over 19849.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2776, pruned_loss=0.07459, over 3894087.33 frames. ], batch size: 85, lr: 1.13e-02, grad_scale: 16.0
+2023-03-28 05:49:16,497 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-28 05:50:22,793 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:51:00,274 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.835e+02 4.699e+02 5.581e+02 6.608e+02 1.039e+03, threshold=1.116e+03, percent-clipped=0.0
+2023-03-28 05:51:08,785 INFO [train.py:892] (3/4) Epoch 14, batch 900, loss[loss=0.1928, simple_loss=0.2502, pruned_loss=0.06774, over 19866.00 frames. ], tot_loss[loss=0.2123, simple_loss=0.2763, pruned_loss=0.07411, over 3906773.07 frames. ], batch size: 129, lr: 1.13e-02, grad_scale: 16.0
+2023-03-28 05:52:39,901 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25056.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 05:53:00,403 INFO [train.py:892] (3/4) Epoch 14, batch 950, loss[loss=0.1974, simple_loss=0.2746, pruned_loss=0.06016, over 19761.00 frames. ], tot_loss[loss=0.2131, simple_loss=0.2769, pruned_loss=0.0747, over 3915872.45 frames. ], batch size: 89, lr: 1.13e-02, grad_scale: 16.0
+2023-03-28 05:54:16,060 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3943, 2.5904, 3.7645, 3.2538, 3.5846, 3.8197, 3.6400, 3.5917],
+       device='cuda:3'), covar=tensor([0.0248, 0.0632, 0.0095, 0.0559, 0.0104, 0.0181, 0.0142, 0.0137],
+       device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0091, 0.0074, 0.0144, 0.0067, 0.0082, 0.0076, 0.0068],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0004, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 05:54:42,044 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.968e+02 4.212e+02 5.107e+02 6.466e+02 1.651e+03, threshold=1.021e+03, percent-clipped=2.0
+2023-03-28 05:54:50,390 INFO [train.py:892] (3/4) Epoch 14, batch 1000, loss[loss=0.1763, simple_loss=0.2425, pruned_loss=0.05509, over 19764.00 frames. ], tot_loss[loss=0.2111, simple_loss=0.2751, pruned_loss=0.07355, over 3924309.60 frames. ], batch size: 70, lr: 1.13e-02, grad_scale: 16.0
+2023-03-28 05:55:04,851 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:56:40,708 INFO [train.py:892] (3/4) Epoch 14, batch 1050, loss[loss=0.2251, simple_loss=0.287, pruned_loss=0.08161, over 19765.00 frames. ], tot_loss[loss=0.2111, simple_loss=0.2752, pruned_loss=0.07347, over 3931145.18 frames. ], batch size: 263, lr: 1.13e-02, grad_scale: 16.0
+2023-03-28 05:58:05,190 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25203.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:58:07,465 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:58:23,776 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.002e+02 4.493e+02 5.579e+02 6.984e+02 1.281e+03, threshold=1.116e+03, percent-clipped=2.0
+2023-03-28 05:58:32,619 INFO [train.py:892] (3/4) Epoch 14, batch 1100, loss[loss=0.1933, simple_loss=0.2597, pruned_loss=0.06348, over 19806.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2754, pruned_loss=0.07349, over 3935517.96 frames. ], batch size: 98, lr: 1.12e-02, grad_scale: 16.0
+2023-03-28 05:58:53,670 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 05:58:55,936 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6431, 2.0706, 2.3591, 2.9169, 3.3913, 3.5406, 3.4863, 3.6145],
+       device='cuda:3'), covar=tensor([0.0797, 0.1707, 0.1295, 0.0545, 0.0326, 0.0195, 0.0246, 0.0271],
+       device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0171, 0.0169, 0.0132, 0.0116, 0.0110, 0.0102, 0.0098],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 05:59:55,685 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:00:27,167 INFO [train.py:892] (3/4) Epoch 14, batch 1150, loss[loss=0.2082, simple_loss=0.2708, pruned_loss=0.07282, over 19738.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2762, pruned_loss=0.07409, over 3938781.49 frames. ], batch size: 118, lr: 1.12e-02, grad_scale: 16.0
+2023-03-28 06:00:46,036 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:02:10,583 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.130e+02 4.700e+02 5.345e+02 6.282e+02 1.184e+03, threshold=1.069e+03, percent-clipped=1.0
+2023-03-28 06:02:18,771 INFO [train.py:892] (3/4) Epoch 14, batch 1200, loss[loss=0.191, simple_loss=0.2611, pruned_loss=0.06049, over 19875.00 frames. ], tot_loss[loss=0.2113, simple_loss=0.2752, pruned_loss=0.0737, over 3941516.78 frames. ], batch size: 125, lr: 1.12e-02, grad_scale: 16.0
+2023-03-28 06:03:15,106 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0
+2023-03-28 06:03:35,803 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25351.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 06:04:08,272 INFO [train.py:892] (3/4) Epoch 14, batch 1250, loss[loss=0.2074, simple_loss=0.2714, pruned_loss=0.07175, over 19845.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2766, pruned_loss=0.07489, over 3943667.09 frames. ], batch size: 190, lr: 1.12e-02, grad_scale: 32.0
+2023-03-28 06:05:51,971 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.364e+02 4.665e+02 5.361e+02 6.465e+02 1.187e+03, threshold=1.072e+03, percent-clipped=1.0
+2023-03-28 06:06:00,015 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0172, 3.9727, 3.9403, 3.7552, 4.0726, 3.0280, 3.2216, 1.9057],
+       device='cuda:3'), covar=tensor([0.0351, 0.0247, 0.0206, 0.0210, 0.0240, 0.0933, 0.1041, 0.2053],
+       device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0124, 0.0102, 0.0118, 0.0105, 0.0122, 0.0133, 0.0118],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 06:06:00,925 INFO [train.py:892] (3/4) Epoch 14, batch 1300, loss[loss=0.2029, simple_loss=0.2593, pruned_loss=0.07322, over 19764.00 frames. ], tot_loss[loss=0.2127, simple_loss=0.2762, pruned_loss=0.07454, over 3944577.04 frames. ], batch size: 179, lr: 1.12e-02, grad_scale: 32.0
+2023-03-28 06:06:14,951 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:06:31,010 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25428.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 06:07:56,733 INFO [train.py:892] (3/4) Epoch 14, batch 1350, loss[loss=0.217, simple_loss=0.2905, pruned_loss=0.07172, over 19783.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2758, pruned_loss=0.07426, over 3947165.77 frames. ], batch size: 53, lr: 1.12e-02, grad_scale: 32.0
+2023-03-28 06:08:04,481 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25469.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:08:31,143 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0
+2023-03-28 06:08:34,530 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.9421, 6.3206, 6.3196, 6.2092, 5.9939, 6.2655, 5.5269, 5.6330],
+       device='cuda:3'), covar=tensor([0.0335, 0.0325, 0.0411, 0.0372, 0.0538, 0.0503, 0.0562, 0.0849],
+       device='cuda:3'), in_proj_covar=tensor([0.0218, 0.0221, 0.0254, 0.0218, 0.0210, 0.0203, 0.0228, 0.0265],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 06:08:50,246 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25489.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 06:09:23,828 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:09:38,483 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.060e+02 4.588e+02 5.403e+02 6.911e+02 1.423e+03, threshold=1.081e+03, percent-clipped=2.0
+2023-03-28 06:09:46,911 INFO [train.py:892] (3/4) Epoch 14, batch 1400, loss[loss=0.2074, simple_loss=0.2817, pruned_loss=0.06654, over 19758.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2759, pruned_loss=0.07425, over 3946416.87 frames. ], batch size: 49, lr: 1.12e-02, grad_scale: 32.0
+2023-03-28 06:11:11,935 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:11:39,099 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1180, 3.3386, 3.7038, 2.6228, 3.7707, 2.9480, 3.0528, 3.6556],
+       device='cuda:3'), covar=tensor([0.0663, 0.0312, 0.0414, 0.0751, 0.0369, 0.0319, 0.0384, 0.0244],
+       device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0064, 0.0066, 0.0094, 0.0063, 0.0060, 0.0059, 0.0052],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 06:11:41,889 INFO [train.py:892] (3/4) Epoch 14, batch 1450, loss[loss=0.2264, simple_loss=0.2899, pruned_loss=0.0815, over 19859.00 frames. ], tot_loss[loss=0.2118, simple_loss=0.2758, pruned_loss=0.07392, over 3946596.91 frames. ], batch size: 51, lr: 1.12e-02, grad_scale: 32.0
+2023-03-28 06:13:26,198 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.132e+02 4.450e+02 5.479e+02 6.667e+02 1.081e+03, threshold=1.096e+03, percent-clipped=1.0
+2023-03-28 06:13:32,614 INFO [train.py:892] (3/4) Epoch 14, batch 1500, loss[loss=0.1892, simple_loss=0.2553, pruned_loss=0.06159, over 19688.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2748, pruned_loss=0.07317, over 3948222.73 frames. ], batch size: 74, lr: 1.12e-02, grad_scale: 16.0
+2023-03-28 06:14:51,791 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25651.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 06:15:22,147 INFO [train.py:892] (3/4) Epoch 14, batch 1550, loss[loss=0.1905, simple_loss=0.2472, pruned_loss=0.06694, over 19764.00 frames. ], tot_loss[loss=0.2119, simple_loss=0.2766, pruned_loss=0.07362, over 3946026.97 frames. ], batch size: 122, lr: 1.11e-02, grad_scale: 16.0
+2023-03-28 06:16:39,237 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:16:56,823 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5492, 4.8668, 4.9082, 4.8205, 4.4585, 4.8471, 4.3406, 4.3834],
+       device='cuda:3'), covar=tensor([0.0460, 0.0467, 0.0514, 0.0474, 0.0611, 0.0594, 0.0683, 0.0978],
+       device='cuda:3'), in_proj_covar=tensor([0.0218, 0.0221, 0.0255, 0.0219, 0.0212, 0.0202, 0.0229, 0.0264],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 06:17:03,353 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-28 06:17:12,374 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.797e+02 4.271e+02 5.006e+02 6.472e+02 1.345e+03, threshold=1.001e+03, percent-clipped=3.0
+2023-03-28 06:17:14,224 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0
+2023-03-28 06:17:19,593 INFO [train.py:892] (3/4) Epoch 14, batch 1600, loss[loss=0.219, simple_loss=0.2756, pruned_loss=0.08124, over 19768.00 frames. ], tot_loss[loss=0.2108, simple_loss=0.2755, pruned_loss=0.07306, over 3946556.48 frames. ], batch size: 188, lr: 1.11e-02, grad_scale: 16.0
+2023-03-28 06:18:01,926 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5336, 2.6659, 3.9473, 3.5008, 3.7868, 3.8716, 3.8248, 3.6221],
+       device='cuda:3'), covar=tensor([0.0255, 0.0731, 0.0078, 0.0491, 0.0096, 0.0204, 0.0107, 0.0147],
+       device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0092, 0.0073, 0.0143, 0.0068, 0.0083, 0.0076, 0.0068],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0004, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 06:19:09,490 INFO [train.py:892] (3/4) Epoch 14, batch 1650, loss[loss=0.2073, simple_loss=0.2755, pruned_loss=0.06949, over 19882.00 frames. ], tot_loss[loss=0.21, simple_loss=0.2747, pruned_loss=0.0726, over 3947807.98 frames. ], batch size: 61, lr: 1.11e-02, grad_scale: 16.0
+2023-03-28 06:19:49,955 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25784.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 06:20:52,547 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.972e+02 4.640e+02 5.535e+02 6.802e+02 1.148e+03, threshold=1.107e+03, percent-clipped=2.0
+2023-03-28 06:21:01,430 INFO [train.py:892] (3/4) Epoch 14, batch 1700, loss[loss=0.206, simple_loss=0.2733, pruned_loss=0.06933, over 19719.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.274, pruned_loss=0.07182, over 3947633.32 frames. ], batch size: 109, lr: 1.11e-02, grad_scale: 16.0
+2023-03-28 06:22:29,044 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25858.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 06:22:38,720 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9957, 2.6217, 3.1685, 3.6003, 4.0486, 4.4852, 4.2380, 4.5665],
+       device='cuda:3'), covar=tensor([0.0805, 0.1539, 0.1157, 0.0434, 0.0235, 0.0143, 0.0210, 0.0231],
+       device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0168, 0.0167, 0.0132, 0.0116, 0.0109, 0.0106, 0.0099],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 06:22:43,569 INFO [train.py:892] (3/4) Epoch 14, batch 1750, loss[loss=0.1874, simple_loss=0.2523, pruned_loss=0.06121, over 19897.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2719, pruned_loss=0.07108, over 3949694.50 frames. ], batch size: 87, lr: 1.11e-02, grad_scale: 16.0
+2023-03-28 06:22:46,910 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-28 06:24:15,401 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.161e+02 4.534e+02 5.430e+02 6.216e+02 1.247e+03, threshold=1.086e+03, percent-clipped=3.0
+2023-03-28 06:24:21,542 INFO [train.py:892] (3/4) Epoch 14, batch 1800, loss[loss=0.2047, simple_loss=0.2767, pruned_loss=0.06638, over 19639.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2728, pruned_loss=0.07159, over 3949755.69 frames. ], batch size: 66, lr: 1.11e-02, grad_scale: 16.0
+2023-03-28 06:24:28,125 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25919.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 06:24:36,257 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8008, 3.0679, 2.5804, 2.0405, 2.6594, 2.9837, 2.9680, 2.9750],
+       device='cuda:3'), covar=tensor([0.0177, 0.0199, 0.0237, 0.0475, 0.0328, 0.0214, 0.0140, 0.0159],
+       device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0067, 0.0075, 0.0082, 0.0086, 0.0059, 0.0056, 0.0059],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-28 06:25:52,595 INFO [train.py:892] (3/4) Epoch 14, batch 1850, loss[loss=0.2205, simple_loss=0.2949, pruned_loss=0.07308, over 19686.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2749, pruned_loss=0.0715, over 3948264.34 frames. ], batch size: 55, lr: 1.11e-02, grad_scale: 16.0
+2023-03-28 06:26:59,300 INFO [train.py:892] (3/4) Epoch 15, batch 0, loss[loss=0.1809, simple_loss=0.2473, pruned_loss=0.05724, over 19754.00 frames. ], tot_loss[loss=0.1809, simple_loss=0.2473, pruned_loss=0.05724, over 19754.00 frames. ], batch size: 110, lr: 1.07e-02, grad_scale: 16.0
+2023-03-28 06:26:59,301 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-28 06:27:35,345 INFO [train.py:926] (3/4) Epoch 15, validation: loss=0.1719, simple_loss=0.2516, pruned_loss=0.0461, over 2883724.00 frames. 
+2023-03-28 06:27:35,346 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-28 06:29:17,397 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.761e+02 4.443e+02 5.358e+02 6.382e+02 1.006e+03, threshold=1.072e+03, percent-clipped=0.0
+2023-03-28 06:29:35,977 INFO [train.py:892] (3/4) Epoch 15, batch 50, loss[loss=0.1827, simple_loss=0.2494, pruned_loss=0.05797, over 19737.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2709, pruned_loss=0.07269, over 891534.87 frames. ], batch size: 118, lr: 1.07e-02, grad_scale: 16.0
+2023-03-28 06:29:51,459 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-28 06:30:27,618 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7832, 3.7814, 2.2746, 4.0338, 4.1618, 1.7741, 3.3534, 3.1479],
+       device='cuda:3'), covar=tensor([0.0741, 0.0953, 0.2782, 0.0715, 0.0490, 0.3205, 0.1168, 0.0791],
+       device='cuda:3'), in_proj_covar=tensor([0.0202, 0.0226, 0.0213, 0.0222, 0.0187, 0.0195, 0.0222, 0.0167],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 06:31:24,910 INFO [train.py:892] (3/4) Epoch 15, batch 100, loss[loss=0.2967, simple_loss=0.3549, pruned_loss=0.1193, over 19579.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.272, pruned_loss=0.07141, over 1569244.06 frames. ], batch size: 376, lr: 1.07e-02, grad_scale: 16.0
+2023-03-28 06:31:57,103 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26084.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 06:32:35,015 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:32:48,861 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3925, 4.3239, 4.7563, 4.4963, 4.5850, 4.1553, 4.4508, 4.2083],
+       device='cuda:3'), covar=tensor([0.1396, 0.1484, 0.0948, 0.1300, 0.0920, 0.1007, 0.1841, 0.2142],
+       device='cuda:3'), in_proj_covar=tensor([0.0260, 0.0263, 0.0317, 0.0250, 0.0239, 0.0233, 0.0310, 0.0336],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 06:32:57,715 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.557e+02 4.232e+02 5.319e+02 6.491e+02 1.379e+03, threshold=1.064e+03, percent-clipped=5.0
+2023-03-28 06:33:15,666 INFO [train.py:892] (3/4) Epoch 15, batch 150, loss[loss=0.175, simple_loss=0.2469, pruned_loss=0.0516, over 19758.00 frames. ], tot_loss[loss=0.207, simple_loss=0.2715, pruned_loss=0.07127, over 2097144.92 frames. ], batch size: 88, lr: 1.07e-02, grad_scale: 16.0
+2023-03-28 06:33:41,877 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26132.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 06:34:21,604 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0
+2023-03-28 06:34:52,849 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:35:03,873 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:35:09,210 INFO [train.py:892] (3/4) Epoch 15, batch 200, loss[loss=0.2061, simple_loss=0.2749, pruned_loss=0.06866, over 19787.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2728, pruned_loss=0.07125, over 2507775.59 frames. ], batch size: 91, lr: 1.07e-02, grad_scale: 16.0
+2023-03-28 06:36:22,746 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0
+2023-03-28 06:36:40,518 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-28 06:36:44,890 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.302e+02 4.298e+02 5.201e+02 6.207e+02 1.187e+03, threshold=1.040e+03, percent-clipped=4.0
+2023-03-28 06:36:48,225 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26214.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 06:37:02,577 INFO [train.py:892] (3/4) Epoch 15, batch 250, loss[loss=0.1965, simple_loss=0.2581, pruned_loss=0.06741, over 19701.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.271, pruned_loss=0.07022, over 2828429.00 frames. ], batch size: 82, lr: 1.07e-02, grad_scale: 16.0
+2023-03-28 06:37:24,940 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:38:12,282 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:38:16,528 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0
+2023-03-28 06:38:58,534 INFO [train.py:892] (3/4) Epoch 15, batch 300, loss[loss=0.1989, simple_loss=0.2556, pruned_loss=0.07112, over 19881.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2726, pruned_loss=0.07098, over 3074313.66 frames. ], batch size: 136, lr: 1.06e-02, grad_scale: 16.0
+2023-03-28 06:40:32,948 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:40:36,425 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.754e+02 4.352e+02 4.963e+02 6.294e+02 9.634e+02, threshold=9.926e+02, percent-clipped=0.0
+2023-03-28 06:40:55,207 INFO [train.py:892] (3/4) Epoch 15, batch 350, loss[loss=0.1932, simple_loss=0.2678, pruned_loss=0.05935, over 19655.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2719, pruned_loss=0.06996, over 3268377.90 frames. ], batch size: 67, lr: 1.06e-02, grad_scale: 16.0
+2023-03-28 06:42:17,606 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.83 vs. limit=5.0
+2023-03-28 06:42:49,752 INFO [train.py:892] (3/4) Epoch 15, batch 400, loss[loss=0.2177, simple_loss=0.284, pruned_loss=0.07571, over 19787.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.272, pruned_loss=0.07076, over 3419459.02 frames. ], batch size: 168, lr: 1.06e-02, grad_scale: 16.0
+2023-03-28 06:44:23,281 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.029e+02 4.352e+02 5.044e+02 6.255e+02 1.053e+03, threshold=1.009e+03, percent-clipped=1.0
+2023-03-28 06:44:28,717 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26414.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:44:30,776 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4695, 1.8564, 2.2498, 2.8903, 3.2473, 3.4102, 3.3730, 3.4957],
+       device='cuda:3'), covar=tensor([0.0941, 0.1885, 0.1383, 0.0582, 0.0367, 0.0256, 0.0298, 0.0277],
+       device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0171, 0.0169, 0.0134, 0.0117, 0.0111, 0.0108, 0.0101],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 06:44:43,045 INFO [train.py:892] (3/4) Epoch 15, batch 450, loss[loss=0.184, simple_loss=0.2412, pruned_loss=0.06344, over 19647.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2714, pruned_loss=0.07023, over 3538108.96 frames. ], batch size: 47, lr: 1.06e-02, grad_scale: 16.0
+2023-03-28 06:45:57,143 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3085, 3.8253, 4.0090, 4.3254, 3.8946, 4.3215, 4.4010, 4.5532],
+       device='cuda:3'), covar=tensor([0.0593, 0.0391, 0.0468, 0.0254, 0.0730, 0.0340, 0.0411, 0.0295],
+       device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0152, 0.0178, 0.0146, 0.0148, 0.0131, 0.0136, 0.0169],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 06:46:09,194 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:46:36,563 INFO [train.py:892] (3/4) Epoch 15, batch 500, loss[loss=0.2329, simple_loss=0.3023, pruned_loss=0.08178, over 19840.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2705, pruned_loss=0.0703, over 3630307.10 frames. ], batch size: 58, lr: 1.06e-02, grad_scale: 16.0
+2023-03-28 06:46:47,817 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:48:13,616 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.684e+02 4.387e+02 5.126e+02 6.696e+02 1.197e+03, threshold=1.025e+03, percent-clipped=1.0
+2023-03-28 06:48:17,063 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26514.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 06:48:30,963 INFO [train.py:892] (3/4) Epoch 15, batch 550, loss[loss=0.2043, simple_loss=0.2778, pruned_loss=0.06539, over 19931.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2701, pruned_loss=0.07023, over 3702343.03 frames. ], batch size: 51, lr: 1.06e-02, grad_scale: 16.0
+2023-03-28 06:48:40,955 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:48:47,425 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2973, 3.3064, 4.6122, 3.4734, 3.8980, 3.8658, 2.4699, 2.6624],
+       device='cuda:3'), covar=tensor([0.0698, 0.2611, 0.0455, 0.0763, 0.1382, 0.1005, 0.1959, 0.2280],
+       device='cuda:3'), in_proj_covar=tensor([0.0319, 0.0348, 0.0292, 0.0236, 0.0349, 0.0296, 0.0311, 0.0284],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 06:50:07,890 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26562.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 06:50:26,052 INFO [train.py:892] (3/4) Epoch 15, batch 600, loss[loss=0.1986, simple_loss=0.2617, pruned_loss=0.06781, over 19793.00 frames. ], tot_loss[loss=0.2051, simple_loss=0.2702, pruned_loss=0.07, over 3756433.86 frames. ], batch size: 241, lr: 1.06e-02, grad_scale: 16.0
+2023-03-28 06:51:45,269 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:52:00,971 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.578e+02 4.242e+02 5.300e+02 6.521e+02 1.210e+03, threshold=1.060e+03, percent-clipped=1.0
+2023-03-28 06:52:21,135 INFO [train.py:892] (3/4) Epoch 15, batch 650, loss[loss=0.2258, simple_loss=0.2806, pruned_loss=0.08551, over 19877.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2696, pruned_loss=0.06947, over 3797797.13 frames. ], batch size: 52, lr: 1.06e-02, grad_scale: 16.0
+2023-03-28 06:52:40,707 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0376, 3.9066, 3.8864, 3.6888, 4.0317, 3.0175, 3.3501, 1.9910],
+       device='cuda:3'), covar=tensor([0.0213, 0.0196, 0.0131, 0.0180, 0.0148, 0.0876, 0.0693, 0.1518],
+       device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0124, 0.0101, 0.0118, 0.0106, 0.0122, 0.0131, 0.0116],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 06:53:58,414 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0
+2023-03-28 06:54:11,814 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0
+2023-03-28 06:54:12,426 INFO [train.py:892] (3/4) Epoch 15, batch 700, loss[loss=0.2827, simple_loss=0.3321, pruned_loss=0.1166, over 19705.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2728, pruned_loss=0.07149, over 3829479.92 frames. ], batch size: 310, lr: 1.06e-02, grad_scale: 16.0
+2023-03-28 06:55:53,018 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.991e+02 4.486e+02 5.723e+02 6.688e+02 1.072e+03, threshold=1.145e+03, percent-clipped=1.0
+2023-03-28 06:56:12,166 INFO [train.py:892] (3/4) Epoch 15, batch 750, loss[loss=0.201, simple_loss=0.2648, pruned_loss=0.06856, over 19652.00 frames. ], tot_loss[loss=0.208, simple_loss=0.2731, pruned_loss=0.07142, over 3856532.24 frames. ], batch size: 43, lr: 1.06e-02, grad_scale: 16.0
+2023-03-28 06:57:19,110 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9260, 2.9783, 1.9012, 2.9547, 3.1066, 1.3622, 2.5697, 2.3106],
+       device='cuda:3'), covar=tensor([0.0775, 0.0784, 0.2496, 0.0709, 0.0452, 0.2545, 0.1002, 0.0884],
+       device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0230, 0.0216, 0.0225, 0.0192, 0.0197, 0.0223, 0.0171],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 06:57:31,285 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6828, 2.0244, 2.5828, 3.0294, 3.5680, 3.8527, 3.7783, 3.8168],
+       device='cuda:3'), covar=tensor([0.0883, 0.1866, 0.1302, 0.0579, 0.0336, 0.0194, 0.0258, 0.0277],
+       device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0171, 0.0169, 0.0135, 0.0117, 0.0110, 0.0108, 0.0102],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 06:57:34,856 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:58:01,891 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26770.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:58:03,008 INFO [train.py:892] (3/4) Epoch 15, batch 800, loss[loss=0.1974, simple_loss=0.2616, pruned_loss=0.06656, over 19842.00 frames. ], tot_loss[loss=0.2103, simple_loss=0.275, pruned_loss=0.07273, over 3876325.96 frames. ], batch size: 137, lr: 1.06e-02, grad_scale: 16.0
+2023-03-28 06:58:18,549 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0131, 3.0815, 1.6329, 3.7710, 3.4450, 3.8395, 3.7787, 2.9059],
+       device='cuda:3'), covar=tensor([0.0588, 0.0534, 0.1650, 0.0527, 0.0476, 0.0314, 0.0617, 0.0741],
+       device='cuda:3'), in_proj_covar=tensor([0.0127, 0.0124, 0.0131, 0.0128, 0.0112, 0.0106, 0.0122, 0.0128],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 06:59:23,180 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 06:59:36,950 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.366e+02 4.760e+02 5.593e+02 6.609e+02 1.473e+03, threshold=1.119e+03, percent-clipped=2.0
+2023-03-28 06:59:56,341 INFO [train.py:892] (3/4) Epoch 15, batch 850, loss[loss=0.1812, simple_loss=0.249, pruned_loss=0.05667, over 19924.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2748, pruned_loss=0.0723, over 3892671.02 frames. ], batch size: 45, lr: 1.05e-02, grad_scale: 16.0
+2023-03-28 07:00:03,367 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 07:01:17,049 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2275, 3.3143, 1.8247, 4.0638, 3.6714, 3.9789, 4.0883, 3.1769],
+       device='cuda:3'), covar=tensor([0.0604, 0.0574, 0.1674, 0.0577, 0.0497, 0.0369, 0.0550, 0.0736],
+       device='cuda:3'), in_proj_covar=tensor([0.0128, 0.0124, 0.0132, 0.0129, 0.0112, 0.0107, 0.0123, 0.0129],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 07:01:47,515 INFO [train.py:892] (3/4) Epoch 15, batch 900, loss[loss=0.2251, simple_loss=0.2894, pruned_loss=0.08037, over 19790.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2733, pruned_loss=0.07142, over 3905348.90 frames. ], batch size: 73, lr: 1.05e-02, grad_scale: 16.0
+2023-03-28 07:01:50,783 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 07:02:29,559 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0
+2023-03-28 07:02:41,668 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0
+2023-03-28 07:02:49,712 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.3635, 5.5585, 5.8633, 5.6487, 5.4998, 5.1437, 5.4322, 5.3548],
+       device='cuda:3'), covar=tensor([0.1403, 0.1024, 0.0868, 0.0986, 0.0704, 0.0950, 0.1875, 0.2091],
+       device='cuda:3'), in_proj_covar=tensor([0.0259, 0.0264, 0.0317, 0.0246, 0.0236, 0.0232, 0.0306, 0.0336],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 07:03:03,233 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26904.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 07:03:07,731 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 07:03:23,384 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.859e+02 4.358e+02 4.979e+02 5.962e+02 9.737e+02, threshold=9.958e+02, percent-clipped=0.0
+2023-03-28 07:03:36,915 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9133, 3.9216, 2.2658, 4.1111, 4.2206, 1.8236, 3.4602, 3.3425],
+       device='cuda:3'), covar=tensor([0.0622, 0.0754, 0.2724, 0.0755, 0.0531, 0.2934, 0.1062, 0.0696],
+       device='cuda:3'), in_proj_covar=tensor([0.0205, 0.0231, 0.0215, 0.0228, 0.0192, 0.0198, 0.0224, 0.0171],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 07:03:41,702 INFO [train.py:892] (3/4) Epoch 15, batch 950, loss[loss=0.2448, simple_loss=0.3007, pruned_loss=0.09451, over 19749.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2739, pruned_loss=0.07137, over 3916329.60 frames. ], batch size: 259, lr: 1.05e-02, grad_scale: 16.0
+2023-03-28 07:04:47,416 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 07:04:57,003 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26954.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 07:05:18,921 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26965.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 07:05:32,181 INFO [train.py:892] (3/4) Epoch 15, batch 1000, loss[loss=0.1957, simple_loss=0.266, pruned_loss=0.06274, over 19802.00 frames. ], tot_loss[loss=0.2083, simple_loss=0.2735, pruned_loss=0.07149, over 3923926.90 frames. ], batch size: 98, lr: 1.05e-02, grad_scale: 16.0
+2023-03-28 07:06:01,951 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2277, 4.3510, 2.3797, 4.6494, 4.8701, 1.9726, 4.2030, 3.3480],
+       device='cuda:3'), covar=tensor([0.0665, 0.0679, 0.2834, 0.0787, 0.0518, 0.2914, 0.0799, 0.0820],
+       device='cuda:3'), in_proj_covar=tensor([0.0206, 0.0232, 0.0215, 0.0228, 0.0193, 0.0198, 0.0225, 0.0172],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 07:06:48,624 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0
+2023-03-28 07:07:01,964 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 07:07:05,034 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.353e+02 4.723e+02 5.899e+02 7.505e+02 1.756e+03, threshold=1.180e+03, percent-clipped=5.0
+2023-03-28 07:07:24,355 INFO [train.py:892] (3/4) Epoch 15, batch 1050, loss[loss=0.1888, simple_loss=0.2565, pruned_loss=0.06051, over 19710.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2729, pruned_loss=0.0709, over 3929269.82 frames. ], batch size: 78, lr: 1.05e-02, grad_scale: 16.0
+2023-03-28 07:07:42,620 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0
+2023-03-28 07:09:11,957 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27070.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 07:09:13,196 INFO [train.py:892] (3/4) Epoch 15, batch 1100, loss[loss=0.1741, simple_loss=0.2494, pruned_loss=0.04942, over 19743.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2723, pruned_loss=0.07041, over 3935200.02 frames. ], batch size: 44, lr: 1.05e-02, grad_scale: 16.0
+2023-03-28 07:09:41,593 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0
+2023-03-28 07:10:49,060 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.986e+02 4.421e+02 5.325e+02 6.263e+02 1.209e+03, threshold=1.065e+03, percent-clipped=1.0
+2023-03-28 07:10:58,657 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.3103, 4.7566, 4.8796, 5.2452, 4.9125, 5.5334, 5.3191, 5.6074],
+       device='cuda:3'), covar=tensor([0.0593, 0.0345, 0.0393, 0.0256, 0.0529, 0.0271, 0.0352, 0.0236],
+       device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0156, 0.0181, 0.0151, 0.0152, 0.0135, 0.0140, 0.0173],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 07:11:00,535 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 07:11:06,891 INFO [train.py:892] (3/4) Epoch 15, batch 1150, loss[loss=0.1751, simple_loss=0.2506, pruned_loss=0.04982, over 19402.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.271, pruned_loss=0.07031, over 3938989.94 frames. ], batch size: 40, lr: 1.05e-02, grad_scale: 8.0
+2023-03-28 07:12:06,806 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 07:12:58,800 INFO [train.py:892] (3/4) Epoch 15, batch 1200, loss[loss=0.2037, simple_loss=0.2707, pruned_loss=0.06831, over 19756.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2723, pruned_loss=0.07076, over 3940531.36 frames. ], batch size: 100, lr: 1.05e-02, grad_scale: 8.0
+2023-03-28 07:13:08,740 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0
+2023-03-28 07:14:25,577 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 07:14:34,162 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.952e+02 4.645e+02 5.567e+02 6.439e+02 1.177e+03, threshold=1.113e+03, percent-clipped=3.0
+2023-03-28 07:14:49,426 INFO [train.py:892] (3/4) Epoch 15, batch 1250, loss[loss=0.2273, simple_loss=0.2884, pruned_loss=0.08313, over 19692.00 frames. ], tot_loss[loss=0.2066, simple_loss=0.2717, pruned_loss=0.07068, over 3944030.97 frames. ], batch size: 305, lr: 1.05e-02, grad_scale: 8.0
+2023-03-28 07:16:10,637 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0
+2023-03-28 07:16:16,257 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27260.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 07:16:40,193 INFO [train.py:892] (3/4) Epoch 15, batch 1300, loss[loss=0.2055, simple_loss=0.2688, pruned_loss=0.07108, over 19605.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2711, pruned_loss=0.0701, over 3943466.15 frames. ], batch size: 46, lr: 1.05e-02, grad_scale: 8.0
+2023-03-28 07:18:00,110 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 07:18:17,025 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.149e+02 4.572e+02 5.387e+02 6.367e+02 1.430e+03, threshold=1.077e+03, percent-clipped=1.0
+2023-03-28 07:18:32,961 INFO [train.py:892] (3/4) Epoch 15, batch 1350, loss[loss=0.2275, simple_loss=0.2926, pruned_loss=0.08123, over 19662.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2721, pruned_loss=0.07063, over 3945100.40 frames. ], batch size: 299, lr: 1.04e-02, grad_scale: 8.0
+2023-03-28 07:20:09,358 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0237, 3.3554, 3.4684, 4.2308, 2.7618, 3.3452, 2.5858, 2.4798],
+       device='cuda:3'), covar=tensor([0.0483, 0.1998, 0.0922, 0.0271, 0.2031, 0.0720, 0.1353, 0.1753],
+       device='cuda:3'), in_proj_covar=tensor([0.0211, 0.0327, 0.0228, 0.0165, 0.0237, 0.0182, 0.0202, 0.0208],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 07:20:21,498 INFO [train.py:892] (3/4) Epoch 15, batch 1400, loss[loss=0.1989, simple_loss=0.2624, pruned_loss=0.06765, over 19814.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.271, pruned_loss=0.07033, over 3947015.04 frames. ], batch size: 147, lr: 1.04e-02, grad_scale: 8.0
+2023-03-28 07:21:59,974 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.359e+02 4.655e+02 5.654e+02 6.548e+02 1.649e+03, threshold=1.131e+03, percent-clipped=3.0
+2023-03-28 07:22:06,439 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.68 vs. limit=5.0
+2023-03-28 07:22:14,687 INFO [train.py:892] (3/4) Epoch 15, batch 1450, loss[loss=0.2207, simple_loss=0.2852, pruned_loss=0.07808, over 19764.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2714, pruned_loss=0.07025, over 3949031.22 frames. ], batch size: 244, lr: 1.04e-02, grad_scale: 8.0
+2023-03-28 07:24:06,275 INFO [train.py:892] (3/4) Epoch 15, batch 1500, loss[loss=0.1967, simple_loss=0.2701, pruned_loss=0.0617, over 19816.00 frames. ], tot_loss[loss=0.2048, simple_loss=0.2703, pruned_loss=0.06967, over 3951017.87 frames. ], batch size: 50, lr: 1.04e-02, grad_scale: 8.0
+2023-03-28 07:25:13,755 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.1782, 2.2455, 2.4382, 2.2107, 2.0736, 2.2164, 2.1776, 2.5665],
+       device='cuda:3'), covar=tensor([0.0232, 0.0273, 0.0208, 0.0215, 0.0322, 0.0269, 0.0309, 0.0204],
+       device='cuda:3'), in_proj_covar=tensor([0.0055, 0.0054, 0.0056, 0.0049, 0.0061, 0.0057, 0.0073, 0.0051],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-28 07:25:20,006 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 07:25:41,666 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.670e+02 4.316e+02 5.110e+02 6.263e+02 1.157e+03, threshold=1.022e+03, percent-clipped=1.0
+2023-03-28 07:25:58,303 INFO [train.py:892] (3/4) Epoch 15, batch 1550, loss[loss=0.174, simple_loss=0.2446, pruned_loss=0.05166, over 19741.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2698, pruned_loss=0.06923, over 3949697.80 frames. ], batch size: 106, lr: 1.04e-02, grad_scale: 8.0
+2023-03-28 07:27:27,765 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27560.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 07:27:50,930 INFO [train.py:892] (3/4) Epoch 15, batch 1600, loss[loss=0.1912, simple_loss=0.2577, pruned_loss=0.06233, over 19811.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2694, pruned_loss=0.0684, over 3949461.15 frames. ], batch size: 72, lr: 1.04e-02, grad_scale: 8.0
+2023-03-28 07:29:11,841 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 07:29:15,493 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27608.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 07:29:28,188 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.222e+02 4.211e+02 5.223e+02 6.136e+02 1.197e+03, threshold=1.045e+03, percent-clipped=1.0
+2023-03-28 07:29:43,673 INFO [train.py:892] (3/4) Epoch 15, batch 1650, loss[loss=0.2516, simple_loss=0.3163, pruned_loss=0.09348, over 19852.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2699, pruned_loss=0.06856, over 3950271.36 frames. ], batch size: 81, lr: 1.04e-02, grad_scale: 8.0
+2023-03-28 07:30:27,612 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.2839, 2.4757, 2.6722, 2.4378, 2.2689, 2.5845, 2.3967, 2.7440],
+       device='cuda:3'), covar=tensor([0.0329, 0.0291, 0.0275, 0.0209, 0.0368, 0.0298, 0.0358, 0.0288],
+       device='cuda:3'), in_proj_covar=tensor([0.0055, 0.0054, 0.0057, 0.0049, 0.0062, 0.0058, 0.0075, 0.0052],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-28 07:30:47,299 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5399, 4.2811, 4.3788, 4.0629, 4.5474, 3.3057, 3.7639, 2.2686],
+       device='cuda:3'), covar=tensor([0.0185, 0.0193, 0.0121, 0.0190, 0.0121, 0.0679, 0.0719, 0.1328],
+       device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0126, 0.0102, 0.0120, 0.0107, 0.0123, 0.0134, 0.0117],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 07:31:01,155 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 07:31:24,223 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0
+2023-03-28 07:31:38,702 INFO [train.py:892] (3/4) Epoch 15, batch 1700, loss[loss=0.2022, simple_loss=0.2549, pruned_loss=0.07479, over 19746.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2708, pruned_loss=0.06902, over 3949436.79 frames. ], batch size: 140, lr: 1.04e-02, grad_scale: 8.0
+2023-03-28 07:32:27,212 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2269, 3.3918, 3.5005, 4.2934, 2.7183, 3.2511, 2.8510, 2.6462],
+       device='cuda:3'), covar=tensor([0.0500, 0.2355, 0.1053, 0.0333, 0.2329, 0.0897, 0.1327, 0.1849],
+       device='cuda:3'), in_proj_covar=tensor([0.0213, 0.0330, 0.0228, 0.0168, 0.0238, 0.0184, 0.0203, 0.0210],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 07:33:14,465 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.834e+02 4.579e+02 5.497e+02 6.695e+02 1.218e+03, threshold=1.099e+03, percent-clipped=2.0
+2023-03-28 07:33:15,310 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27714.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 07:33:27,524 INFO [train.py:892] (3/4) Epoch 15, batch 1750, loss[loss=0.1898, simple_loss=0.2636, pruned_loss=0.058, over 19868.00 frames. ], tot_loss[loss=0.2065, simple_loss=0.2724, pruned_loss=0.07028, over 3948074.89 frames. ], batch size: 48, lr: 1.04e-02, grad_scale: 8.0
+2023-03-28 07:35:04,131 INFO [train.py:892] (3/4) Epoch 15, batch 1800, loss[loss=0.2066, simple_loss=0.2768, pruned_loss=0.06818, over 19635.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.2719, pruned_loss=0.0699, over 3946805.43 frames. ], batch size: 68, lr: 1.04e-02, grad_scale: 8.0
+2023-03-28 07:35:11,037 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27775.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 07:36:04,547 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-28 07:36:05,816 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 07:36:14,526 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3807, 3.1882, 3.4491, 2.1672, 3.9975, 2.9142, 2.9465, 3.4875],
+       device='cuda:3'), covar=tensor([0.0424, 0.0321, 0.0436, 0.0964, 0.0196, 0.0379, 0.0438, 0.0324],
+       device='cuda:3'), in_proj_covar=tensor([0.0062, 0.0066, 0.0067, 0.0096, 0.0063, 0.0062, 0.0061, 0.0054],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 07:36:22,623 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.736e+02 4.432e+02 5.190e+02 6.215e+02 1.891e+03, threshold=1.038e+03, percent-clipped=1.0
+2023-03-28 07:36:34,461 INFO [train.py:892] (3/4) Epoch 15, batch 1850, loss[loss=0.1987, simple_loss=0.2793, pruned_loss=0.05907, over 19682.00 frames. ], tot_loss[loss=0.2078, simple_loss=0.2745, pruned_loss=0.07052, over 3945588.55 frames. ], batch size: 55, lr: 1.04e-02, grad_scale: 8.0
+2023-03-28 07:37:42,182 INFO [train.py:892] (3/4) Epoch 16, batch 0, loss[loss=0.2562, simple_loss=0.3134, pruned_loss=0.09947, over 19712.00 frames. ], tot_loss[loss=0.2562, simple_loss=0.3134, pruned_loss=0.09947, over 19712.00 frames. ], batch size: 310, lr: 1.00e-02, grad_scale: 8.0
+2023-03-28 07:37:42,182 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-28 07:38:12,989 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0596, 3.4699, 2.8995, 2.4126, 2.7804, 3.4142, 3.0492, 3.3069],
+       device='cuda:3'), covar=tensor([0.0296, 0.0195, 0.0244, 0.0428, 0.0344, 0.0233, 0.0173, 0.0173],
+       device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0070, 0.0077, 0.0083, 0.0087, 0.0060, 0.0058, 0.0060],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-28 07:38:14,624 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8908, 3.1993, 2.6050, 2.2155, 2.6225, 3.0226, 2.7744, 3.2154],
+       device='cuda:3'), covar=tensor([0.0137, 0.0249, 0.0242, 0.0493, 0.0330, 0.0205, 0.0199, 0.0069],
+       device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0070, 0.0077, 0.0083, 0.0087, 0.0060, 0.0058, 0.0060],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-28 07:38:15,255 INFO [train.py:926] (3/4) Epoch 16, validation: loss=0.1716, simple_loss=0.2504, pruned_loss=0.04639, over 2883724.00 frames. 
+2023-03-28 07:38:15,256 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-28 07:38:45,710 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.92 vs. limit=5.0
+2023-03-28 07:39:17,620 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 07:40:09,186 INFO [train.py:892] (3/4) Epoch 16, batch 50, loss[loss=0.195, simple_loss=0.2735, pruned_loss=0.05824, over 19832.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2665, pruned_loss=0.06848, over 890713.79 frames. ], batch size: 58, lr: 1.00e-02, grad_scale: 8.0
+2023-03-28 07:41:32,616 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.518e+02 4.210e+02 5.030e+02 5.753e+02 1.491e+03, threshold=1.006e+03, percent-clipped=2.0
+2023-03-28 07:42:00,015 INFO [train.py:892] (3/4) Epoch 16, batch 100, loss[loss=0.1781, simple_loss=0.2487, pruned_loss=0.05378, over 19907.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.266, pruned_loss=0.06707, over 1569969.60 frames. ], batch size: 116, lr: 1.00e-02, grad_scale: 8.0
+2023-03-28 07:42:22,885 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4311, 2.5422, 1.4494, 2.8958, 2.6325, 2.8224, 2.9406, 2.2382],
+       device='cuda:3'), covar=tensor([0.0697, 0.0625, 0.1501, 0.0484, 0.0662, 0.0443, 0.0522, 0.0910],
+       device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0126, 0.0135, 0.0131, 0.0114, 0.0111, 0.0126, 0.0132],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 07:42:46,016 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1470, 3.3818, 3.4354, 4.3618, 2.8687, 3.0921, 2.6965, 2.5721],
+       device='cuda:3'), covar=tensor([0.0526, 0.2376, 0.1047, 0.0278, 0.2099, 0.0925, 0.1316, 0.1859],
+       device='cuda:3'), in_proj_covar=tensor([0.0213, 0.0331, 0.0229, 0.0168, 0.0240, 0.0184, 0.0203, 0.0211],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 07:43:53,907 INFO [train.py:892] (3/4) Epoch 16, batch 150, loss[loss=0.2106, simple_loss=0.2942, pruned_loss=0.06354, over 19688.00 frames. ], tot_loss[loss=0.199, simple_loss=0.266, pruned_loss=0.06604, over 2098202.24 frames. ], batch size: 56, lr: 1.00e-02, grad_scale: 8.0
+2023-03-28 07:45:25,173 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.668e+02 4.765e+02 5.560e+02 7.224e+02 1.633e+03, threshold=1.112e+03, percent-clipped=5.0
+2023-03-28 07:45:44,418 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9552, 5.2845, 5.3041, 5.2222, 4.9153, 5.2867, 4.7176, 4.7952],
+       device='cuda:3'), covar=tensor([0.0459, 0.0462, 0.0559, 0.0441, 0.0587, 0.0533, 0.0646, 0.0928],
+       device='cuda:3'), in_proj_covar=tensor([0.0221, 0.0230, 0.0256, 0.0220, 0.0215, 0.0202, 0.0230, 0.0269],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 07:45:54,096 INFO [train.py:892] (3/4) Epoch 16, batch 200, loss[loss=0.256, simple_loss=0.3108, pruned_loss=0.1006, over 19751.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2697, pruned_loss=0.068, over 2507549.95 frames. ], batch size: 256, lr: 9.99e-03, grad_scale: 8.0
+2023-03-28 07:47:36,344 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28070.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 07:47:47,819 INFO [train.py:892] (3/4) Epoch 16, batch 250, loss[loss=0.2103, simple_loss=0.2833, pruned_loss=0.06868, over 19949.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2687, pruned_loss=0.0675, over 2828209.01 frames. ], batch size: 52, lr: 9.98e-03, grad_scale: 8.0
+2023-03-28 07:49:14,119 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.535e+02 4.468e+02 5.270e+02 6.395e+02 1.144e+03, threshold=1.054e+03, percent-clipped=2.0
+2023-03-28 07:49:40,346 INFO [train.py:892] (3/4) Epoch 16, batch 300, loss[loss=0.1785, simple_loss=0.2655, pruned_loss=0.04572, over 19811.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2703, pruned_loss=0.06842, over 3075689.10 frames. ], batch size: 50, lr: 9.97e-03, grad_scale: 8.0
+2023-03-28 07:49:44,193 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3265, 4.2932, 4.7378, 4.3401, 4.0739, 4.5608, 4.4384, 4.8395],
+       device='cuda:3'), covar=tensor([0.0924, 0.0346, 0.0333, 0.0340, 0.0787, 0.0449, 0.0376, 0.0265],
+       device='cuda:3'), in_proj_covar=tensor([0.0264, 0.0201, 0.0201, 0.0207, 0.0195, 0.0210, 0.0207, 0.0191],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 07:51:33,564 INFO [train.py:892] (3/4) Epoch 16, batch 350, loss[loss=0.1718, simple_loss=0.245, pruned_loss=0.04933, over 19685.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.2699, pruned_loss=0.06779, over 3269157.56 frames. ], batch size: 75, lr: 9.96e-03, grad_scale: 8.0
+2023-03-28 07:53:01,623 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.842e+02 4.257e+02 5.267e+02 6.673e+02 1.330e+03, threshold=1.053e+03, percent-clipped=2.0
+2023-03-28 07:53:27,017 INFO [train.py:892] (3/4) Epoch 16, batch 400, loss[loss=0.203, simple_loss=0.2749, pruned_loss=0.06551, over 19857.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.2696, pruned_loss=0.06784, over 3420292.08 frames. ], batch size: 142, lr: 9.95e-03, grad_scale: 8.0
+2023-03-28 07:53:35,164 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 07:54:36,293 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28255.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 07:54:41,322 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-28 07:55:21,932 INFO [train.py:892] (3/4) Epoch 16, batch 450, loss[loss=0.2121, simple_loss=0.2659, pruned_loss=0.07916, over 19829.00 frames. ], tot_loss[loss=0.203, simple_loss=0.27, pruned_loss=0.06806, over 3537301.64 frames. ], batch size: 202, lr: 9.95e-03, grad_scale: 8.0
+2023-03-28 07:55:27,749 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2412, 3.1371, 3.3880, 2.5553, 3.7449, 2.9216, 3.2200, 3.6699],
+       device='cuda:3'), covar=tensor([0.0785, 0.0398, 0.0566, 0.0766, 0.0250, 0.0408, 0.0480, 0.0392],
+       device='cuda:3'), in_proj_covar=tensor([0.0064, 0.0068, 0.0067, 0.0098, 0.0065, 0.0063, 0.0061, 0.0054],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 07:55:48,443 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2508, 3.1923, 3.5902, 3.2845, 3.1367, 3.5285, 3.3623, 3.6476],
+       device='cuda:3'), covar=tensor([0.1112, 0.0469, 0.0434, 0.0443, 0.1537, 0.0582, 0.0469, 0.0409],
+       device='cuda:3'), in_proj_covar=tensor([0.0263, 0.0202, 0.0200, 0.0206, 0.0194, 0.0210, 0.0208, 0.0191],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 07:55:48,527 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28287.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 07:55:55,234 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 07:56:47,446 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.970e+02 4.542e+02 5.285e+02 6.486e+02 1.059e+03, threshold=1.057e+03, percent-clipped=1.0
+2023-03-28 07:56:54,412 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28316.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 07:57:04,969 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0
+2023-03-28 07:57:14,803 INFO [train.py:892] (3/4) Epoch 16, batch 500, loss[loss=0.2025, simple_loss=0.2715, pruned_loss=0.06676, over 19737.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.27, pruned_loss=0.06823, over 3628271.98 frames. ], batch size: 205, lr: 9.94e-03, grad_scale: 8.0
+2023-03-28 07:57:17,905 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8138, 2.6129, 3.0465, 2.6845, 3.1923, 3.1227, 3.7179, 4.0753],
+       device='cuda:3'), covar=tensor([0.0590, 0.1720, 0.1499, 0.2007, 0.1511, 0.1375, 0.0539, 0.0406],
+       device='cuda:3'), in_proj_covar=tensor([0.0224, 0.0224, 0.0247, 0.0238, 0.0271, 0.0238, 0.0198, 0.0209],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 07:58:04,717 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 07:58:55,486 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28370.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 07:59:07,692 INFO [train.py:892] (3/4) Epoch 16, batch 550, loss[loss=0.2022, simple_loss=0.266, pruned_loss=0.06925, over 19875.00 frames. ], tot_loss[loss=0.2027, simple_loss=0.269, pruned_loss=0.06821, over 3699767.48 frames. ], batch size: 159, lr: 9.93e-03, grad_scale: 8.0
+2023-03-28 08:00:36,161 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.975e+02 4.219e+02 4.967e+02 6.198e+02 1.278e+03, threshold=9.934e+02, percent-clipped=3.0
+2023-03-28 08:00:46,625 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28418.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 08:00:46,758 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9209, 4.7215, 4.7468, 4.4150, 4.7960, 3.0679, 3.7026, 2.1738],
+       device='cuda:3'), covar=tensor([0.0358, 0.0260, 0.0280, 0.0308, 0.0426, 0.1039, 0.1305, 0.2328],
+       device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0126, 0.0101, 0.0121, 0.0107, 0.0123, 0.0132, 0.0118],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 08:01:02,579 INFO [train.py:892] (3/4) Epoch 16, batch 600, loss[loss=0.1879, simple_loss=0.2453, pruned_loss=0.06521, over 19840.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2691, pruned_loss=0.0684, over 3755654.93 frames. ], batch size: 160, lr: 9.92e-03, grad_scale: 8.0
+2023-03-28 08:02:53,443 INFO [train.py:892] (3/4) Epoch 16, batch 650, loss[loss=0.2037, simple_loss=0.2804, pruned_loss=0.06349, over 19700.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2668, pruned_loss=0.06724, over 3799091.19 frames. ], batch size: 56, lr: 9.91e-03, grad_scale: 8.0
+2023-03-28 08:03:17,496 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:03:31,454 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:03:33,416 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9380, 3.9279, 4.2900, 4.0389, 4.1794, 3.7000, 4.0507, 3.7574],
+       device='cuda:3'), covar=tensor([0.1398, 0.1422, 0.1032, 0.1243, 0.1189, 0.1163, 0.1807, 0.2276],
+       device='cuda:3'), in_proj_covar=tensor([0.0270, 0.0271, 0.0331, 0.0257, 0.0244, 0.0239, 0.0321, 0.0348],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 08:04:17,631 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.821e+02 4.448e+02 5.454e+02 6.222e+02 1.151e+03, threshold=1.091e+03, percent-clipped=4.0
+2023-03-28 08:04:21,212 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.02 vs. limit=5.0
+2023-03-28 08:04:44,538 INFO [train.py:892] (3/4) Epoch 16, batch 700, loss[loss=0.2049, simple_loss=0.2777, pruned_loss=0.06605, over 19728.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2672, pruned_loss=0.06737, over 3833827.04 frames. ], batch size: 50, lr: 9.90e-03, grad_scale: 8.0
+2023-03-28 08:05:04,055 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0
+2023-03-28 08:05:34,361 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28547.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:05:45,750 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:06:25,761 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0
+2023-03-28 08:06:37,717 INFO [train.py:892] (3/4) Epoch 16, batch 750, loss[loss=0.2156, simple_loss=0.2768, pruned_loss=0.07718, over 19779.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2675, pruned_loss=0.06759, over 3858425.32 frames. ], batch size: 247, lr: 9.89e-03, grad_scale: 8.0
+2023-03-28 08:06:59,802 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:07:59,385 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28611.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 08:08:06,521 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.992e+02 4.406e+02 5.087e+02 6.345e+02 1.381e+03, threshold=1.017e+03, percent-clipped=1.0
+2023-03-28 08:08:34,056 INFO [train.py:892] (3/4) Epoch 16, batch 800, loss[loss=0.1703, simple_loss=0.2303, pruned_loss=0.05512, over 19738.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2696, pruned_loss=0.06897, over 3877456.10 frames. ], batch size: 140, lr: 9.89e-03, grad_scale: 8.0
+2023-03-28 08:09:12,846 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:10:29,836 INFO [train.py:892] (3/4) Epoch 16, batch 850, loss[loss=0.1846, simple_loss=0.2497, pruned_loss=0.05972, over 19845.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2694, pruned_loss=0.0685, over 3893180.71 frames. ], batch size: 124, lr: 9.88e-03, grad_scale: 8.0
+2023-03-28 08:11:14,635 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0
+2023-03-28 08:11:56,546 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.530e+02 4.659e+02 5.536e+02 6.490e+02 1.175e+03, threshold=1.107e+03, percent-clipped=2.0
+2023-03-28 08:12:21,644 INFO [train.py:892] (3/4) Epoch 16, batch 900, loss[loss=0.2032, simple_loss=0.2786, pruned_loss=0.0639, over 19850.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2687, pruned_loss=0.06809, over 3905247.77 frames. ], batch size: 58, lr: 9.87e-03, grad_scale: 8.0
+2023-03-28 08:12:48,581 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.76 vs. limit=5.0
+2023-03-28 08:14:18,776 INFO [train.py:892] (3/4) Epoch 16, batch 950, loss[loss=0.1962, simple_loss=0.2513, pruned_loss=0.07052, over 19870.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2695, pruned_loss=0.06869, over 3915903.89 frames. ], batch size: 157, lr: 9.86e-03, grad_scale: 8.0
+2023-03-28 08:15:44,967 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.335e+02 4.634e+02 5.431e+02 6.353e+02 1.268e+03, threshold=1.086e+03, percent-clipped=1.0
+2023-03-28 08:16:11,367 INFO [train.py:892] (3/4) Epoch 16, batch 1000, loss[loss=0.2173, simple_loss=0.2897, pruned_loss=0.07249, over 19576.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2683, pruned_loss=0.06807, over 3924513.90 frames. ], batch size: 49, lr: 9.85e-03, grad_scale: 8.0
+2023-03-28 08:16:14,348 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1704, 2.4401, 2.5912, 3.0324, 2.1386, 2.7750, 1.9918, 2.0610],
+       device='cuda:3'), covar=tensor([0.0503, 0.1347, 0.1078, 0.0415, 0.2131, 0.0627, 0.1352, 0.1558],
+       device='cuda:3'), in_proj_covar=tensor([0.0215, 0.0335, 0.0232, 0.0174, 0.0240, 0.0187, 0.0205, 0.0211],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 08:16:45,748 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:16:59,638 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:17:31,433 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:18:01,602 INFO [train.py:892] (3/4) Epoch 16, batch 1050, loss[loss=0.1974, simple_loss=0.2709, pruned_loss=0.06196, over 19801.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2699, pruned_loss=0.06854, over 3928086.04 frames. ], batch size: 229, lr: 9.84e-03, grad_scale: 8.0
+2023-03-28 08:18:23,549 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:18:35,328 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0
+2023-03-28 08:19:23,618 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28911.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 08:19:23,924 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.90 vs. limit=5.0
+2023-03-28 08:19:30,254 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.700e+02 4.466e+02 5.126e+02 6.804e+02 1.153e+03, threshold=1.025e+03, percent-clipped=1.0
+2023-03-28 08:19:48,283 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:19:56,637 INFO [train.py:892] (3/4) Epoch 16, batch 1100, loss[loss=0.1914, simple_loss=0.2493, pruned_loss=0.06672, over 19833.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2703, pruned_loss=0.0687, over 3932508.05 frames. ], batch size: 146, lr: 9.84e-03, grad_scale: 8.0
+2023-03-28 08:20:14,252 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:20:37,596 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28943.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:21:12,749 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28959.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 08:21:48,728 INFO [train.py:892] (3/4) Epoch 16, batch 1150, loss[loss=0.2186, simple_loss=0.2972, pruned_loss=0.07004, over 19729.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2698, pruned_loss=0.06851, over 3936590.46 frames. ], batch size: 52, lr: 9.83e-03, grad_scale: 8.0
+2023-03-28 08:22:22,205 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0
+2023-03-28 08:22:23,821 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28991.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:22:51,081 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:23:15,271 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.211e+02 4.550e+02 5.400e+02 6.345e+02 1.264e+03, threshold=1.080e+03, percent-clipped=1.0
+2023-03-28 08:23:38,327 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-28 08:23:43,700 INFO [train.py:892] (3/4) Epoch 16, batch 1200, loss[loss=0.172, simple_loss=0.2432, pruned_loss=0.05045, over 19673.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2708, pruned_loss=0.06877, over 3939234.07 frames. ], batch size: 49, lr: 9.82e-03, grad_scale: 8.0
+2023-03-28 08:25:09,581 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:25:38,185 INFO [train.py:892] (3/4) Epoch 16, batch 1250, loss[loss=0.1644, simple_loss=0.232, pruned_loss=0.04838, over 19736.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2706, pruned_loss=0.06884, over 3940475.50 frames. ], batch size: 99, lr: 9.81e-03, grad_scale: 8.0
+2023-03-28 08:26:52,499 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9127, 4.8515, 5.3396, 4.8494, 4.2825, 5.2274, 4.9864, 5.5460],
+       device='cuda:3'), covar=tensor([0.0947, 0.0349, 0.0346, 0.0367, 0.0698, 0.0410, 0.0420, 0.0296],
+       device='cuda:3'), in_proj_covar=tensor([0.0261, 0.0202, 0.0201, 0.0208, 0.0196, 0.0210, 0.0207, 0.0190],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 08:27:06,757 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.820e+02 4.171e+02 4.962e+02 5.864e+02 1.336e+03, threshold=9.924e+02, percent-clipped=4.0
+2023-03-28 08:27:33,971 INFO [train.py:892] (3/4) Epoch 16, batch 1300, loss[loss=0.1886, simple_loss=0.2576, pruned_loss=0.05976, over 19882.00 frames. ], tot_loss[loss=0.2032, simple_loss=0.2698, pruned_loss=0.06834, over 3942437.26 frames. ], batch size: 88, lr: 9.80e-03, grad_scale: 16.0
+2023-03-28 08:28:12,079 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:28:16,752 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.37 vs. limit=5.0
+2023-03-28 08:28:24,941 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:28:32,117 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.69 vs. limit=5.0
+2023-03-28 08:29:27,194 INFO [train.py:892] (3/4) Epoch 16, batch 1350, loss[loss=0.1753, simple_loss=0.2481, pruned_loss=0.05123, over 19806.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2705, pruned_loss=0.06851, over 3943163.09 frames. ], batch size: 117, lr: 9.80e-03, grad_scale: 16.0
+2023-03-28 08:30:01,012 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:30:15,729 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:30:55,172 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.081e+02 4.385e+02 5.390e+02 6.526e+02 1.400e+03, threshold=1.078e+03, percent-clipped=3.0
+2023-03-28 08:31:03,473 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:31:23,385 INFO [train.py:892] (3/4) Epoch 16, batch 1400, loss[loss=0.1758, simple_loss=0.242, pruned_loss=0.05477, over 19886.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2702, pruned_loss=0.06846, over 3944222.47 frames. ], batch size: 87, lr: 9.79e-03, grad_scale: 16.0
+2023-03-28 08:31:47,885 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:33:17,618 INFO [train.py:892] (3/4) Epoch 16, batch 1450, loss[loss=0.3064, simple_loss=0.3657, pruned_loss=0.1235, over 19409.00 frames. ], tot_loss[loss=0.2031, simple_loss=0.2701, pruned_loss=0.06808, over 3945850.31 frames. ], batch size: 412, lr: 9.78e-03, grad_scale: 16.0
+2023-03-28 08:34:08,535 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:34:43,970 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.554e+02 4.411e+02 5.175e+02 6.101e+02 1.424e+03, threshold=1.035e+03, percent-clipped=1.0
+2023-03-28 08:35:11,604 INFO [train.py:892] (3/4) Epoch 16, batch 1500, loss[loss=0.1916, simple_loss=0.2534, pruned_loss=0.06492, over 19840.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2684, pruned_loss=0.06698, over 3948149.92 frames. ], batch size: 160, lr: 9.77e-03, grad_scale: 16.0
+2023-03-28 08:36:25,410 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:37:05,542 INFO [train.py:892] (3/4) Epoch 16, batch 1550, loss[loss=0.1816, simple_loss=0.2493, pruned_loss=0.05692, over 19780.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2683, pruned_loss=0.0674, over 3947618.37 frames. ], batch size: 215, lr: 9.76e-03, grad_scale: 16.0
+2023-03-28 08:38:32,272 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.261e+02 4.365e+02 5.019e+02 6.130e+02 1.225e+03, threshold=1.004e+03, percent-clipped=3.0
+2023-03-28 08:39:00,864 INFO [train.py:892] (3/4) Epoch 16, batch 1600, loss[loss=0.1885, simple_loss=0.262, pruned_loss=0.0575, over 19847.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2688, pruned_loss=0.06769, over 3947859.77 frames. ], batch size: 59, lr: 9.76e-03, grad_scale: 16.0
+2023-03-28 08:40:51,220 INFO [train.py:892] (3/4) Epoch 16, batch 1650, loss[loss=0.1862, simple_loss=0.2493, pruned_loss=0.06154, over 19861.00 frames. ], tot_loss[loss=0.2024, simple_loss=0.2695, pruned_loss=0.0676, over 3948028.34 frames. ], batch size: 157, lr: 9.75e-03, grad_scale: 16.0
+2023-03-28 08:41:00,096 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0073, 3.0132, 1.8286, 3.6145, 3.2931, 3.5566, 3.6328, 2.8318],
+       device='cuda:3'), covar=tensor([0.0632, 0.0615, 0.1686, 0.0571, 0.0555, 0.0449, 0.0558, 0.0749],
+       device='cuda:3'), in_proj_covar=tensor([0.0132, 0.0129, 0.0137, 0.0134, 0.0117, 0.0115, 0.0130, 0.0133],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 08:42:19,142 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.587e+02 4.298e+02 4.820e+02 5.850e+02 1.066e+03, threshold=9.641e+02, percent-clipped=1.0
+2023-03-28 08:42:27,250 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29517.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:42:45,630 INFO [train.py:892] (3/4) Epoch 16, batch 1700, loss[loss=0.1836, simple_loss=0.2431, pruned_loss=0.06205, over 19611.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2694, pruned_loss=0.06737, over 3947022.91 frames. ], batch size: 46, lr: 9.74e-03, grad_scale: 16.0
+2023-03-28 08:43:26,586 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6369, 2.9580, 2.9709, 3.6391, 2.4837, 3.0750, 2.3708, 2.2275],
+       device='cuda:3'), covar=tensor([0.0507, 0.1911, 0.1109, 0.0348, 0.2155, 0.0698, 0.1346, 0.1807],
+       device='cuda:3'), in_proj_covar=tensor([0.0218, 0.0333, 0.0234, 0.0173, 0.0242, 0.0189, 0.0205, 0.0210],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 08:43:45,440 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:44:15,333 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:44:35,362 INFO [train.py:892] (3/4) Epoch 16, batch 1750, loss[loss=0.2051, simple_loss=0.2742, pruned_loss=0.06798, over 19428.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2683, pruned_loss=0.06654, over 3946489.45 frames. ], batch size: 40, lr: 9.73e-03, grad_scale: 16.0
+2023-03-28 08:44:49,521 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3649, 3.0180, 3.2675, 3.0830, 3.5614, 3.6335, 4.2984, 4.7229],
+       device='cuda:3'), covar=tensor([0.0535, 0.1654, 0.1496, 0.2095, 0.1642, 0.1299, 0.0485, 0.0400],
+       device='cuda:3'), in_proj_covar=tensor([0.0221, 0.0221, 0.0244, 0.0235, 0.0268, 0.0233, 0.0195, 0.0210],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 08:45:10,845 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:45:33,430 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3147, 3.0323, 4.9243, 4.1862, 4.5970, 4.8089, 4.7048, 4.3673],
+       device='cuda:3'), covar=tensor([0.0271, 0.0777, 0.0085, 0.0811, 0.0110, 0.0179, 0.0112, 0.0144],
+       device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0094, 0.0077, 0.0148, 0.0072, 0.0086, 0.0079, 0.0073],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0002, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 08:45:50,208 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:45:52,890 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.648e+02 4.260e+02 5.446e+02 6.419e+02 1.506e+03, threshold=1.089e+03, percent-clipped=4.0
+2023-03-28 08:46:14,301 INFO [train.py:892] (3/4) Epoch 16, batch 1800, loss[loss=0.1962, simple_loss=0.2651, pruned_loss=0.0637, over 19764.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2679, pruned_loss=0.06676, over 3947473.36 frames. ], batch size: 244, lr: 9.72e-03, grad_scale: 16.0
+2023-03-28 08:46:19,049 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4839, 3.1145, 3.5341, 2.4146, 3.6444, 2.7684, 3.0851, 3.6939],
+       device='cuda:3'), covar=tensor([0.0462, 0.0370, 0.0466, 0.0953, 0.0382, 0.0423, 0.0383, 0.0250],
+       device='cuda:3'), in_proj_covar=tensor([0.0064, 0.0068, 0.0067, 0.0098, 0.0066, 0.0064, 0.0062, 0.0055],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 08:47:16,108 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:47:23,448 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:47:46,316 INFO [train.py:892] (3/4) Epoch 16, batch 1850, loss[loss=0.1833, simple_loss=0.2573, pruned_loss=0.05469, over 19824.00 frames. ], tot_loss[loss=0.201, simple_loss=0.269, pruned_loss=0.06654, over 3947679.10 frames. ], batch size: 57, lr: 9.72e-03, grad_scale: 16.0
+2023-03-28 08:48:53,440 INFO [train.py:892] (3/4) Epoch 17, batch 0, loss[loss=0.168, simple_loss=0.2454, pruned_loss=0.04529, over 19872.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2454, pruned_loss=0.04529, over 19872.00 frames. ], batch size: 108, lr: 9.42e-03, grad_scale: 16.0
+2023-03-28 08:48:53,441 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-28 08:49:26,238 INFO [train.py:926] (3/4) Epoch 17, validation: loss=0.1709, simple_loss=0.2495, pruned_loss=0.0462, over 2883724.00 frames. 
+2023-03-28 08:49:26,239 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-28 08:50:27,532 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:50:46,147 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.983e+02 4.092e+02 5.169e+02 6.324e+02 1.457e+03, threshold=1.034e+03, percent-clipped=3.0
+2023-03-28 08:51:08,853 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29723.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 08:51:25,302 INFO [train.py:892] (3/4) Epoch 17, batch 50, loss[loss=0.1885, simple_loss=0.2564, pruned_loss=0.0603, over 19875.00 frames. ], tot_loss[loss=0.193, simple_loss=0.2606, pruned_loss=0.06267, over 889970.94 frames. ], batch size: 89, lr: 9.41e-03, grad_scale: 16.0
+2023-03-28 08:53:21,579 INFO [train.py:892] (3/4) Epoch 17, batch 100, loss[loss=0.2681, simple_loss=0.3621, pruned_loss=0.08708, over 17919.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2683, pruned_loss=0.06657, over 1565586.17 frames. ], batch size: 633, lr: 9.41e-03, grad_scale: 16.0
+2023-03-28 08:54:13,406 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29802.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:54:32,840 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4052, 3.4812, 2.0704, 3.5101, 3.6318, 1.6420, 3.0296, 2.8639],
+       device='cuda:3'), covar=tensor([0.0684, 0.0735, 0.2659, 0.0671, 0.0473, 0.2714, 0.0981, 0.0700],
+       device='cuda:3'), in_proj_covar=tensor([0.0209, 0.0232, 0.0216, 0.0231, 0.0202, 0.0199, 0.0225, 0.0172],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 08:54:38,192 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.079e+02 4.290e+02 5.009e+02 6.167e+02 1.031e+03, threshold=1.002e+03, percent-clipped=0.0
+2023-03-28 08:55:00,938 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0547, 4.6662, 4.8456, 4.4940, 5.0089, 3.3442, 4.0481, 2.7120],
+       device='cuda:3'), covar=tensor([0.0167, 0.0196, 0.0115, 0.0174, 0.0114, 0.0745, 0.0801, 0.1300],
+       device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0127, 0.0102, 0.0122, 0.0108, 0.0122, 0.0135, 0.0120],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 08:55:19,240 INFO [train.py:892] (3/4) Epoch 17, batch 150, loss[loss=0.2001, simple_loss=0.2674, pruned_loss=0.06633, over 19724.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2673, pruned_loss=0.06584, over 2093498.76 frames. ], batch size: 63, lr: 9.40e-03, grad_scale: 16.0
+2023-03-28 08:55:28,810 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.2121, 2.7239, 2.3326, 1.7555, 2.3771, 2.6266, 2.6000, 2.6323],
+       device='cuda:3'), covar=tensor([0.0317, 0.0237, 0.0265, 0.0534, 0.0387, 0.0233, 0.0204, 0.0191],
+       device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0072, 0.0080, 0.0085, 0.0089, 0.0063, 0.0060, 0.0063],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-28 08:55:38,810 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2447, 3.9009, 4.0410, 4.2496, 3.9040, 4.2373, 4.3762, 4.4983],
+       device='cuda:3'), covar=tensor([0.0591, 0.0316, 0.0417, 0.0302, 0.0684, 0.0382, 0.0336, 0.0260],
+       device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0155, 0.0179, 0.0151, 0.0151, 0.0132, 0.0134, 0.0172],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 08:56:29,426 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29863.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:57:07,849 INFO [train.py:892] (3/4) Epoch 17, batch 200, loss[loss=0.2133, simple_loss=0.2752, pruned_loss=0.07568, over 19773.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2683, pruned_loss=0.06628, over 2505619.19 frames. ], batch size: 263, lr: 9.39e-03, grad_scale: 16.0
+2023-03-28 08:57:33,277 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:58:05,740 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29907.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 08:58:19,264 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.430e+02 4.252e+02 5.177e+02 6.087e+02 1.116e+03, threshold=1.035e+03, percent-clipped=2.0
+2023-03-28 08:58:59,506 INFO [train.py:892] (3/4) Epoch 17, batch 250, loss[loss=0.1958, simple_loss=0.2703, pruned_loss=0.06068, over 19665.00 frames. ], tot_loss[loss=0.2017, simple_loss=0.269, pruned_loss=0.06716, over 2825433.19 frames. ], batch size: 50, lr: 9.38e-03, grad_scale: 16.0
+2023-03-28 08:59:22,778 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:00:54,570 INFO [train.py:892] (3/4) Epoch 17, batch 300, loss[loss=0.1958, simple_loss=0.269, pruned_loss=0.06133, over 19854.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2699, pruned_loss=0.06691, over 3073470.31 frames. ], batch size: 112, lr: 9.37e-03, grad_scale: 16.0
+2023-03-28 09:01:27,128 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3601, 4.6758, 4.6931, 4.5780, 4.3049, 4.6080, 4.1425, 4.1513],
+       device='cuda:3'), covar=tensor([0.0455, 0.0432, 0.0542, 0.0459, 0.0635, 0.0556, 0.0682, 0.0913],
+       device='cuda:3'), in_proj_covar=tensor([0.0223, 0.0231, 0.0259, 0.0222, 0.0217, 0.0209, 0.0231, 0.0272],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 09:02:13,033 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.673e+02 4.129e+02 4.962e+02 6.041e+02 9.266e+02, threshold=9.924e+02, percent-clipped=0.0
+2023-03-28 09:02:24,238 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30018.0, num_to_drop=1, layers_to_drop={3}
+2023-03-28 09:02:53,706 INFO [train.py:892] (3/4) Epoch 17, batch 350, loss[loss=0.1836, simple_loss=0.255, pruned_loss=0.05612, over 19863.00 frames. ], tot_loss[loss=0.2022, simple_loss=0.2703, pruned_loss=0.06707, over 3267759.94 frames. ], batch size: 106, lr: 9.37e-03, grad_scale: 16.0
+2023-03-28 09:04:44,399 INFO [train.py:892] (3/4) Epoch 17, batch 400, loss[loss=0.1847, simple_loss=0.2503, pruned_loss=0.0595, over 19817.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2694, pruned_loss=0.06704, over 3419647.59 frames. ], batch size: 143, lr: 9.36e-03, grad_scale: 16.0
+2023-03-28 09:04:52,302 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-28 09:05:47,588 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.75 vs. limit=5.0
+2023-03-28 09:05:59,370 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.983e+02 4.154e+02 5.101e+02 6.100e+02 1.061e+03, threshold=1.020e+03, percent-clipped=2.0
+2023-03-28 09:06:25,977 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:06:38,292 INFO [train.py:892] (3/4) Epoch 17, batch 450, loss[loss=0.1884, simple_loss=0.26, pruned_loss=0.05839, over 19766.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2681, pruned_loss=0.06643, over 3538239.31 frames. ], batch size: 116, lr: 9.35e-03, grad_scale: 16.0
+2023-03-28 09:07:29,893 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1565, 2.4505, 3.1985, 2.8096, 2.8867, 2.9356, 1.9394, 2.0900],
+       device='cuda:3'), covar=tensor([0.0874, 0.2309, 0.0590, 0.0762, 0.1380, 0.1029, 0.2015, 0.2094],
+       device='cuda:3'), in_proj_covar=tensor([0.0329, 0.0357, 0.0306, 0.0246, 0.0353, 0.0315, 0.0324, 0.0294],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 09:07:39,173 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1122, 3.4272, 3.4331, 4.1916, 2.6854, 3.2636, 2.7202, 2.5639],
+       device='cuda:3'), covar=tensor([0.0450, 0.2076, 0.0915, 0.0286, 0.2173, 0.0717, 0.1271, 0.1700],
+       device='cuda:3'), in_proj_covar=tensor([0.0218, 0.0334, 0.0233, 0.0174, 0.0241, 0.0188, 0.0206, 0.0211],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 09:07:40,918 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:08:32,545 INFO [train.py:892] (3/4) Epoch 17, batch 500, loss[loss=0.192, simple_loss=0.2557, pruned_loss=0.06418, over 19782.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2681, pruned_loss=0.06653, over 3630299.78 frames. ], batch size: 198, lr: 9.34e-03, grad_scale: 16.0
+2023-03-28 09:08:44,009 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:08:54,411 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0
+2023-03-28 09:09:31,042 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30207.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:09:35,220 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:09:45,652 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.879e+02 4.496e+02 5.220e+02 6.735e+02 1.165e+03, threshold=1.044e+03, percent-clipped=2.0
+2023-03-28 09:10:25,873 INFO [train.py:892] (3/4) Epoch 17, batch 550, loss[loss=0.2006, simple_loss=0.2564, pruned_loss=0.07244, over 19833.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2684, pruned_loss=0.06738, over 3700389.24 frames. ], batch size: 127, lr: 9.34e-03, grad_scale: 16.0
+2023-03-28 09:11:19,689 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30255.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:11:53,905 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:12:01,980 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.9630, 2.0809, 2.1630, 2.0684, 2.0529, 2.0506, 2.0716, 2.1699],
+       device='cuda:3'), covar=tensor([0.0231, 0.0242, 0.0235, 0.0244, 0.0358, 0.0269, 0.0361, 0.0261],
+       device='cuda:3'), in_proj_covar=tensor([0.0059, 0.0056, 0.0061, 0.0053, 0.0066, 0.0062, 0.0079, 0.0054],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-28 09:12:19,495 INFO [train.py:892] (3/4) Epoch 17, batch 600, loss[loss=0.2063, simple_loss=0.268, pruned_loss=0.07233, over 19810.00 frames. ], tot_loss[loss=0.2012, simple_loss=0.2679, pruned_loss=0.06723, over 3756984.76 frames. ], batch size: 181, lr: 9.33e-03, grad_scale: 16.0
+2023-03-28 09:13:33,410 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.677e+02 4.467e+02 5.041e+02 6.108e+02 1.228e+03, threshold=1.008e+03, percent-clipped=2.0
+2023-03-28 09:13:44,476 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30318.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 09:14:12,614 INFO [train.py:892] (3/4) Epoch 17, batch 650, loss[loss=0.1825, simple_loss=0.2484, pruned_loss=0.05832, over 19865.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2667, pruned_loss=0.06644, over 3800182.82 frames. ], batch size: 157, lr: 9.32e-03, grad_scale: 16.0
+2023-03-28 09:15:18,862 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:15:32,421 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30366.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:16:03,582 INFO [train.py:892] (3/4) Epoch 17, batch 700, loss[loss=0.1798, simple_loss=0.2524, pruned_loss=0.05364, over 19837.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2657, pruned_loss=0.06609, over 3835067.60 frames. ], batch size: 43, lr: 9.31e-03, grad_scale: 16.0
+2023-03-28 09:17:22,246 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.759e+02 4.180e+02 5.300e+02 6.975e+02 1.622e+03, threshold=1.060e+03, percent-clipped=3.0
+2023-03-28 09:17:39,043 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:17:59,219 INFO [train.py:892] (3/4) Epoch 17, batch 750, loss[loss=0.1933, simple_loss=0.2572, pruned_loss=0.06474, over 19858.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2656, pruned_loss=0.06561, over 3861820.12 frames. ], batch size: 46, lr: 9.31e-03, grad_scale: 16.0
+2023-03-28 09:18:38,894 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0
+2023-03-28 09:18:48,654 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.1216, 2.5240, 2.1411, 1.6379, 2.2691, 2.3649, 2.3222, 2.4589],
+       device='cuda:3'), covar=tensor([0.0248, 0.0215, 0.0277, 0.0540, 0.0323, 0.0235, 0.0200, 0.0202],
+       device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0071, 0.0080, 0.0084, 0.0087, 0.0063, 0.0060, 0.0062],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001],
+       device='cuda:3')
+2023-03-28 09:19:02,417 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:19:53,024 INFO [train.py:892] (3/4) Epoch 17, batch 800, loss[loss=0.2233, simple_loss=0.2774, pruned_loss=0.08454, over 19858.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2667, pruned_loss=0.06566, over 3878207.19 frames. ], batch size: 165, lr: 9.30e-03, grad_scale: 16.0
+2023-03-28 09:19:54,177 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:20:52,605 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:21:10,975 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.987e+02 4.028e+02 4.946e+02 5.803e+02 1.096e+03, threshold=9.891e+02, percent-clipped=1.0
+2023-03-28 09:21:23,829 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0
+2023-03-28 09:21:46,471 INFO [train.py:892] (3/4) Epoch 17, batch 850, loss[loss=0.2029, simple_loss=0.2651, pruned_loss=0.0703, over 19836.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2669, pruned_loss=0.06555, over 3894406.97 frames. ], batch size: 171, lr: 9.29e-03, grad_scale: 16.0
+2023-03-28 09:23:05,880 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:23:40,246 INFO [train.py:892] (3/4) Epoch 17, batch 900, loss[loss=0.2099, simple_loss=0.2802, pruned_loss=0.0698, over 19832.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2669, pruned_loss=0.0656, over 3907199.30 frames. ], batch size: 204, lr: 9.28e-03, grad_scale: 16.0
+2023-03-28 09:24:54,339 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.960e+02 4.086e+02 5.123e+02 6.200e+02 1.238e+03, threshold=1.025e+03, percent-clipped=1.0
+2023-03-28 09:25:34,346 INFO [train.py:892] (3/4) Epoch 17, batch 950, loss[loss=0.1917, simple_loss=0.2574, pruned_loss=0.06298, over 19888.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2672, pruned_loss=0.06552, over 3916844.09 frames. ], batch size: 176, lr: 9.28e-03, grad_scale: 16.0
+2023-03-28 09:26:28,472 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:27:25,990 INFO [train.py:892] (3/4) Epoch 17, batch 1000, loss[loss=0.2013, simple_loss=0.2636, pruned_loss=0.06954, over 19806.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2666, pruned_loss=0.06547, over 3925392.95 frames. ], batch size: 148, lr: 9.27e-03, grad_scale: 16.0
+2023-03-28 09:28:28,139 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0
+2023-03-28 09:28:42,762 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.493e+02 4.412e+02 5.308e+02 6.255e+02 1.077e+03, threshold=1.062e+03, percent-clipped=1.0
+2023-03-28 09:28:48,020 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:28:48,187 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30716.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 09:29:20,839 INFO [train.py:892] (3/4) Epoch 17, batch 1050, loss[loss=0.1965, simple_loss=0.2661, pruned_loss=0.06347, over 19771.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2652, pruned_loss=0.06463, over 3931219.96 frames. ], batch size: 108, lr: 9.26e-03, grad_scale: 16.0
+2023-03-28 09:31:15,319 INFO [train.py:892] (3/4) Epoch 17, batch 1100, loss[loss=0.2138, simple_loss=0.2719, pruned_loss=0.07792, over 19756.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2648, pruned_loss=0.06394, over 3934910.52 frames. ], batch size: 256, lr: 9.26e-03, grad_scale: 16.0
+2023-03-28 09:31:16,150 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30781.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:32:31,408 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.889e+02 4.403e+02 5.206e+02 6.352e+02 1.143e+03, threshold=1.041e+03, percent-clipped=2.0
+2023-03-28 09:33:05,662 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30829.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:33:08,525 INFO [train.py:892] (3/4) Epoch 17, batch 1150, loss[loss=0.1893, simple_loss=0.2407, pruned_loss=0.06899, over 19840.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.265, pruned_loss=0.06435, over 3938026.80 frames. ], batch size: 190, lr: 9.25e-03, grad_scale: 16.0
+2023-03-28 09:34:17,723 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.0331, 2.1298, 2.4792, 2.2079, 2.1379, 2.1386, 2.1810, 2.4399],
+       device='cuda:3'), covar=tensor([0.0270, 0.0226, 0.0207, 0.0214, 0.0323, 0.0292, 0.0378, 0.0211],
+       device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0056, 0.0060, 0.0053, 0.0067, 0.0062, 0.0079, 0.0054],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-28 09:34:25,754 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:35:00,471 INFO [train.py:892] (3/4) Epoch 17, batch 1200, loss[loss=0.1747, simple_loss=0.2518, pruned_loss=0.04884, over 19783.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.2648, pruned_loss=0.06367, over 3941825.06 frames. ], batch size: 91, lr: 9.24e-03, grad_scale: 16.0
+2023-03-28 09:35:30,347 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7745, 2.1149, 3.4547, 2.8953, 3.3785, 3.4984, 3.2136, 3.3908],
+       device='cuda:3'), covar=tensor([0.0484, 0.0963, 0.0108, 0.0557, 0.0132, 0.0216, 0.0208, 0.0163],
+       device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0095, 0.0077, 0.0148, 0.0073, 0.0086, 0.0081, 0.0073],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0002, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 09:35:50,836 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:36:17,103 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30913.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:36:18,325 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.660e+02 3.834e+02 4.757e+02 6.314e+02 9.786e+02, threshold=9.513e+02, percent-clipped=0.0
+2023-03-28 09:36:38,740 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4992, 3.6288, 2.2366, 3.7119, 3.8774, 1.7709, 3.1550, 2.8746],
+       device='cuda:3'), covar=tensor([0.0740, 0.0790, 0.2601, 0.0734, 0.0518, 0.2791, 0.1120, 0.0794],
+       device='cuda:3'), in_proj_covar=tensor([0.0212, 0.0236, 0.0217, 0.0235, 0.0208, 0.0200, 0.0229, 0.0175],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 09:36:42,604 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:36:53,364 INFO [train.py:892] (3/4) Epoch 17, batch 1250, loss[loss=0.193, simple_loss=0.2461, pruned_loss=0.06996, over 19824.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2661, pruned_loss=0.06474, over 3943882.06 frames. ], batch size: 143, lr: 9.23e-03, grad_scale: 16.0
+2023-03-28 09:37:01,733 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:38:10,827 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:38:45,966 INFO [train.py:892] (3/4) Epoch 17, batch 1300, loss[loss=0.1672, simple_loss=0.2404, pruned_loss=0.04698, over 19399.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2663, pruned_loss=0.06499, over 3945275.79 frames. ], batch size: 40, lr: 9.23e-03, grad_scale: 16.0
+2023-03-28 09:39:00,950 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:39:22,303 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:39:56,580 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31011.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 09:39:56,733 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31011.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 09:40:01,497 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.052e+02 4.725e+02 5.560e+02 6.623e+02 9.875e+02, threshold=1.112e+03, percent-clipped=1.0
+2023-03-28 09:40:06,323 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:40:39,845 INFO [train.py:892] (3/4) Epoch 17, batch 1350, loss[loss=0.185, simple_loss=0.2489, pruned_loss=0.06059, over 19838.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2651, pruned_loss=0.06438, over 3947070.21 frames. ], batch size: 144, lr: 9.22e-03, grad_scale: 16.0
+2023-03-28 09:41:54,685 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:42:13,100 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31072.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 09:42:31,167 INFO [train.py:892] (3/4) Epoch 17, batch 1400, loss[loss=0.1876, simple_loss=0.2525, pruned_loss=0.06131, over 19800.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2648, pruned_loss=0.06458, over 3946240.81 frames. ], batch size: 200, lr: 9.21e-03, grad_scale: 16.0
+2023-03-28 09:43:40,799 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.284e+02 4.073e+02 5.032e+02 6.013e+02 1.066e+03, threshold=1.006e+03, percent-clipped=0.0
+2023-03-28 09:44:19,113 INFO [train.py:892] (3/4) Epoch 17, batch 1450, loss[loss=0.2183, simple_loss=0.2809, pruned_loss=0.0779, over 19634.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2649, pruned_loss=0.06435, over 3948032.07 frames. ], batch size: 299, lr: 9.20e-03, grad_scale: 32.0
+2023-03-28 09:46:17,141 INFO [train.py:892] (3/4) Epoch 17, batch 1500, loss[loss=0.2127, simple_loss=0.2696, pruned_loss=0.0779, over 19845.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2649, pruned_loss=0.06435, over 3948589.75 frames. ], batch size: 161, lr: 9.20e-03, grad_scale: 32.0
+2023-03-28 09:47:32,202 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.841e+02 4.138e+02 4.953e+02 5.911e+02 9.562e+02, threshold=9.905e+02, percent-clipped=0.0
+2023-03-28 09:48:10,817 INFO [train.py:892] (3/4) Epoch 17, batch 1550, loss[loss=0.2095, simple_loss=0.2731, pruned_loss=0.07294, over 19741.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2655, pruned_loss=0.06464, over 3948684.08 frames. ], batch size: 269, lr: 9.19e-03, grad_scale: 32.0
+2023-03-28 09:48:32,107 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2983, 2.9750, 4.8183, 3.8896, 4.4702, 4.6999, 4.6515, 4.3547],
+       device='cuda:3'), covar=tensor([0.0289, 0.0822, 0.0083, 0.1135, 0.0116, 0.0176, 0.0136, 0.0136],
+       device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0094, 0.0077, 0.0147, 0.0073, 0.0087, 0.0081, 0.0073],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0002, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 09:49:12,683 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:50:00,502 INFO [train.py:892] (3/4) Epoch 17, batch 1600, loss[loss=0.2082, simple_loss=0.2589, pruned_loss=0.07881, over 19760.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2647, pruned_loss=0.06427, over 3949770.33 frames. ], batch size: 213, lr: 9.18e-03, grad_scale: 32.0
+2023-03-28 09:50:02,882 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:50:23,820 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:51:07,718 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31311.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:51:12,796 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.169e+02 4.273e+02 5.402e+02 6.708e+02 1.273e+03, threshold=1.080e+03, percent-clipped=3.0
+2023-03-28 09:51:41,157 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7699, 4.3684, 4.5562, 4.2453, 4.7369, 3.1900, 3.8393, 2.7797],
+       device='cuda:3'), covar=tensor([0.0162, 0.0206, 0.0121, 0.0164, 0.0130, 0.0853, 0.0727, 0.1210],
+       device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0130, 0.0103, 0.0124, 0.0109, 0.0125, 0.0135, 0.0119],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 09:51:54,976 INFO [train.py:892] (3/4) Epoch 17, batch 1650, loss[loss=0.17, simple_loss=0.2377, pruned_loss=0.05121, over 19904.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2647, pruned_loss=0.06431, over 3949697.56 frames. ], batch size: 116, lr: 9.18e-03, grad_scale: 32.0
+2023-03-28 09:52:57,493 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:53:13,474 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31367.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 09:53:43,712 INFO [train.py:892] (3/4) Epoch 17, batch 1700, loss[loss=0.2078, simple_loss=0.2749, pruned_loss=0.07036, over 19877.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2646, pruned_loss=0.06407, over 3952035.67 frames. ], batch size: 139, lr: 9.17e-03, grad_scale: 32.0
+2023-03-28 09:54:50,495 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31409.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:54:59,290 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.015e+02 4.580e+02 5.253e+02 6.145e+02 1.351e+03, threshold=1.051e+03, percent-clipped=1.0
+2023-03-28 09:55:32,409 INFO [train.py:892] (3/4) Epoch 17, batch 1750, loss[loss=0.1855, simple_loss=0.2498, pruned_loss=0.06053, over 19833.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2663, pruned_loss=0.06488, over 3949358.43 frames. ], batch size: 90, lr: 9.16e-03, grad_scale: 32.0
+2023-03-28 09:56:48,071 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 09:57:07,930 INFO [train.py:892] (3/4) Epoch 17, batch 1800, loss[loss=0.2062, simple_loss=0.28, pruned_loss=0.06616, over 19826.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2656, pruned_loss=0.06467, over 3950270.29 frames. ], batch size: 57, lr: 9.15e-03, grad_scale: 32.0
+2023-03-28 09:57:26,733 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31491.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 09:58:07,131 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.684e+02 3.864e+02 5.000e+02 5.920e+02 9.753e+02, threshold=1.000e+03, percent-clipped=0.0
+2023-03-28 09:58:38,268 INFO [train.py:892] (3/4) Epoch 17, batch 1850, loss[loss=0.1678, simple_loss=0.2456, pruned_loss=0.04498, over 19577.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2672, pruned_loss=0.06413, over 3948444.04 frames. ], batch size: 53, lr: 9.15e-03, grad_scale: 32.0
+2023-03-28 09:59:46,575 INFO [train.py:892] (3/4) Epoch 18, batch 0, loss[loss=0.1709, simple_loss=0.2463, pruned_loss=0.04773, over 19731.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2463, pruned_loss=0.04773, over 19731.00 frames. ], batch size: 80, lr: 8.89e-03, grad_scale: 32.0
+2023-03-28 09:59:46,575 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-28 10:00:15,456 INFO [train.py:926] (3/4) Epoch 18, validation: loss=0.171, simple_loss=0.2489, pruned_loss=0.04657, over 2883724.00 frames. 
+2023-03-28 10:00:15,457 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-28 10:00:43,175 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8863, 4.4931, 4.6531, 4.3637, 4.8701, 3.2814, 3.8955, 2.5923],
+       device='cuda:3'), covar=tensor([0.0168, 0.0197, 0.0125, 0.0164, 0.0124, 0.0801, 0.0812, 0.1330],
+       device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0130, 0.0103, 0.0124, 0.0110, 0.0124, 0.0135, 0.0118],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 10:00:52,982 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31552.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 10:01:09,742 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31559.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:02:01,462 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:02:11,838 INFO [train.py:892] (3/4) Epoch 18, batch 50, loss[loss=0.2235, simple_loss=0.2893, pruned_loss=0.07882, over 19764.00 frames. ], tot_loss[loss=0.193, simple_loss=0.2604, pruned_loss=0.06278, over 891264.53 frames. ], batch size: 244, lr: 8.88e-03, grad_scale: 32.0
+2023-03-28 10:02:21,721 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:02:47,509 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4739, 4.0968, 4.1089, 4.4655, 4.1419, 4.4988, 4.5608, 4.7229],
+       device='cuda:3'), covar=tensor([0.0647, 0.0327, 0.0489, 0.0315, 0.0603, 0.0386, 0.0403, 0.0290],
+       device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0160, 0.0187, 0.0157, 0.0158, 0.0139, 0.0141, 0.0179],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 10:02:52,050 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6415, 3.6822, 2.2922, 3.9023, 4.0579, 1.8582, 3.2024, 3.2900],
+       device='cuda:3'), covar=tensor([0.0735, 0.1062, 0.2822, 0.0859, 0.0525, 0.2965, 0.1290, 0.0714],
+       device='cuda:3'), in_proj_covar=tensor([0.0217, 0.0239, 0.0220, 0.0241, 0.0212, 0.0203, 0.0230, 0.0177],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 10:03:03,304 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:03:16,248 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.475e+02 3.973e+02 4.915e+02 6.184e+02 1.019e+03, threshold=9.829e+02, percent-clipped=1.0
+2023-03-28 10:03:52,356 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:04:05,498 INFO [train.py:892] (3/4) Epoch 18, batch 100, loss[loss=0.2008, simple_loss=0.2647, pruned_loss=0.06845, over 19872.00 frames. ], tot_loss[loss=0.1938, simple_loss=0.2611, pruned_loss=0.06322, over 1568544.39 frames. ], batch size: 158, lr: 8.87e-03, grad_scale: 32.0
+2023-03-28 10:04:10,149 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31638.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:04:57,969 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.1408, 5.2138, 5.6624, 5.3810, 5.3693, 4.9546, 5.2812, 5.1857],
+       device='cuda:3'), covar=tensor([0.1489, 0.1279, 0.0903, 0.1156, 0.0719, 0.1072, 0.1957, 0.2120],
+       device='cuda:3'), in_proj_covar=tensor([0.0266, 0.0276, 0.0328, 0.0260, 0.0240, 0.0240, 0.0321, 0.0352],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 10:05:17,223 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31667.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 10:06:00,557 INFO [train.py:892] (3/4) Epoch 18, batch 150, loss[loss=0.2344, simple_loss=0.3071, pruned_loss=0.08087, over 19615.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2652, pruned_loss=0.06495, over 2093735.46 frames. ], batch size: 351, lr: 8.86e-03, grad_scale: 32.0
+2023-03-28 10:06:42,133 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31704.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:07:06,651 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.935e+02 4.429e+02 5.296e+02 6.261e+02 1.038e+03, threshold=1.059e+03, percent-clipped=5.0
+2023-03-28 10:07:07,557 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31715.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 10:07:37,054 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4405, 3.1541, 3.4704, 2.4553, 3.5827, 2.9273, 3.0746, 3.5146],
+       device='cuda:3'), covar=tensor([0.0476, 0.0370, 0.0507, 0.0788, 0.0374, 0.0338, 0.0420, 0.0377],
+       device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0071, 0.0069, 0.0100, 0.0066, 0.0066, 0.0065, 0.0057],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 10:07:53,071 INFO [train.py:892] (3/4) Epoch 18, batch 200, loss[loss=0.1958, simple_loss=0.2674, pruned_loss=0.06206, over 19859.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2645, pruned_loss=0.06419, over 2505038.81 frames. ], batch size: 51, lr: 8.86e-03, grad_scale: 16.0
+2023-03-28 10:09:00,505 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:09:00,688 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:09:47,141 INFO [train.py:892] (3/4) Epoch 18, batch 250, loss[loss=0.1861, simple_loss=0.2551, pruned_loss=0.05856, over 19773.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2646, pruned_loss=0.06428, over 2824547.77 frames. ], batch size: 116, lr: 8.85e-03, grad_scale: 16.0
+2023-03-28 10:09:53,034 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:10:52,978 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.400e+02 4.013e+02 4.672e+02 5.827e+02 1.248e+03, threshold=9.344e+02, percent-clipped=1.0
+2023-03-28 10:11:15,802 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:11:20,270 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.4540, 1.4016, 1.6356, 1.6790, 1.3868, 1.5708, 1.4695, 1.5768],
+       device='cuda:3'), covar=tensor([0.0253, 0.0244, 0.0225, 0.0187, 0.0356, 0.0213, 0.0385, 0.0233],
+       device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0057, 0.0061, 0.0053, 0.0067, 0.0062, 0.0079, 0.0055],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-28 10:11:41,785 INFO [train.py:892] (3/4) Epoch 18, batch 300, loss[loss=0.1812, simple_loss=0.2624, pruned_loss=0.05001, over 19797.00 frames. ], tot_loss[loss=0.195, simple_loss=0.264, pruned_loss=0.06303, over 3075077.42 frames. ], batch size: 83, lr: 8.84e-03, grad_scale: 16.0
+2023-03-28 10:12:08,487 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31847.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 10:12:14,523 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:12:34,787 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6718, 4.8051, 5.2570, 4.7909, 4.4294, 5.1520, 5.0110, 5.4521],
+       device='cuda:3'), covar=tensor([0.1261, 0.0405, 0.0478, 0.0403, 0.0608, 0.0445, 0.0460, 0.0389],
+       device='cuda:3'), in_proj_covar=tensor([0.0268, 0.0207, 0.0209, 0.0215, 0.0193, 0.0218, 0.0214, 0.0197],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 10:12:42,716 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.0078, 1.9876, 2.3513, 2.2724, 1.9901, 2.3078, 2.1897, 2.1662],
+       device='cuda:3'), covar=tensor([0.0291, 0.0254, 0.0203, 0.0175, 0.0335, 0.0185, 0.0328, 0.0309],
+       device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0056, 0.0060, 0.0052, 0.0066, 0.0061, 0.0077, 0.0054],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-28 10:13:18,390 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:13:32,377 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:13:35,328 INFO [train.py:892] (3/4) Epoch 18, batch 350, loss[loss=0.168, simple_loss=0.2456, pruned_loss=0.04524, over 19799.00 frames. ], tot_loss[loss=0.193, simple_loss=0.262, pruned_loss=0.06196, over 3268778.65 frames. ], batch size: 111, lr: 8.84e-03, grad_scale: 16.0
+2023-03-28 10:13:37,944 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.3748, 1.2781, 1.5026, 1.5647, 1.2731, 1.4431, 1.3256, 1.4680],
+       device='cuda:3'), covar=tensor([0.0269, 0.0286, 0.0236, 0.0188, 0.0394, 0.0245, 0.0416, 0.0250],
+       device='cuda:3'), in_proj_covar=tensor([0.0061, 0.0057, 0.0061, 0.0053, 0.0067, 0.0062, 0.0079, 0.0054],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-28 10:14:41,485 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.968e+02 4.069e+02 4.792e+02 5.707e+02 1.077e+03, threshold=9.584e+02, percent-clipped=3.0
+2023-03-28 10:14:42,360 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:15:31,576 INFO [train.py:892] (3/4) Epoch 18, batch 400, loss[loss=0.3118, simple_loss=0.3737, pruned_loss=0.125, over 19278.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2653, pruned_loss=0.06354, over 3418224.37 frames. ], batch size: 483, lr: 8.83e-03, grad_scale: 16.0
+2023-03-28 10:15:38,966 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9613, 4.5456, 4.7579, 4.4397, 4.9519, 3.2255, 3.9662, 2.6922],
+       device='cuda:3'), covar=tensor([0.0173, 0.0193, 0.0121, 0.0171, 0.0114, 0.0825, 0.0788, 0.1341],
+       device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0132, 0.0104, 0.0125, 0.0110, 0.0125, 0.0137, 0.0121],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 10:15:39,076 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:15:54,924 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.2070, 5.5130, 5.5525, 5.4861, 5.1660, 5.4806, 4.9243, 5.0452],
+       device='cuda:3'), covar=tensor([0.0423, 0.0429, 0.0512, 0.0426, 0.0610, 0.0597, 0.0725, 0.0975],
+       device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0234, 0.0263, 0.0227, 0.0223, 0.0212, 0.0235, 0.0279],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 10:16:04,169 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:16:39,623 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0969, 3.3038, 2.8470, 2.4485, 2.9203, 3.1694, 3.2041, 3.2353],
+       device='cuda:3'), covar=tensor([0.0197, 0.0285, 0.0245, 0.0469, 0.0311, 0.0283, 0.0185, 0.0179],
+       device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0073, 0.0081, 0.0085, 0.0089, 0.0065, 0.0061, 0.0064],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 10:17:03,305 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:17:16,146 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9236, 4.8877, 5.4287, 4.9054, 4.3374, 5.1181, 5.0830, 5.6067],
+       device='cuda:3'), covar=tensor([0.1053, 0.0367, 0.0370, 0.0371, 0.0718, 0.0433, 0.0403, 0.0289],
+       device='cuda:3'), in_proj_covar=tensor([0.0267, 0.0207, 0.0207, 0.0214, 0.0193, 0.0217, 0.0214, 0.0197],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 10:17:23,480 INFO [train.py:892] (3/4) Epoch 18, batch 450, loss[loss=0.1787, simple_loss=0.2454, pruned_loss=0.05601, over 19873.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2657, pruned_loss=0.06409, over 3535844.50 frames. ], batch size: 134, lr: 8.82e-03, grad_scale: 16.0
+2023-03-28 10:17:32,028 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3671, 3.4928, 2.1011, 3.6182, 3.7348, 1.6545, 3.0121, 2.8623],
+       device='cuda:3'), covar=tensor([0.0756, 0.0835, 0.2761, 0.0762, 0.0538, 0.2833, 0.1202, 0.0828],
+       device='cuda:3'), in_proj_covar=tensor([0.0215, 0.0239, 0.0220, 0.0242, 0.0213, 0.0200, 0.0230, 0.0177],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 10:18:28,621 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32011.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:18:36,658 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.048e+02 4.283e+02 4.991e+02 5.829e+02 1.431e+03, threshold=9.983e+02, percent-clipped=4.0
+2023-03-28 10:18:59,663 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5172, 4.1600, 4.1746, 4.5085, 4.1336, 4.5318, 4.6816, 4.7942],
+       device='cuda:3'), covar=tensor([0.0612, 0.0360, 0.0513, 0.0298, 0.0626, 0.0446, 0.0395, 0.0281],
+       device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0158, 0.0183, 0.0154, 0.0157, 0.0138, 0.0136, 0.0175],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 10:19:26,548 INFO [train.py:892] (3/4) Epoch 18, batch 500, loss[loss=0.1868, simple_loss=0.2593, pruned_loss=0.05714, over 19818.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.2646, pruned_loss=0.06355, over 3627993.86 frames. ], batch size: 72, lr: 8.82e-03, grad_scale: 16.0
+2023-03-28 10:20:20,745 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32060.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:20:34,990 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:21:22,495 INFO [train.py:892] (3/4) Epoch 18, batch 550, loss[loss=0.1871, simple_loss=0.2558, pruned_loss=0.05923, over 19899.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.2647, pruned_loss=0.0636, over 3698367.34 frames. ], batch size: 91, lr: 8.81e-03, grad_scale: 16.0
+2023-03-28 10:21:32,542 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6813, 2.7024, 3.0406, 2.8547, 2.7038, 2.5865, 2.7707, 3.0125],
+       device='cuda:3'), covar=tensor([0.0212, 0.0282, 0.0243, 0.0206, 0.0330, 0.0290, 0.0294, 0.0211],
+       device='cuda:3'), in_proj_covar=tensor([0.0060, 0.0056, 0.0061, 0.0053, 0.0067, 0.0062, 0.0078, 0.0054],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-28 10:22:25,779 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:22:30,548 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.880e+02 4.056e+02 4.821e+02 6.235e+02 9.910e+02, threshold=9.642e+02, percent-clipped=0.0
+2023-03-28 10:23:18,291 INFO [train.py:892] (3/4) Epoch 18, batch 600, loss[loss=0.2082, simple_loss=0.2714, pruned_loss=0.07251, over 19797.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.2647, pruned_loss=0.06368, over 3754411.02 frames. ], batch size: 224, lr: 8.80e-03, grad_scale: 16.0
+2023-03-28 10:23:38,026 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:23:44,724 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32147.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 10:24:56,494 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:25:12,430 INFO [train.py:892] (3/4) Epoch 18, batch 650, loss[loss=0.1947, simple_loss=0.263, pruned_loss=0.06321, over 19765.00 frames. ], tot_loss[loss=0.1942, simple_loss=0.263, pruned_loss=0.06266, over 3798640.73 frames. ], batch size: 217, lr: 8.80e-03, grad_scale: 16.0
+2023-03-28 10:25:36,497 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32195.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 10:25:38,911 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0
+2023-03-28 10:25:54,466 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.82 vs. limit=5.0
+2023-03-28 10:26:20,621 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.941e+02 4.475e+02 5.110e+02 5.947e+02 1.058e+03, threshold=1.022e+03, percent-clipped=2.0
+2023-03-28 10:27:06,413 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:27:09,562 INFO [train.py:892] (3/4) Epoch 18, batch 700, loss[loss=0.1763, simple_loss=0.2524, pruned_loss=0.05004, over 19797.00 frames. ], tot_loss[loss=0.1933, simple_loss=0.2625, pruned_loss=0.06204, over 3832924.68 frames. ], batch size: 74, lr: 8.79e-03, grad_scale: 16.0
+2023-03-28 10:28:31,688 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32271.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:29:04,741 INFO [train.py:892] (3/4) Epoch 18, batch 750, loss[loss=0.2031, simple_loss=0.2699, pruned_loss=0.0682, over 19845.00 frames. ], tot_loss[loss=0.1932, simple_loss=0.2626, pruned_loss=0.06191, over 3858916.24 frames. ], batch size: 208, lr: 8.78e-03, grad_scale: 16.0
+2023-03-28 10:29:52,387 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:30:09,301 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.935e+02 4.187e+02 5.149e+02 5.952e+02 1.160e+03, threshold=1.030e+03, percent-clipped=2.0
+2023-03-28 10:30:59,337 INFO [train.py:892] (3/4) Epoch 18, batch 800, loss[loss=0.1781, simple_loss=0.2517, pruned_loss=0.05221, over 19828.00 frames. ], tot_loss[loss=0.1933, simple_loss=0.2628, pruned_loss=0.06189, over 3878917.41 frames. ], batch size: 143, lr: 8.78e-03, grad_scale: 16.0
+2023-03-28 10:31:15,029 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9184, 3.2882, 2.7206, 2.2783, 2.8402, 3.1514, 3.1639, 3.0723],
+       device='cuda:3'), covar=tensor([0.0217, 0.0232, 0.0255, 0.0518, 0.0312, 0.0220, 0.0150, 0.0172],
+       device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0074, 0.0083, 0.0088, 0.0091, 0.0066, 0.0063, 0.0066],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 10:31:51,813 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:32:44,939 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4605, 3.5713, 2.1829, 3.6951, 3.8269, 1.7062, 3.0364, 2.8676],
+       device='cuda:3'), covar=tensor([0.0730, 0.0859, 0.2608, 0.0736, 0.0471, 0.2647, 0.1162, 0.0776],
+       device='cuda:3'), in_proj_covar=tensor([0.0216, 0.0237, 0.0220, 0.0241, 0.0212, 0.0199, 0.0229, 0.0177],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 10:32:47,932 INFO [train.py:892] (3/4) Epoch 18, batch 850, loss[loss=0.1635, simple_loss=0.2278, pruned_loss=0.04961, over 19744.00 frames. ], tot_loss[loss=0.1941, simple_loss=0.2637, pruned_loss=0.06229, over 3894733.78 frames. ], batch size: 140, lr: 8.77e-03, grad_scale: 16.0
+2023-03-28 10:33:22,239 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0619, 3.1717, 1.9642, 3.2289, 3.3261, 1.5546, 2.6783, 2.4845],
+       device='cuda:3'), covar=tensor([0.0780, 0.0768, 0.2615, 0.0702, 0.0531, 0.2561, 0.1156, 0.0895],
+       device='cuda:3'), in_proj_covar=tensor([0.0215, 0.0237, 0.0219, 0.0240, 0.0212, 0.0199, 0.0228, 0.0177],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 10:33:36,800 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:33:53,381 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.074e+02 4.132e+02 4.901e+02 5.957e+02 1.046e+03, threshold=9.802e+02, percent-clipped=1.0
+2023-03-28 10:34:11,263 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0785, 2.9228, 1.6903, 3.5759, 3.3352, 3.5418, 3.6301, 2.8432],
+       device='cuda:3'), covar=tensor([0.0566, 0.0659, 0.1845, 0.0547, 0.0533, 0.0396, 0.0533, 0.0772],
+       device='cuda:3'), in_proj_covar=tensor([0.0130, 0.0129, 0.0136, 0.0134, 0.0117, 0.0115, 0.0130, 0.0134],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 10:34:39,917 INFO [train.py:892] (3/4) Epoch 18, batch 900, loss[loss=0.1933, simple_loss=0.2513, pruned_loss=0.06766, over 19871.00 frames. ], tot_loss[loss=0.1926, simple_loss=0.2623, pruned_loss=0.06143, over 3905977.85 frames. ], batch size: 136, lr: 8.76e-03, grad_scale: 16.0
+2023-03-28 10:34:40,778 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3736, 4.4161, 4.8520, 4.3547, 4.0662, 4.6214, 4.4421, 4.9904],
+       device='cuda:3'), covar=tensor([0.1059, 0.0390, 0.0421, 0.0411, 0.0849, 0.0500, 0.0545, 0.0328],
+       device='cuda:3'), in_proj_covar=tensor([0.0266, 0.0207, 0.0207, 0.0217, 0.0196, 0.0219, 0.0215, 0.0194],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 10:34:58,557 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:36:20,337 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:36:36,666 INFO [train.py:892] (3/4) Epoch 18, batch 950, loss[loss=0.2064, simple_loss=0.2689, pruned_loss=0.07197, over 19765.00 frames. ], tot_loss[loss=0.1937, simple_loss=0.2631, pruned_loss=0.06214, over 3916575.92 frames. ], batch size: 226, lr: 8.76e-03, grad_scale: 16.0
+2023-03-28 10:36:53,375 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:37:41,376 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.554e+02 4.412e+02 5.212e+02 6.577e+02 2.177e+03, threshold=1.042e+03, percent-clipped=5.0
+2023-03-28 10:38:01,615 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.05 vs. limit=5.0
+2023-03-28 10:38:11,322 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:38:28,446 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:38:32,107 INFO [train.py:892] (3/4) Epoch 18, batch 1000, loss[loss=0.1943, simple_loss=0.2619, pruned_loss=0.06331, over 19868.00 frames. ], tot_loss[loss=0.1934, simple_loss=0.2629, pruned_loss=0.06191, over 3924906.14 frames. ], batch size: 64, lr: 8.75e-03, grad_scale: 16.0
+2023-03-28 10:38:39,863 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6945, 4.5984, 2.8822, 4.9676, 5.1785, 2.2431, 4.3012, 3.6523],
+       device='cuda:3'), covar=tensor([0.0484, 0.0774, 0.2361, 0.0589, 0.0377, 0.2764, 0.0895, 0.0727],
+       device='cuda:3'), in_proj_covar=tensor([0.0215, 0.0235, 0.0219, 0.0240, 0.0212, 0.0199, 0.0229, 0.0176],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 10:39:52,849 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32571.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:40:16,947 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:40:24,638 INFO [train.py:892] (3/4) Epoch 18, batch 1050, loss[loss=0.2152, simple_loss=0.2856, pruned_loss=0.07238, over 19708.00 frames. ], tot_loss[loss=0.1937, simple_loss=0.263, pruned_loss=0.06218, over 3931368.09 frames. ], batch size: 305, lr: 8.74e-03, grad_scale: 16.0
+2023-03-28 10:41:12,754 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:41:30,002 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.098e+02 4.345e+02 4.935e+02 6.065e+02 2.066e+03, threshold=9.870e+02, percent-clipped=4.0
+2023-03-28 10:41:38,233 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:42:16,175 INFO [train.py:892] (3/4) Epoch 18, batch 1100, loss[loss=0.1667, simple_loss=0.2374, pruned_loss=0.04801, over 19745.00 frames. ], tot_loss[loss=0.1932, simple_loss=0.2628, pruned_loss=0.06182, over 3934972.89 frames. ], batch size: 110, lr: 8.74e-03, grad_scale: 16.0
+2023-03-28 10:42:54,946 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:43:19,006 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0819, 4.1204, 2.3024, 4.4185, 4.4891, 1.9365, 3.7339, 3.3036],
+       device='cuda:3'), covar=tensor([0.0609, 0.0736, 0.2910, 0.0570, 0.0595, 0.3093, 0.1018, 0.0791],
+       device='cuda:3'), in_proj_covar=tensor([0.0218, 0.0239, 0.0222, 0.0242, 0.0216, 0.0201, 0.0233, 0.0178],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 10:44:04,843 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0706, 3.7988, 3.8513, 4.1516, 3.7714, 4.1168, 4.2121, 4.3630],
+       device='cuda:3'), covar=tensor([0.0594, 0.0404, 0.0510, 0.0336, 0.0700, 0.0509, 0.0383, 0.0329],
+       device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0157, 0.0181, 0.0153, 0.0157, 0.0137, 0.0136, 0.0177],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 10:44:10,625 INFO [train.py:892] (3/4) Epoch 18, batch 1150, loss[loss=0.1897, simple_loss=0.2616, pruned_loss=0.05888, over 19882.00 frames. ], tot_loss[loss=0.1932, simple_loss=0.2624, pruned_loss=0.06199, over 3939102.44 frames. ], batch size: 97, lr: 8.73e-03, grad_scale: 16.0
+2023-03-28 10:45:05,366 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4869, 4.0801, 4.2318, 4.5649, 4.1662, 4.6322, 4.6467, 4.8349],
+       device='cuda:3'), covar=tensor([0.0621, 0.0424, 0.0501, 0.0300, 0.0680, 0.0413, 0.0375, 0.0300],
+       device='cuda:3'), in_proj_covar=tensor([0.0137, 0.0158, 0.0182, 0.0154, 0.0157, 0.0137, 0.0137, 0.0178],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 10:45:16,102 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.651e+02 4.519e+02 5.355e+02 6.274e+02 1.728e+03, threshold=1.071e+03, percent-clipped=4.0
+2023-03-28 10:46:02,484 INFO [train.py:892] (3/4) Epoch 18, batch 1200, loss[loss=0.1922, simple_loss=0.2554, pruned_loss=0.06449, over 19817.00 frames. ], tot_loss[loss=0.193, simple_loss=0.2621, pruned_loss=0.06191, over 3940825.26 frames. ], batch size: 72, lr: 8.72e-03, grad_scale: 16.0
+2023-03-28 10:46:06,551 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.93 vs. limit=5.0
+2023-03-28 10:46:18,134 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.86 vs. limit=5.0
+2023-03-28 10:47:52,957 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0
+2023-03-28 10:48:00,615 INFO [train.py:892] (3/4) Epoch 18, batch 1250, loss[loss=0.2095, simple_loss=0.2875, pruned_loss=0.06576, over 19886.00 frames. ], tot_loss[loss=0.1914, simple_loss=0.2606, pruned_loss=0.06114, over 3943008.59 frames. ], batch size: 63, lr: 8.72e-03, grad_scale: 16.0
+2023-03-28 10:48:27,895 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.1312, 2.6087, 2.2719, 1.5876, 2.2239, 2.4459, 2.5257, 2.4499],
+       device='cuda:3'), covar=tensor([0.0311, 0.0216, 0.0260, 0.0591, 0.0377, 0.0231, 0.0167, 0.0215],
+       device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0074, 0.0082, 0.0087, 0.0090, 0.0065, 0.0063, 0.0065],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 10:48:38,515 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9113, 4.7482, 4.7901, 5.1681, 4.7799, 5.4647, 5.1114, 5.3035],
+       device='cuda:3'), covar=tensor([0.0755, 0.0389, 0.0496, 0.0336, 0.0645, 0.0354, 0.0645, 0.0484],
+       device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0158, 0.0182, 0.0154, 0.0158, 0.0138, 0.0136, 0.0178],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 10:49:03,776 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.128e+02 4.132e+02 4.830e+02 5.865e+02 8.224e+02, threshold=9.660e+02, percent-clipped=0.0
+2023-03-28 10:49:53,668 INFO [train.py:892] (3/4) Epoch 18, batch 1300, loss[loss=0.2221, simple_loss=0.2874, pruned_loss=0.07844, over 19777.00 frames. ], tot_loss[loss=0.1914, simple_loss=0.2607, pruned_loss=0.06107, over 3944841.76 frames. ], batch size: 198, lr: 8.71e-03, grad_scale: 16.0
+2023-03-28 10:51:47,894 INFO [train.py:892] (3/4) Epoch 18, batch 1350, loss[loss=0.2116, simple_loss=0.2716, pruned_loss=0.07586, over 19775.00 frames. ], tot_loss[loss=0.1931, simple_loss=0.2626, pruned_loss=0.06187, over 3944949.44 frames. ], batch size: 152, lr: 8.71e-03, grad_scale: 16.0
+2023-03-28 10:52:29,514 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.32 vs. limit=5.0
+2023-03-28 10:52:55,306 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.479e+02 4.452e+02 5.064e+02 6.052e+02 1.021e+03, threshold=1.013e+03, percent-clipped=2.0
+2023-03-28 10:53:44,829 INFO [train.py:892] (3/4) Epoch 18, batch 1400, loss[loss=0.1681, simple_loss=0.228, pruned_loss=0.05406, over 19858.00 frames. ], tot_loss[loss=0.192, simple_loss=0.2616, pruned_loss=0.06123, over 3947436.93 frames. ], batch size: 122, lr: 8.70e-03, grad_scale: 16.0
+2023-03-28 10:54:21,002 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3214, 3.5741, 3.7697, 4.4057, 2.8611, 3.4625, 2.7211, 2.6327],
+       device='cuda:3'), covar=tensor([0.0492, 0.2127, 0.0862, 0.0317, 0.2194, 0.0852, 0.1317, 0.1800],
+       device='cuda:3'), in_proj_covar=tensor([0.0225, 0.0336, 0.0235, 0.0181, 0.0242, 0.0195, 0.0208, 0.0216],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 10:54:52,735 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32966.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:55:38,293 INFO [train.py:892] (3/4) Epoch 18, batch 1450, loss[loss=0.1889, simple_loss=0.27, pruned_loss=0.05389, over 19824.00 frames. ], tot_loss[loss=0.1932, simple_loss=0.2626, pruned_loss=0.06187, over 3949217.53 frames. ], batch size: 57, lr: 8.69e-03, grad_scale: 16.0
+2023-03-28 10:56:44,785 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.979e+02 4.312e+02 5.122e+02 6.077e+02 1.174e+03, threshold=1.024e+03, percent-clipped=2.0
+2023-03-28 10:57:12,048 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33027.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:57:32,874 INFO [train.py:892] (3/4) Epoch 18, batch 1500, loss[loss=0.1857, simple_loss=0.2526, pruned_loss=0.05946, over 19811.00 frames. ], tot_loss[loss=0.1934, simple_loss=0.263, pruned_loss=0.06191, over 3949770.93 frames. ], batch size: 132, lr: 8.69e-03, grad_scale: 16.0
+2023-03-28 10:57:56,208 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5497, 2.6106, 4.0981, 3.6639, 3.9347, 4.0647, 3.9647, 3.8746],
+       device='cuda:3'), covar=tensor([0.0318, 0.0798, 0.0105, 0.0670, 0.0127, 0.0189, 0.0145, 0.0131],
+       device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0096, 0.0079, 0.0150, 0.0075, 0.0088, 0.0083, 0.0076],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0002, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 10:58:45,800 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 10:59:25,361 INFO [train.py:892] (3/4) Epoch 18, batch 1550, loss[loss=0.1777, simple_loss=0.2426, pruned_loss=0.05643, over 19875.00 frames. ], tot_loss[loss=0.1928, simple_loss=0.2624, pruned_loss=0.06157, over 3950415.93 frames. ], batch size: 125, lr: 8.68e-03, grad_scale: 16.0
+2023-03-28 11:00:25,343 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.832e+02 4.176e+02 4.933e+02 6.239e+02 1.615e+03, threshold=9.866e+02, percent-clipped=3.0
+2023-03-28 11:00:59,182 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:01:09,744 INFO [train.py:892] (3/4) Epoch 18, batch 1600, loss[loss=0.182, simple_loss=0.2533, pruned_loss=0.05535, over 19885.00 frames. ], tot_loss[loss=0.1922, simple_loss=0.2617, pruned_loss=0.06136, over 3952344.85 frames. ], batch size: 134, lr: 8.67e-03, grad_scale: 16.0
+2023-03-28 11:01:53,568 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33155.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:02:29,045 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8072, 3.8633, 2.3649, 4.0114, 4.1293, 1.9880, 3.4476, 3.2260],
+       device='cuda:3'), covar=tensor([0.0730, 0.0848, 0.2723, 0.0857, 0.0598, 0.2859, 0.1108, 0.0798],
+       device='cuda:3'), in_proj_covar=tensor([0.0217, 0.0238, 0.0221, 0.0243, 0.0216, 0.0199, 0.0232, 0.0177],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 11:02:55,055 INFO [train.py:892] (3/4) Epoch 18, batch 1650, loss[loss=0.1729, simple_loss=0.2476, pruned_loss=0.04907, over 19855.00 frames. ], tot_loss[loss=0.193, simple_loss=0.2622, pruned_loss=0.06195, over 3951844.09 frames. ], batch size: 78, lr: 8.67e-03, grad_scale: 16.0
+2023-03-28 11:04:01,028 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.567e+02 4.484e+02 5.220e+02 6.414e+02 1.768e+03, threshold=1.044e+03, percent-clipped=5.0
+2023-03-28 11:04:05,645 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33216.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:04:50,003 INFO [train.py:892] (3/4) Epoch 18, batch 1700, loss[loss=0.1823, simple_loss=0.2569, pruned_loss=0.05388, over 19796.00 frames. ], tot_loss[loss=0.1928, simple_loss=0.2625, pruned_loss=0.06158, over 3950755.17 frames. ], batch size: 79, lr: 8.66e-03, grad_scale: 16.0
+2023-03-28 11:06:44,500 INFO [train.py:892] (3/4) Epoch 18, batch 1750, loss[loss=0.1588, simple_loss=0.2413, pruned_loss=0.03815, over 19829.00 frames. ], tot_loss[loss=0.1923, simple_loss=0.2619, pruned_loss=0.06138, over 3949797.29 frames. ], batch size: 52, lr: 8.65e-03, grad_scale: 16.0
+2023-03-28 11:07:09,419 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9245, 2.3948, 2.9016, 3.2338, 3.6898, 4.0043, 3.8854, 4.0594],
+       device='cuda:3'), covar=tensor([0.0864, 0.1746, 0.1178, 0.0574, 0.0357, 0.0237, 0.0358, 0.0349],
+       device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0172, 0.0172, 0.0143, 0.0124, 0.0116, 0.0109, 0.0105],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 11:07:40,437 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.050e+02 4.001e+02 4.864e+02 5.584e+02 1.277e+03, threshold=9.729e+02, percent-clipped=1.0
+2023-03-28 11:07:54,280 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:08:21,753 INFO [train.py:892] (3/4) Epoch 18, batch 1800, loss[loss=0.1807, simple_loss=0.2507, pruned_loss=0.05534, over 19739.00 frames. ], tot_loss[loss=0.1919, simple_loss=0.2616, pruned_loss=0.06106, over 3948762.33 frames. ], batch size: 140, lr: 8.65e-03, grad_scale: 16.0
+2023-03-28 11:09:11,549 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0320, 2.3374, 2.9625, 3.2849, 3.7569, 4.1507, 4.1278, 4.2105],
+       device='cuda:3'), covar=tensor([0.0833, 0.1893, 0.1296, 0.0617, 0.0347, 0.0217, 0.0270, 0.0301],
+       device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0170, 0.0171, 0.0142, 0.0123, 0.0115, 0.0108, 0.0104],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 11:09:57,881 INFO [train.py:892] (3/4) Epoch 18, batch 1850, loss[loss=0.2043, simple_loss=0.2855, pruned_loss=0.06161, over 19819.00 frames. ], tot_loss[loss=0.1934, simple_loss=0.2642, pruned_loss=0.06131, over 3948940.13 frames. ], batch size: 57, lr: 8.64e-03, grad_scale: 16.0
+2023-03-28 11:11:02,086 INFO [train.py:892] (3/4) Epoch 19, batch 0, loss[loss=0.1625, simple_loss=0.2362, pruned_loss=0.04443, over 19735.00 frames. ], tot_loss[loss=0.1625, simple_loss=0.2362, pruned_loss=0.04443, over 19735.00 frames. ], batch size: 89, lr: 8.41e-03, grad_scale: 16.0
+2023-03-28 11:11:02,087 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-28 11:11:35,982 INFO [train.py:926] (3/4) Epoch 19, validation: loss=0.1703, simple_loss=0.2482, pruned_loss=0.04619, over 2883724.00 frames. 
+2023-03-28 11:11:35,983 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-28 11:12:33,176 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.922e+02 3.899e+02 4.712e+02 6.072e+02 1.255e+03, threshold=9.424e+02, percent-clipped=1.0
+2023-03-28 11:12:54,565 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:13:32,573 INFO [train.py:892] (3/4) Epoch 19, batch 50, loss[loss=0.1722, simple_loss=0.2462, pruned_loss=0.04908, over 19750.00 frames. ], tot_loss[loss=0.1836, simple_loss=0.2539, pruned_loss=0.05667, over 891299.36 frames. ], batch size: 44, lr: 8.40e-03, grad_scale: 16.0
+2023-03-28 11:14:40,741 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:15:25,858 INFO [train.py:892] (3/4) Epoch 19, batch 100, loss[loss=0.1665, simple_loss=0.229, pruned_loss=0.05201, over 19875.00 frames. ], tot_loss[loss=0.1846, simple_loss=0.2541, pruned_loss=0.05754, over 1571126.21 frames. ], batch size: 125, lr: 8.39e-03, grad_scale: 16.0
+2023-03-28 11:15:47,856 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5298, 3.5809, 5.0769, 3.8551, 4.2237, 4.1293, 2.7233, 2.9989],
+       device='cuda:3'), covar=tensor([0.0709, 0.2425, 0.0336, 0.0759, 0.1253, 0.0916, 0.2070, 0.2087],
+       device='cuda:3'), in_proj_covar=tensor([0.0336, 0.0365, 0.0316, 0.0254, 0.0357, 0.0327, 0.0336, 0.0306],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 11:16:09,704 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33511.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:16:17,086 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.445e+02 4.091e+02 4.969e+02 5.976e+02 1.351e+03, threshold=9.939e+02, percent-clipped=3.0
+2023-03-28 11:16:42,238 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3603, 3.2840, 3.6114, 2.7165, 3.8154, 3.1109, 3.2463, 3.6811],
+       device='cuda:3'), covar=tensor([0.0687, 0.0324, 0.0652, 0.0749, 0.0367, 0.0378, 0.0461, 0.0343],
+       device='cuda:3'), in_proj_covar=tensor([0.0065, 0.0072, 0.0072, 0.0101, 0.0067, 0.0068, 0.0066, 0.0059],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 11:16:59,235 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33533.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:17:16,225 INFO [train.py:892] (3/4) Epoch 19, batch 150, loss[loss=0.2111, simple_loss=0.2866, pruned_loss=0.06773, over 19855.00 frames. ], tot_loss[loss=0.1874, simple_loss=0.2567, pruned_loss=0.05902, over 2099455.60 frames. ], batch size: 58, lr: 8.39e-03, grad_scale: 16.0
+2023-03-28 11:18:57,045 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0088, 3.7076, 3.7483, 4.0105, 3.7667, 4.1141, 4.1269, 4.2787],
+       device='cuda:3'), covar=tensor([0.0672, 0.0408, 0.0516, 0.0363, 0.0668, 0.0451, 0.0423, 0.0321],
+       device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0158, 0.0181, 0.0153, 0.0158, 0.0138, 0.0137, 0.0176],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 11:19:06,875 INFO [train.py:892] (3/4) Epoch 19, batch 200, loss[loss=0.1868, simple_loss=0.2497, pruned_loss=0.06199, over 19830.00 frames. ], tot_loss[loss=0.1897, simple_loss=0.2594, pruned_loss=0.05999, over 2509119.09 frames. ], batch size: 146, lr: 8.38e-03, grad_scale: 16.0
+2023-03-28 11:19:56,414 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.308e+02 4.349e+02 4.912e+02 5.878e+02 1.071e+03, threshold=9.825e+02, percent-clipped=1.0
+2023-03-28 11:20:15,498 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:20:45,834 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-28 11:20:54,825 INFO [train.py:892] (3/4) Epoch 19, batch 250, loss[loss=0.1756, simple_loss=0.2425, pruned_loss=0.05431, over 19771.00 frames. ], tot_loss[loss=0.1927, simple_loss=0.2629, pruned_loss=0.06123, over 2825917.22 frames. ], batch size: 130, lr: 8.38e-03, grad_scale: 16.0
+2023-03-28 11:21:47,377 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.3374, 2.6909, 2.3571, 1.8260, 2.4686, 2.6742, 2.6132, 2.6521],
+       device='cuda:3'), covar=tensor([0.0326, 0.0251, 0.0278, 0.0546, 0.0361, 0.0215, 0.0227, 0.0200],
+       device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0074, 0.0082, 0.0087, 0.0090, 0.0065, 0.0063, 0.0065],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 11:21:58,899 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:22:13,370 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:22:15,648 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0365, 3.9368, 2.3755, 4.2849, 4.4198, 1.9911, 3.5732, 3.3764],
+       device='cuda:3'), covar=tensor([0.0581, 0.0839, 0.2604, 0.0700, 0.0449, 0.2782, 0.1133, 0.0774],
+       device='cuda:3'), in_proj_covar=tensor([0.0215, 0.0240, 0.0222, 0.0243, 0.0216, 0.0200, 0.0230, 0.0178],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 11:22:29,295 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0328, 2.2953, 3.5993, 3.0612, 3.5033, 3.7032, 3.4015, 3.5110],
+       device='cuda:3'), covar=tensor([0.0424, 0.0886, 0.0099, 0.0549, 0.0119, 0.0185, 0.0174, 0.0137],
+       device='cuda:3'), in_proj_covar=tensor([0.0089, 0.0096, 0.0078, 0.0148, 0.0074, 0.0087, 0.0082, 0.0075],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0002, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 11:22:42,184 INFO [train.py:892] (3/4) Epoch 19, batch 300, loss[loss=0.2016, simple_loss=0.2754, pruned_loss=0.06391, over 19915.00 frames. ], tot_loss[loss=0.1924, simple_loss=0.2622, pruned_loss=0.0613, over 3074705.38 frames. ], batch size: 53, lr: 8.37e-03, grad_scale: 16.0
+2023-03-28 11:23:39,624 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.947e+02 4.475e+02 5.212e+02 6.221e+02 1.054e+03, threshold=1.042e+03, percent-clipped=4.0
+2023-03-28 11:24:02,805 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:24:16,336 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.62 vs. limit=2.0
+2023-03-28 11:24:33,478 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33738.0, num_to_drop=1, layers_to_drop={3}
+2023-03-28 11:24:38,598 INFO [train.py:892] (3/4) Epoch 19, batch 350, loss[loss=0.1705, simple_loss=0.2378, pruned_loss=0.05163, over 19846.00 frames. ], tot_loss[loss=0.1912, simple_loss=0.2615, pruned_loss=0.06047, over 3269038.03 frames. ], batch size: 137, lr: 8.36e-03, grad_scale: 32.0
+2023-03-28 11:24:59,240 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1484, 3.7630, 3.8817, 4.1251, 3.7874, 4.1358, 4.2446, 4.4281],
+       device='cuda:3'), covar=tensor([0.0610, 0.0460, 0.0540, 0.0348, 0.0679, 0.0489, 0.0429, 0.0283],
+       device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0158, 0.0181, 0.0152, 0.0157, 0.0136, 0.0137, 0.0174],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 11:25:51,613 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:25:56,579 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-28 11:26:07,354 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8441, 2.3496, 2.9282, 3.2253, 3.7884, 4.2587, 4.0789, 4.1578],
+       device='cuda:3'), covar=tensor([0.0999, 0.1874, 0.1353, 0.0637, 0.0356, 0.0167, 0.0283, 0.0245],
+       device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0171, 0.0173, 0.0143, 0.0125, 0.0116, 0.0109, 0.0106],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 11:26:29,469 INFO [train.py:892] (3/4) Epoch 19, batch 400, loss[loss=0.1889, simple_loss=0.248, pruned_loss=0.06493, over 19824.00 frames. ], tot_loss[loss=0.1902, simple_loss=0.2604, pruned_loss=0.06, over 3421242.49 frames. ], batch size: 202, lr: 8.36e-03, grad_scale: 32.0
+2023-03-28 11:27:18,807 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:27:26,570 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.571e+02 3.967e+02 4.675e+02 5.503e+02 1.191e+03, threshold=9.350e+02, percent-clipped=1.0
+2023-03-28 11:27:27,970 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0477, 4.6029, 4.7626, 4.4619, 4.9316, 3.2338, 4.0553, 2.5173],
+       device='cuda:3'), covar=tensor([0.0166, 0.0201, 0.0131, 0.0174, 0.0139, 0.0903, 0.0836, 0.1442],
+       device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0133, 0.0106, 0.0125, 0.0111, 0.0128, 0.0137, 0.0121],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 11:27:37,282 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.2141, 4.8023, 4.8139, 5.2279, 4.7843, 5.5267, 5.3343, 5.5432],
+       device='cuda:3'), covar=tensor([0.0609, 0.0350, 0.0438, 0.0304, 0.0596, 0.0286, 0.0375, 0.0284],
+       device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0159, 0.0182, 0.0153, 0.0157, 0.0137, 0.0137, 0.0174],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 11:27:37,535 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4162, 3.4412, 4.9461, 3.5358, 4.0935, 3.9128, 2.6681, 2.7980],
+       device='cuda:3'), covar=tensor([0.0701, 0.2524, 0.0385, 0.0967, 0.1466, 0.0959, 0.2070, 0.2313],
+       device='cuda:3'), in_proj_covar=tensor([0.0333, 0.0362, 0.0314, 0.0254, 0.0356, 0.0323, 0.0334, 0.0303],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 11:27:40,018 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2067, 3.2574, 2.0531, 3.3489, 3.4644, 1.6550, 2.8386, 2.6595],
+       device='cuda:3'), covar=tensor([0.0726, 0.0792, 0.2584, 0.0745, 0.0526, 0.2567, 0.1030, 0.0812],
+       device='cuda:3'), in_proj_covar=tensor([0.0213, 0.0236, 0.0220, 0.0241, 0.0215, 0.0198, 0.0228, 0.0177],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 11:27:57,236 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:28:04,303 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.54 vs. limit=5.0
+2023-03-28 11:28:17,392 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.94 vs. limit=5.0
+2023-03-28 11:28:24,667 INFO [train.py:892] (3/4) Epoch 19, batch 450, loss[loss=0.1656, simple_loss=0.2355, pruned_loss=0.04784, over 19859.00 frames. ], tot_loss[loss=0.1917, simple_loss=0.2618, pruned_loss=0.06082, over 3537844.98 frames. ], batch size: 106, lr: 8.35e-03, grad_scale: 16.0
+2023-03-28 11:29:08,880 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:30:18,957 INFO [train.py:892] (3/4) Epoch 19, batch 500, loss[loss=0.1473, simple_loss=0.218, pruned_loss=0.03829, over 19728.00 frames. ], tot_loss[loss=0.1916, simple_loss=0.2615, pruned_loss=0.06081, over 3630121.92 frames. ], batch size: 47, lr: 8.35e-03, grad_scale: 16.0
+2023-03-28 11:31:15,651 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.605e+02 4.236e+02 4.841e+02 6.500e+02 9.898e+02, threshold=9.682e+02, percent-clipped=2.0
+2023-03-28 11:31:23,802 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-28 11:32:08,682 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:32:11,666 INFO [train.py:892] (3/4) Epoch 19, batch 550, loss[loss=0.1865, simple_loss=0.2678, pruned_loss=0.0526, over 19812.00 frames. ], tot_loss[loss=0.1927, simple_loss=0.2628, pruned_loss=0.0613, over 3698362.63 frames. ], batch size: 72, lr: 8.34e-03, grad_scale: 16.0
+2023-03-28 11:32:43,524 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-28 11:34:04,578 INFO [train.py:892] (3/4) Epoch 19, batch 600, loss[loss=0.1879, simple_loss=0.2606, pruned_loss=0.05759, over 19768.00 frames. ], tot_loss[loss=0.1934, simple_loss=0.2636, pruned_loss=0.06161, over 3753692.26 frames. ], batch size: 280, lr: 8.33e-03, grad_scale: 16.0
+2023-03-28 11:34:05,657 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4541, 1.8456, 1.9875, 2.7729, 3.0038, 3.0691, 2.9777, 3.0721],
+       device='cuda:3'), covar=tensor([0.0980, 0.1918, 0.1682, 0.0653, 0.0472, 0.0330, 0.0434, 0.0385],
+       device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0169, 0.0171, 0.0142, 0.0123, 0.0116, 0.0109, 0.0105],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 11:34:31,012 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34000.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:35:01,391 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.382e+02 4.429e+02 5.203e+02 6.509e+02 1.553e+03, threshold=1.041e+03, percent-clipped=5.0
+2023-03-28 11:35:42,666 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34033.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 11:36:02,211 INFO [train.py:892] (3/4) Epoch 19, batch 650, loss[loss=0.2062, simple_loss=0.2676, pruned_loss=0.07242, over 19844.00 frames. ], tot_loss[loss=0.1939, simple_loss=0.2636, pruned_loss=0.06208, over 3795955.37 frames. ], batch size: 190, lr: 8.33e-03, grad_scale: 16.0
+2023-03-28 11:36:10,062 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34044.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:37:53,397 INFO [train.py:892] (3/4) Epoch 19, batch 700, loss[loss=0.1672, simple_loss=0.235, pruned_loss=0.04967, over 19749.00 frames. ], tot_loss[loss=0.1939, simple_loss=0.2634, pruned_loss=0.06216, over 3830492.50 frames. ], batch size: 129, lr: 8.32e-03, grad_scale: 16.0
+2023-03-28 11:38:28,868 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34105.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:38:52,051 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.720e+02 4.319e+02 5.100e+02 5.966e+02 1.380e+03, threshold=1.020e+03, percent-clipped=2.0
+2023-03-28 11:39:19,074 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:39:45,519 INFO [train.py:892] (3/4) Epoch 19, batch 750, loss[loss=0.1729, simple_loss=0.2492, pruned_loss=0.04837, over 19803.00 frames. ], tot_loss[loss=0.1918, simple_loss=0.262, pruned_loss=0.0608, over 3859053.87 frames. ], batch size: 195, lr: 8.32e-03, grad_scale: 16.0
+2023-03-28 11:39:57,411 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:40:10,275 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-28 11:41:06,996 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:41:40,075 INFO [train.py:892] (3/4) Epoch 19, batch 800, loss[loss=0.2223, simple_loss=0.2903, pruned_loss=0.07715, over 19778.00 frames. ], tot_loss[loss=0.1904, simple_loss=0.261, pruned_loss=0.05989, over 3879655.75 frames. ], batch size: 233, lr: 8.31e-03, grad_scale: 16.0
+2023-03-28 11:42:14,768 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:42:21,147 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4581, 2.6494, 4.0944, 2.9764, 3.3484, 3.1247, 2.1895, 2.3495],
+       device='cuda:3'), covar=tensor([0.1138, 0.3013, 0.0489, 0.0953, 0.1707, 0.1311, 0.2130, 0.2547],
+       device='cuda:3'), in_proj_covar=tensor([0.0334, 0.0364, 0.0316, 0.0254, 0.0357, 0.0327, 0.0335, 0.0304],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 11:42:36,504 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.024e+02 4.346e+02 5.123e+02 5.919e+02 1.290e+03, threshold=1.025e+03, percent-clipped=2.0
+2023-03-28 11:43:11,651 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8898, 4.9791, 5.3355, 5.0156, 5.1430, 4.7766, 5.0158, 4.7997],
+       device='cuda:3'), covar=tensor([0.1437, 0.1256, 0.0840, 0.1213, 0.0707, 0.0809, 0.1888, 0.2068],
+       device='cuda:3'), in_proj_covar=tensor([0.0272, 0.0288, 0.0338, 0.0267, 0.0248, 0.0245, 0.0328, 0.0361],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 11:43:36,137 INFO [train.py:892] (3/4) Epoch 19, batch 850, loss[loss=0.2971, simple_loss=0.3685, pruned_loss=0.1128, over 19239.00 frames. ], tot_loss[loss=0.1923, simple_loss=0.2629, pruned_loss=0.0609, over 3893391.02 frames. ], batch size: 483, lr: 8.30e-03, grad_scale: 16.0
+2023-03-28 11:44:26,634 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.64 vs. limit=5.0
+2023-03-28 11:45:25,868 INFO [train.py:892] (3/4) Epoch 19, batch 900, loss[loss=0.2011, simple_loss=0.2842, pruned_loss=0.059, over 19656.00 frames. ], tot_loss[loss=0.1917, simple_loss=0.262, pruned_loss=0.06073, over 3906539.51 frames. ], batch size: 57, lr: 8.30e-03, grad_scale: 8.0
+2023-03-28 11:45:38,097 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34295.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:45:58,122 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.90 vs. limit=5.0
+2023-03-28 11:46:25,649 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.957e+02 4.098e+02 5.258e+02 6.127e+02 1.043e+03, threshold=1.052e+03, percent-clipped=1.0
+2023-03-28 11:47:00,528 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:47:16,666 INFO [train.py:892] (3/4) Epoch 19, batch 950, loss[loss=0.1921, simple_loss=0.2684, pruned_loss=0.05786, over 19728.00 frames. ], tot_loss[loss=0.1905, simple_loss=0.2611, pruned_loss=0.05999, over 3916354.82 frames. ], batch size: 63, lr: 8.29e-03, grad_scale: 8.0
+2023-03-28 11:48:46,323 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.19 vs. limit=5.0
+2023-03-28 11:48:49,466 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:49:08,919 INFO [train.py:892] (3/4) Epoch 19, batch 1000, loss[loss=0.1872, simple_loss=0.253, pruned_loss=0.06068, over 19827.00 frames. ], tot_loss[loss=0.1912, simple_loss=0.262, pruned_loss=0.06019, over 3922034.33 frames. ], batch size: 127, lr: 8.29e-03, grad_scale: 8.0
+2023-03-28 11:49:29,376 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34400.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:50:08,246 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.058e+02 4.216e+02 4.828e+02 6.014e+02 1.612e+03, threshold=9.655e+02, percent-clipped=1.0
+2023-03-28 11:50:44,678 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-28 11:51:00,686 INFO [train.py:892] (3/4) Epoch 19, batch 1050, loss[loss=0.1889, simple_loss=0.2621, pruned_loss=0.05784, over 19792.00 frames. ], tot_loss[loss=0.1905, simple_loss=0.2615, pruned_loss=0.05977, over 3928205.93 frames. ], batch size: 65, lr: 8.28e-03, grad_scale: 8.0
+2023-03-28 11:51:29,877 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5621, 2.9006, 2.9503, 3.4463, 2.2848, 2.9585, 2.2158, 2.1357],
+       device='cuda:3'), covar=tensor([0.0598, 0.1962, 0.1079, 0.0446, 0.2441, 0.0817, 0.1561, 0.1935],
+       device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0335, 0.0236, 0.0182, 0.0241, 0.0194, 0.0208, 0.0215],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 11:52:51,879 INFO [train.py:892] (3/4) Epoch 19, batch 1100, loss[loss=0.1833, simple_loss=0.2589, pruned_loss=0.05384, over 19931.00 frames. ], tot_loss[loss=0.1906, simple_loss=0.2614, pruned_loss=0.0599, over 3933597.56 frames. ], batch size: 49, lr: 8.27e-03, grad_scale: 8.0
+2023-03-28 11:53:15,458 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:53:47,009 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5857, 2.0358, 2.3199, 2.8705, 3.2716, 3.2717, 3.2974, 3.3572],
+       device='cuda:3'), covar=tensor([0.0941, 0.1724, 0.1320, 0.0642, 0.0414, 0.0337, 0.0352, 0.0369],
+       device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0171, 0.0172, 0.0144, 0.0125, 0.0118, 0.0110, 0.0106],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 11:53:50,398 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.527e+02 4.200e+02 4.924e+02 5.856e+02 9.802e+02, threshold=9.848e+02, percent-clipped=1.0
+2023-03-28 11:54:41,914 INFO [train.py:892] (3/4) Epoch 19, batch 1150, loss[loss=0.1866, simple_loss=0.2472, pruned_loss=0.06295, over 19822.00 frames. ], tot_loss[loss=0.191, simple_loss=0.2613, pruned_loss=0.06036, over 3937092.73 frames. ], batch size: 166, lr: 8.27e-03, grad_scale: 8.0
+2023-03-28 11:56:34,998 INFO [train.py:892] (3/4) Epoch 19, batch 1200, loss[loss=0.158, simple_loss=0.2329, pruned_loss=0.04159, over 19847.00 frames. ], tot_loss[loss=0.1903, simple_loss=0.261, pruned_loss=0.05982, over 3939683.82 frames. ], batch size: 112, lr: 8.26e-03, grad_scale: 8.0
+2023-03-28 11:56:37,644 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4939, 3.1326, 3.3629, 3.1545, 3.6945, 3.5986, 4.2916, 4.7121],
+       device='cuda:3'), covar=tensor([0.0451, 0.1476, 0.1380, 0.1964, 0.1499, 0.1295, 0.0481, 0.0466],
+       device='cuda:3'), in_proj_covar=tensor([0.0228, 0.0227, 0.0250, 0.0240, 0.0276, 0.0240, 0.0206, 0.0224],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 11:56:41,569 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:56:45,640 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:56:56,722 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0518, 2.9229, 1.8790, 3.6374, 3.2807, 3.4696, 3.6102, 2.8537],
+       device='cuda:3'), covar=tensor([0.0623, 0.0701, 0.1601, 0.0487, 0.0577, 0.0537, 0.0579, 0.0778],
+       device='cuda:3'), in_proj_covar=tensor([0.0133, 0.0132, 0.0136, 0.0135, 0.0119, 0.0119, 0.0131, 0.0136],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 11:57:36,125 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.821e+02 4.325e+02 5.082e+02 6.313e+02 1.101e+03, threshold=1.016e+03, percent-clipped=4.0
+2023-03-28 11:57:49,005 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34622.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:57:57,956 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34626.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:57:58,577 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-28 11:58:30,494 INFO [train.py:892] (3/4) Epoch 19, batch 1250, loss[loss=0.1773, simple_loss=0.2446, pruned_loss=0.05503, over 19737.00 frames. ], tot_loss[loss=0.1897, simple_loss=0.26, pruned_loss=0.05966, over 3943052.90 frames. ], batch size: 44, lr: 8.26e-03, grad_scale: 8.0
+2023-03-28 11:58:35,841 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:58:59,626 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 11:59:23,733 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0
+2023-03-28 11:59:47,960 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.70 vs. limit=2.0
+2023-03-28 12:00:01,146 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9005, 2.2789, 2.9478, 3.0226, 3.6784, 3.9788, 3.9465, 4.0969],
+       device='cuda:3'), covar=tensor([0.0923, 0.1852, 0.1264, 0.0668, 0.0372, 0.0231, 0.0334, 0.0267],
+       device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0171, 0.0172, 0.0143, 0.0126, 0.0118, 0.0110, 0.0107],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 12:00:04,874 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34683.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 12:00:13,990 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34687.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:00:21,330 INFO [train.py:892] (3/4) Epoch 19, batch 1300, loss[loss=0.1851, simple_loss=0.2556, pruned_loss=0.05729, over 19800.00 frames. ], tot_loss[loss=0.1909, simple_loss=0.261, pruned_loss=0.06045, over 3943332.01 frames. ], batch size: 200, lr: 8.25e-03, grad_scale: 8.0
+2023-03-28 12:00:44,742 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34700.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:01:06,094 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34710.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 12:01:21,579 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.806e+02 3.918e+02 4.981e+02 5.928e+02 9.169e+02, threshold=9.963e+02, percent-clipped=0.0
+2023-03-28 12:02:16,205 INFO [train.py:892] (3/4) Epoch 19, batch 1350, loss[loss=0.2078, simple_loss=0.269, pruned_loss=0.0733, over 19741.00 frames. ], tot_loss[loss=0.1912, simple_loss=0.2611, pruned_loss=0.06063, over 3945251.59 frames. ], batch size: 259, lr: 8.24e-03, grad_scale: 8.0
+2023-03-28 12:02:34,277 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:03:22,020 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34771.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 12:04:08,638 INFO [train.py:892] (3/4) Epoch 19, batch 1400, loss[loss=0.1919, simple_loss=0.2576, pruned_loss=0.06311, over 19795.00 frames. ], tot_loss[loss=0.191, simple_loss=0.2608, pruned_loss=0.06053, over 3947049.02 frames. ], batch size: 185, lr: 8.24e-03, grad_scale: 8.0
+2023-03-28 12:04:31,260 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:05:06,146 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.813e+02 4.248e+02 4.961e+02 6.105e+02 8.838e+02, threshold=9.921e+02, percent-clipped=0.0
+2023-03-28 12:05:56,214 INFO [train.py:892] (3/4) Epoch 19, batch 1450, loss[loss=0.1807, simple_loss=0.2541, pruned_loss=0.05365, over 19749.00 frames. ], tot_loss[loss=0.1909, simple_loss=0.2609, pruned_loss=0.06048, over 3948021.00 frames. ], batch size: 89, lr: 8.23e-03, grad_scale: 8.0
+2023-03-28 12:06:16,984 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:06:41,875 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.9254, 1.9557, 2.0672, 1.9852, 1.9530, 1.9972, 1.8636, 2.1036],
+       device='cuda:3'), covar=tensor([0.0324, 0.0272, 0.0261, 0.0251, 0.0375, 0.0255, 0.0410, 0.0257],
+       device='cuda:3'), in_proj_covar=tensor([0.0066, 0.0062, 0.0065, 0.0057, 0.0072, 0.0067, 0.0084, 0.0059],
+       device='cuda:3'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-28 12:07:50,400 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0311, 2.2772, 2.3945, 2.1971, 2.6534, 2.6561, 2.9402, 3.1083],
+       device='cuda:3'), covar=tensor([0.0649, 0.1441, 0.1432, 0.1782, 0.1104, 0.1154, 0.0603, 0.0562],
+       device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0230, 0.0253, 0.0243, 0.0279, 0.0245, 0.0209, 0.0227],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 12:07:51,324 INFO [train.py:892] (3/4) Epoch 19, batch 1500, loss[loss=0.3342, simple_loss=0.382, pruned_loss=0.1432, over 19196.00 frames. ], tot_loss[loss=0.1904, simple_loss=0.2602, pruned_loss=0.06034, over 3948810.27 frames. ], batch size: 452, lr: 8.23e-03, grad_scale: 8.0
+2023-03-28 12:08:47,123 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.972e+02 4.472e+02 5.197e+02 6.398e+02 1.021e+03, threshold=1.039e+03, percent-clipped=2.0
+2023-03-28 12:09:15,920 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0158, 2.8066, 3.1063, 2.8715, 3.2739, 3.2755, 3.7710, 4.1549],
+       device='cuda:3'), covar=tensor([0.0555, 0.1701, 0.1473, 0.1943, 0.1761, 0.1387, 0.0572, 0.0580],
+       device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0230, 0.0253, 0.0244, 0.0280, 0.0244, 0.0210, 0.0227],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 12:09:19,969 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0931, 4.6537, 4.6894, 5.0589, 4.6840, 5.2503, 5.1745, 5.4001],
+       device='cuda:3'), covar=tensor([0.0667, 0.0432, 0.0436, 0.0355, 0.0662, 0.0432, 0.0398, 0.0362],
+       device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0161, 0.0185, 0.0156, 0.0159, 0.0141, 0.0140, 0.0180],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 12:09:43,733 INFO [train.py:892] (3/4) Epoch 19, batch 1550, loss[loss=0.2027, simple_loss=0.2741, pruned_loss=0.06567, over 19660.00 frames. ], tot_loss[loss=0.1921, simple_loss=0.2621, pruned_loss=0.06107, over 3946854.63 frames. ], batch size: 67, lr: 8.22e-03, grad_scale: 8.0
+2023-03-28 12:10:01,132 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:11:05,758 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34978.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 12:11:13,812 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:11:32,654 INFO [train.py:892] (3/4) Epoch 19, batch 1600, loss[loss=0.1711, simple_loss=0.2416, pruned_loss=0.0503, over 19893.00 frames. ], tot_loss[loss=0.1904, simple_loss=0.2601, pruned_loss=0.06034, over 3949103.57 frames. ], batch size: 87, lr: 8.22e-03, grad_scale: 8.0
+2023-03-28 12:12:30,719 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.006e+02 4.214e+02 4.806e+02 5.980e+02 1.376e+03, threshold=9.612e+02, percent-clipped=1.0
+2023-03-28 12:13:27,279 INFO [train.py:892] (3/4) Epoch 19, batch 1650, loss[loss=0.2025, simple_loss=0.2619, pruned_loss=0.07155, over 19827.00 frames. ], tot_loss[loss=0.1902, simple_loss=0.2598, pruned_loss=0.06027, over 3947801.15 frames. ], batch size: 202, lr: 8.21e-03, grad_scale: 8.0
+2023-03-28 12:14:22,945 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35066.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 12:14:49,686 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.68 vs. limit=2.0
+2023-03-28 12:15:21,220 INFO [train.py:892] (3/4) Epoch 19, batch 1700, loss[loss=0.2961, simple_loss=0.3592, pruned_loss=0.1166, over 19401.00 frames. ], tot_loss[loss=0.1928, simple_loss=0.2628, pruned_loss=0.06139, over 3944595.19 frames. ], batch size: 412, lr: 8.20e-03, grad_scale: 8.0
+2023-03-28 12:15:25,110 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:15:45,489 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8030, 2.3040, 2.6632, 3.0609, 3.4855, 3.5763, 3.5440, 3.6429],
+       device='cuda:3'), covar=tensor([0.0935, 0.1619, 0.1241, 0.0591, 0.0395, 0.0312, 0.0350, 0.0405],
+       device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0169, 0.0171, 0.0141, 0.0125, 0.0116, 0.0110, 0.0106],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 12:16:05,062 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5522, 4.8455, 4.9034, 4.8244, 4.5359, 4.8525, 4.4205, 4.4707],
+       device='cuda:3'), covar=tensor([0.0489, 0.0459, 0.0512, 0.0425, 0.0683, 0.0527, 0.0635, 0.0867],
+       device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0242, 0.0270, 0.0231, 0.0228, 0.0222, 0.0239, 0.0280],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 12:16:20,384 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.976e+02 3.844e+02 4.787e+02 6.177e+02 1.345e+03, threshold=9.574e+02, percent-clipped=3.0
+2023-03-28 12:17:09,579 INFO [train.py:892] (3/4) Epoch 19, batch 1750, loss[loss=0.2076, simple_loss=0.2968, pruned_loss=0.05915, over 19852.00 frames. ], tot_loss[loss=0.1917, simple_loss=0.2621, pruned_loss=0.06066, over 3945169.64 frames. ], batch size: 56, lr: 8.20e-03, grad_scale: 8.0
+2023-03-28 12:17:32,564 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35153.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:17:51,667 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5423, 2.7264, 3.8524, 3.0157, 3.3423, 3.1258, 2.1332, 2.2783],
+       device='cuda:3'), covar=tensor([0.0956, 0.2761, 0.0537, 0.0925, 0.1499, 0.1300, 0.2252, 0.2560],
+       device='cuda:3'), in_proj_covar=tensor([0.0337, 0.0368, 0.0319, 0.0258, 0.0361, 0.0333, 0.0340, 0.0309],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 12:18:36,657 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-28 12:18:42,353 INFO [train.py:892] (3/4) Epoch 19, batch 1800, loss[loss=0.1898, simple_loss=0.2555, pruned_loss=0.06207, over 19752.00 frames. ], tot_loss[loss=0.1921, simple_loss=0.2625, pruned_loss=0.06087, over 3944473.35 frames. ], batch size: 205, lr: 8.19e-03, grad_scale: 8.0
+2023-03-28 12:19:29,997 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.773e+02 3.767e+02 4.672e+02 5.957e+02 1.054e+03, threshold=9.344e+02, percent-clipped=2.0
+2023-03-28 12:20:07,729 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0871, 2.4651, 3.1127, 3.1521, 3.7122, 4.2506, 3.9701, 4.1067],
+       device='cuda:3'), covar=tensor([0.0846, 0.1875, 0.1198, 0.0694, 0.0386, 0.0194, 0.0306, 0.0484],
+       device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0172, 0.0174, 0.0144, 0.0127, 0.0118, 0.0112, 0.0108],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 12:20:12,656 INFO [train.py:892] (3/4) Epoch 19, batch 1850, loss[loss=0.2027, simple_loss=0.2817, pruned_loss=0.06178, over 19687.00 frames. ], tot_loss[loss=0.1927, simple_loss=0.2641, pruned_loss=0.06063, over 3945123.97 frames. ], batch size: 55, lr: 8.19e-03, grad_scale: 8.0
+2023-03-28 12:21:18,512 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-28 12:21:19,080 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-28 12:21:19,187 INFO [train.py:892] (3/4) Epoch 20, batch 0, loss[loss=0.1695, simple_loss=0.2429, pruned_loss=0.04801, over 19727.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2429, pruned_loss=0.04801, over 19727.00 frames. ], batch size: 47, lr: 7.98e-03, grad_scale: 8.0
+2023-03-28 12:21:19,188 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-28 12:21:49,712 INFO [train.py:926] (3/4) Epoch 20, validation: loss=0.1718, simple_loss=0.2485, pruned_loss=0.04755, over 2883724.00 frames. 
+2023-03-28 12:21:49,713 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-28 12:21:57,715 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:22:01,576 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.9165, 1.9632, 2.1554, 1.9768, 1.9351, 1.9531, 2.0549, 2.1761],
+       device='cuda:3'), covar=tensor([0.0348, 0.0261, 0.0218, 0.0240, 0.0385, 0.0297, 0.0360, 0.0213],
+       device='cuda:3'), in_proj_covar=tensor([0.0067, 0.0063, 0.0066, 0.0059, 0.0073, 0.0068, 0.0085, 0.0059],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-28 12:23:03,201 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35278.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 12:23:12,037 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:23:43,008 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2189, 4.1931, 4.5998, 4.1540, 3.9234, 4.4510, 4.1946, 4.6822],
+       device='cuda:3'), covar=tensor([0.0936, 0.0389, 0.0372, 0.0436, 0.0926, 0.0476, 0.0545, 0.0325],
+       device='cuda:3'), in_proj_covar=tensor([0.0277, 0.0217, 0.0213, 0.0222, 0.0204, 0.0225, 0.0222, 0.0205],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 12:23:44,082 INFO [train.py:892] (3/4) Epoch 20, batch 50, loss[loss=0.1794, simple_loss=0.2495, pruned_loss=0.05466, over 19896.00 frames. ], tot_loss[loss=0.1858, simple_loss=0.2556, pruned_loss=0.05802, over 889868.71 frames. ], batch size: 119, lr: 7.97e-03, grad_scale: 8.0
+2023-03-28 12:23:47,120 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:24:30,954 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.385e+02 4.131e+02 4.865e+02 6.041e+02 1.249e+03, threshold=9.730e+02, percent-clipped=3.0
+2023-03-28 12:24:53,317 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:25:03,352 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:25:04,179 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0
+2023-03-28 12:25:26,268 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0
+2023-03-28 12:25:38,653 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35345.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 12:25:39,714 INFO [train.py:892] (3/4) Epoch 20, batch 100, loss[loss=0.1752, simple_loss=0.247, pruned_loss=0.05169, over 19760.00 frames. ], tot_loss[loss=0.1865, simple_loss=0.2565, pruned_loss=0.05825, over 1568846.72 frames. ], batch size: 113, lr: 7.96e-03, grad_scale: 8.0
+2023-03-28 12:26:04,099 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7758, 2.9105, 3.0388, 2.2945, 3.0600, 2.6082, 2.7743, 3.0764],
+       device='cuda:3'), covar=tensor([0.0579, 0.0369, 0.0390, 0.0794, 0.0409, 0.0418, 0.0506, 0.0341],
+       device='cuda:3'), in_proj_covar=tensor([0.0067, 0.0075, 0.0073, 0.0102, 0.0068, 0.0069, 0.0068, 0.0060],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 12:26:25,085 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35366.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 12:26:42,403 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:27:26,022 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.00 vs. limit=5.0
+2023-03-28 12:27:28,858 INFO [train.py:892] (3/4) Epoch 20, batch 150, loss[loss=0.1954, simple_loss=0.2667, pruned_loss=0.06205, over 19788.00 frames. ], tot_loss[loss=0.189, simple_loss=0.2583, pruned_loss=0.05987, over 2097335.78 frames. ], batch size: 263, lr: 7.96e-03, grad_scale: 8.0
+2023-03-28 12:27:54,657 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35406.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 12:28:12,781 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35414.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 12:28:19,116 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.855e+02 4.347e+02 5.138e+02 6.356e+02 9.999e+02, threshold=1.028e+03, percent-clipped=1.0
+2023-03-28 12:28:58,662 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35435.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:29:25,240 INFO [train.py:892] (3/4) Epoch 20, batch 200, loss[loss=0.1887, simple_loss=0.2488, pruned_loss=0.0643, over 19828.00 frames. ], tot_loss[loss=0.1883, simple_loss=0.2585, pruned_loss=0.05907, over 2507376.26 frames. ], batch size: 202, lr: 7.95e-03, grad_scale: 8.0
+2023-03-28 12:29:30,558 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35448.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:31:20,209 INFO [train.py:892] (3/4) Epoch 20, batch 250, loss[loss=0.1477, simple_loss=0.222, pruned_loss=0.03669, over 19904.00 frames. ], tot_loss[loss=0.189, simple_loss=0.2595, pruned_loss=0.05927, over 2826061.29 frames. ], batch size: 113, lr: 7.95e-03, grad_scale: 8.0
+2023-03-28 12:32:02,738 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.970e+02 4.157e+02 5.011e+02 5.895e+02 1.207e+03, threshold=1.002e+03, percent-clipped=2.0
+2023-03-28 12:32:35,635 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:33:07,882 INFO [train.py:892] (3/4) Epoch 20, batch 300, loss[loss=0.1878, simple_loss=0.2603, pruned_loss=0.05765, over 19784.00 frames. ], tot_loss[loss=0.1898, simple_loss=0.2601, pruned_loss=0.05982, over 3073861.40 frames. ], batch size: 94, lr: 7.94e-03, grad_scale: 8.0
+2023-03-28 12:33:39,242 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0477, 5.1461, 5.2353, 5.2338, 4.9030, 5.2204, 4.7279, 4.3394],
+       device='cuda:3'), covar=tensor([0.0730, 0.0925, 0.0789, 0.0713, 0.0965, 0.0869, 0.1350, 0.2308],
+       device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0246, 0.0272, 0.0236, 0.0236, 0.0225, 0.0244, 0.0286],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 12:34:26,306 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-28 12:34:49,175 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:34:58,995 INFO [train.py:892] (3/4) Epoch 20, batch 350, loss[loss=0.2099, simple_loss=0.284, pruned_loss=0.06787, over 19810.00 frames. ], tot_loss[loss=0.1902, simple_loss=0.2609, pruned_loss=0.05974, over 3268302.17 frames. ], batch size: 65, lr: 7.94e-03, grad_scale: 8.0
+2023-03-28 12:35:32,686 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7834, 3.4251, 3.5939, 3.7918, 3.5499, 3.6611, 3.8493, 4.0199],
+       device='cuda:3'), covar=tensor([0.0667, 0.0437, 0.0533, 0.0354, 0.0729, 0.0621, 0.0447, 0.0315],
+       device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0160, 0.0184, 0.0157, 0.0159, 0.0141, 0.0139, 0.0180],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 12:35:45,653 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.727e+02 4.459e+02 5.252e+02 6.207e+02 1.161e+03, threshold=1.050e+03, percent-clipped=2.0
+2023-03-28 12:36:55,517 INFO [train.py:892] (3/4) Epoch 20, batch 400, loss[loss=0.2112, simple_loss=0.2786, pruned_loss=0.07194, over 19685.00 frames. ], tot_loss[loss=0.1892, simple_loss=0.2605, pruned_loss=0.05894, over 3418517.77 frames. ], batch size: 315, lr: 7.93e-03, grad_scale: 8.0
+2023-03-28 12:38:32,198 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7478, 2.8383, 4.1188, 3.2160, 3.5296, 3.3288, 2.3127, 2.3887],
+       device='cuda:3'), covar=tensor([0.0882, 0.2880, 0.0552, 0.0902, 0.1432, 0.1244, 0.2208, 0.2753],
+       device='cuda:3'), in_proj_covar=tensor([0.0334, 0.0364, 0.0319, 0.0258, 0.0357, 0.0333, 0.0338, 0.0308],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 12:38:46,379 INFO [train.py:892] (3/4) Epoch 20, batch 450, loss[loss=0.1907, simple_loss=0.2641, pruned_loss=0.05863, over 19798.00 frames. ], tot_loss[loss=0.1892, simple_loss=0.2608, pruned_loss=0.05886, over 3536544.43 frames. ], batch size: 51, lr: 7.93e-03, grad_scale: 8.0
+2023-03-28 12:38:58,529 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35701.0, num_to_drop=1, layers_to_drop={3}
+2023-03-28 12:39:22,313 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6299, 3.7555, 3.9310, 4.7497, 3.0543, 3.4824, 2.8538, 2.8734],
+       device='cuda:3'), covar=tensor([0.0416, 0.2363, 0.0899, 0.0284, 0.2096, 0.0894, 0.1297, 0.1754],
+       device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0334, 0.0238, 0.0181, 0.0241, 0.0194, 0.0207, 0.0215],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 12:39:31,593 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.836e+02 4.301e+02 4.934e+02 5.630e+02 1.538e+03, threshold=9.868e+02, percent-clipped=1.0
+2023-03-28 12:39:58,941 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35730.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:40:31,620 INFO [train.py:892] (3/4) Epoch 20, batch 500, loss[loss=0.1733, simple_loss=0.2411, pruned_loss=0.0528, over 19813.00 frames. ], tot_loss[loss=0.19, simple_loss=0.2609, pruned_loss=0.05962, over 3628279.52 frames. ], batch size: 167, lr: 7.92e-03, grad_scale: 8.0
+2023-03-28 12:40:37,956 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:41:26,205 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:42:19,107 INFO [train.py:892] (3/4) Epoch 20, batch 550, loss[loss=0.1793, simple_loss=0.2512, pruned_loss=0.05375, over 19832.00 frames. ], tot_loss[loss=0.1892, simple_loss=0.2599, pruned_loss=0.05926, over 3700100.48 frames. ], batch size: 143, lr: 7.92e-03, grad_scale: 8.0
+2023-03-28 12:42:19,809 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:42:22,725 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-28 12:42:34,018 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7480, 3.7044, 4.0495, 3.6876, 3.5712, 3.9807, 3.7617, 4.1256],
+       device='cuda:3'), covar=tensor([0.0937, 0.0399, 0.0389, 0.0435, 0.1111, 0.0497, 0.0561, 0.0361],
+       device='cuda:3'), in_proj_covar=tensor([0.0277, 0.0216, 0.0213, 0.0222, 0.0200, 0.0224, 0.0222, 0.0205],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 12:42:51,821 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:43:03,595 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6785, 4.0029, 3.9479, 4.9207, 3.0160, 3.5930, 3.2413, 2.9250],
+       device='cuda:3'), covar=tensor([0.0425, 0.2110, 0.0909, 0.0304, 0.2222, 0.0928, 0.1130, 0.1781],
+       device='cuda:3'), in_proj_covar=tensor([0.0225, 0.0333, 0.0237, 0.0181, 0.0240, 0.0194, 0.0207, 0.0214],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 12:43:04,503 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.312e+02 4.108e+02 4.834e+02 5.739e+02 9.285e+02, threshold=9.669e+02, percent-clipped=0.0
+2023-03-28 12:43:11,742 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0
+2023-03-28 12:43:18,621 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.11 vs. limit=2.0
+2023-03-28 12:43:35,155 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6236, 3.6426, 2.2323, 4.5129, 3.8464, 4.3948, 4.4870, 3.4313],
+       device='cuda:3'), covar=tensor([0.0534, 0.0542, 0.1471, 0.0471, 0.0506, 0.0278, 0.0537, 0.0716],
+       device='cuda:3'), in_proj_covar=tensor([0.0134, 0.0134, 0.0138, 0.0136, 0.0121, 0.0121, 0.0133, 0.0137],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 12:43:39,415 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:44:08,189 INFO [train.py:892] (3/4) Epoch 20, batch 600, loss[loss=0.143, simple_loss=0.2182, pruned_loss=0.03393, over 19796.00 frames. ], tot_loss[loss=0.1881, simple_loss=0.2591, pruned_loss=0.0586, over 3756907.73 frames. ], batch size: 86, lr: 7.91e-03, grad_scale: 8.0
+2023-03-28 12:45:06,792 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:45:28,101 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7131, 4.6610, 5.1241, 4.6969, 4.1718, 4.8597, 4.8126, 5.2304],
+       device='cuda:3'), covar=tensor([0.0860, 0.0337, 0.0343, 0.0347, 0.0780, 0.0419, 0.0381, 0.0298],
+       device='cuda:3'), in_proj_covar=tensor([0.0278, 0.0217, 0.0214, 0.0223, 0.0202, 0.0225, 0.0223, 0.0207],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 12:45:36,174 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:45:54,699 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.1123, 5.2535, 5.3487, 5.3472, 4.9522, 5.3436, 4.8007, 4.5080],
+       device='cuda:3'), covar=tensor([0.0818, 0.0968, 0.0909, 0.0679, 0.1026, 0.0947, 0.1556, 0.2257],
+       device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0242, 0.0266, 0.0231, 0.0231, 0.0223, 0.0238, 0.0279],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 12:45:55,670 INFO [train.py:892] (3/4) Epoch 20, batch 650, loss[loss=0.1725, simple_loss=0.2455, pruned_loss=0.04975, over 19782.00 frames. ], tot_loss[loss=0.1886, simple_loss=0.2595, pruned_loss=0.05884, over 3798549.76 frames. ], batch size: 91, lr: 7.90e-03, grad_scale: 8.0
+2023-03-28 12:46:04,798 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.51 vs. limit=5.0
+2023-03-28 12:46:10,836 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8323, 4.0072, 2.3810, 4.1298, 4.3244, 1.8961, 3.4559, 3.2667],
+       device='cuda:3'), covar=tensor([0.0650, 0.0688, 0.2496, 0.0792, 0.0464, 0.2763, 0.1138, 0.0745],
+       device='cuda:3'), in_proj_covar=tensor([0.0220, 0.0242, 0.0223, 0.0249, 0.0220, 0.0202, 0.0230, 0.0181],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 12:46:29,831 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4861, 4.7782, 4.8095, 4.6915, 4.4686, 4.7840, 4.2849, 4.4293],
+       device='cuda:3'), covar=tensor([0.0448, 0.0418, 0.0456, 0.0419, 0.0588, 0.0520, 0.0641, 0.0819],
+       device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0242, 0.0266, 0.0231, 0.0232, 0.0223, 0.0239, 0.0279],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 12:46:42,645 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.720e+02 4.236e+02 5.120e+02 6.447e+02 1.973e+03, threshold=1.024e+03, percent-clipped=4.0
+2023-03-28 12:47:06,721 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5770, 2.9651, 3.3613, 3.0700, 3.7056, 3.7593, 4.3242, 4.8814],
+       device='cuda:3'), covar=tensor([0.0499, 0.1826, 0.1561, 0.2105, 0.1795, 0.1284, 0.0559, 0.0553],
+       device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0232, 0.0253, 0.0245, 0.0280, 0.0245, 0.0213, 0.0229],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 12:47:18,354 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.2723, 1.5902, 1.8655, 2.5559, 2.8219, 2.9425, 2.8400, 2.9535],
+       device='cuda:3'), covar=tensor([0.1062, 0.2128, 0.1620, 0.0699, 0.0515, 0.0373, 0.0387, 0.0377],
+       device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0167, 0.0170, 0.0141, 0.0124, 0.0116, 0.0110, 0.0106],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 12:47:47,276 INFO [train.py:892] (3/4) Epoch 20, batch 700, loss[loss=0.1762, simple_loss=0.2352, pruned_loss=0.0586, over 19749.00 frames. ], tot_loss[loss=0.1889, simple_loss=0.2596, pruned_loss=0.05913, over 3832780.98 frames. ], batch size: 129, lr: 7.90e-03, grad_scale: 8.0
+2023-03-28 12:49:26,349 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-28 12:49:30,463 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0
+2023-03-28 12:49:38,998 INFO [train.py:892] (3/4) Epoch 20, batch 750, loss[loss=0.1819, simple_loss=0.262, pruned_loss=0.05092, over 19849.00 frames. ], tot_loss[loss=0.1889, simple_loss=0.2598, pruned_loss=0.05897, over 3859696.39 frames. ], batch size: 43, lr: 7.89e-03, grad_scale: 8.0
+2023-03-28 12:49:57,238 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36001.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 12:50:31,044 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.681e+02 4.396e+02 5.340e+02 6.377e+02 1.000e+03, threshold=1.068e+03, percent-clipped=0.0
+2023-03-28 12:51:02,872 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:51:36,013 INFO [train.py:892] (3/4) Epoch 20, batch 800, loss[loss=0.1903, simple_loss=0.2664, pruned_loss=0.05714, over 19804.00 frames. ], tot_loss[loss=0.1889, simple_loss=0.2595, pruned_loss=0.05914, over 3881036.99 frames. ], batch size: 107, lr: 7.89e-03, grad_scale: 8.0
+2023-03-28 12:51:39,422 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-28 12:51:42,560 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36049.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 12:51:48,666 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4936, 5.8454, 6.0379, 5.7913, 5.6661, 5.4598, 5.5556, 5.5590],
+       device='cuda:3'), covar=tensor([0.1298, 0.0944, 0.0785, 0.1011, 0.0581, 0.0824, 0.2035, 0.1702],
+       device='cuda:3'), in_proj_covar=tensor([0.0273, 0.0289, 0.0336, 0.0272, 0.0253, 0.0249, 0.0325, 0.0358],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 12:52:48,786 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:52:51,397 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.0194, 2.0392, 2.1736, 2.1195, 2.0550, 2.1261, 2.0655, 2.2364],
+       device='cuda:3'), covar=tensor([0.0296, 0.0280, 0.0240, 0.0218, 0.0348, 0.0270, 0.0368, 0.0219],
+       device='cuda:3'), in_proj_covar=tensor([0.0068, 0.0064, 0.0067, 0.0059, 0.0073, 0.0068, 0.0086, 0.0060],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-28 12:53:26,923 INFO [train.py:892] (3/4) Epoch 20, batch 850, loss[loss=0.1722, simple_loss=0.2443, pruned_loss=0.05005, over 19807.00 frames. ], tot_loss[loss=0.1895, simple_loss=0.2606, pruned_loss=0.05919, over 3894840.36 frames. ], batch size: 86, lr: 7.88e-03, grad_scale: 8.0
+2023-03-28 12:53:34,718 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.1694, 4.7669, 4.9237, 5.1321, 4.6289, 5.3559, 5.2792, 5.4271],
+       device='cuda:3'), covar=tensor([0.0600, 0.0340, 0.0355, 0.0279, 0.0634, 0.0341, 0.0358, 0.0315],
+       device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0161, 0.0185, 0.0159, 0.0160, 0.0142, 0.0140, 0.0182],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 12:54:15,633 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.913e+02 4.436e+02 5.161e+02 6.355e+02 1.299e+03, threshold=1.032e+03, percent-clipped=1.0
+2023-03-28 12:54:36,501 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:54:58,737 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9469, 4.9860, 5.3138, 5.1148, 5.1166, 4.5347, 5.0170, 4.8822],
+       device='cuda:3'), covar=tensor([0.1447, 0.1295, 0.0969, 0.1221, 0.0800, 0.1040, 0.1957, 0.2008],
+       device='cuda:3'), in_proj_covar=tensor([0.0272, 0.0289, 0.0338, 0.0272, 0.0252, 0.0249, 0.0326, 0.0358],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 12:55:19,442 INFO [train.py:892] (3/4) Epoch 20, batch 900, loss[loss=0.3489, simple_loss=0.3902, pruned_loss=0.1538, over 19184.00 frames. ], tot_loss[loss=0.1891, simple_loss=0.2601, pruned_loss=0.05906, over 3907725.43 frames. ], batch size: 452, lr: 7.88e-03, grad_scale: 8.0
+2023-03-28 12:56:08,496 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:56:39,321 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:56:51,734 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:57:11,795 INFO [train.py:892] (3/4) Epoch 20, batch 950, loss[loss=0.1939, simple_loss=0.261, pruned_loss=0.06345, over 19795.00 frames. ], tot_loss[loss=0.1886, simple_loss=0.2595, pruned_loss=0.05889, over 3917966.04 frames. ], batch size: 185, lr: 7.87e-03, grad_scale: 8.0
+2023-03-28 12:57:58,659 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.907e+02 4.513e+02 5.155e+02 5.945e+02 1.102e+03, threshold=1.031e+03, percent-clipped=1.0
+2023-03-28 12:58:39,630 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:58:57,905 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 12:59:05,340 INFO [train.py:892] (3/4) Epoch 20, batch 1000, loss[loss=0.1842, simple_loss=0.2581, pruned_loss=0.05509, over 19758.00 frames. ], tot_loss[loss=0.188, simple_loss=0.2589, pruned_loss=0.05857, over 3925412.45 frames. ], batch size: 253, lr: 7.87e-03, grad_scale: 8.0
+2023-03-28 13:01:03,799 INFO [train.py:892] (3/4) Epoch 20, batch 1050, loss[loss=0.1996, simple_loss=0.2722, pruned_loss=0.06351, over 19802.00 frames. ], tot_loss[loss=0.1872, simple_loss=0.2581, pruned_loss=0.05817, over 3931425.12 frames. ], batch size: 67, lr: 7.86e-03, grad_scale: 16.0
+2023-03-28 13:01:49,680 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.691e+02 4.131e+02 5.046e+02 5.738e+02 1.025e+03, threshold=1.009e+03, percent-clipped=0.0
+2023-03-28 13:02:56,403 INFO [train.py:892] (3/4) Epoch 20, batch 1100, loss[loss=0.174, simple_loss=0.2358, pruned_loss=0.05607, over 19783.00 frames. ], tot_loss[loss=0.1878, simple_loss=0.2586, pruned_loss=0.05846, over 3935952.84 frames. ], batch size: 168, lr: 7.86e-03, grad_scale: 16.0
+2023-03-28 13:03:40,278 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8173, 2.7718, 4.5304, 3.8968, 4.4246, 4.4672, 4.3842, 4.2753],
+       device='cuda:3'), covar=tensor([0.0335, 0.0828, 0.0096, 0.0846, 0.0123, 0.0206, 0.0151, 0.0140],
+       device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0097, 0.0080, 0.0148, 0.0075, 0.0090, 0.0083, 0.0077],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0002, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 13:04:48,944 INFO [train.py:892] (3/4) Epoch 20, batch 1150, loss[loss=0.194, simple_loss=0.2717, pruned_loss=0.05821, over 19813.00 frames. ], tot_loss[loss=0.1878, simple_loss=0.2588, pruned_loss=0.05846, over 3939683.20 frames. ], batch size: 82, lr: 7.85e-03, grad_scale: 16.0
+2023-03-28 13:05:36,784 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.688e+02 4.392e+02 5.039e+02 5.876e+02 1.119e+03, threshold=1.008e+03, percent-clipped=1.0
+2023-03-28 13:06:01,301 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36427.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:06:08,036 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0
+2023-03-28 13:06:43,820 INFO [train.py:892] (3/4) Epoch 20, batch 1200, loss[loss=0.1681, simple_loss=0.2416, pruned_loss=0.04729, over 19875.00 frames. ], tot_loss[loss=0.1884, simple_loss=0.2593, pruned_loss=0.05871, over 3939598.14 frames. ], batch size: 159, lr: 7.85e-03, grad_scale: 16.0
+2023-03-28 13:07:31,500 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36467.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:07:50,177 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36475.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:08:37,661 INFO [train.py:892] (3/4) Epoch 20, batch 1250, loss[loss=0.1955, simple_loss=0.2667, pruned_loss=0.06213, over 19760.00 frames. ], tot_loss[loss=0.1877, simple_loss=0.2588, pruned_loss=0.05826, over 3943126.11 frames. ], batch size: 276, lr: 7.84e-03, grad_scale: 16.0
+2023-03-28 13:09:22,501 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:09:27,467 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.037e+02 4.208e+02 4.891e+02 6.117e+02 9.708e+02, threshold=9.782e+02, percent-clipped=0.0
+2023-03-28 13:10:11,609 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36537.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:10:29,788 INFO [train.py:892] (3/4) Epoch 20, batch 1300, loss[loss=0.1657, simple_loss=0.2515, pruned_loss=0.03993, over 19832.00 frames. ], tot_loss[loss=0.1857, simple_loss=0.2569, pruned_loss=0.05726, over 3945055.38 frames. ], batch size: 52, lr: 7.84e-03, grad_scale: 16.0
+2023-03-28 13:12:24,062 INFO [train.py:892] (3/4) Epoch 20, batch 1350, loss[loss=0.1955, simple_loss=0.2665, pruned_loss=0.06231, over 19724.00 frames. ], tot_loss[loss=0.1862, simple_loss=0.2574, pruned_loss=0.05749, over 3946926.33 frames. ], batch size: 61, lr: 7.83e-03, grad_scale: 16.0
+2023-03-28 13:12:45,788 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:12:51,740 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36608.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:13:02,312 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:13:12,351 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.844e+02 4.275e+02 5.272e+02 6.307e+02 1.102e+03, threshold=1.054e+03, percent-clipped=1.0
+2023-03-28 13:14:16,024 INFO [train.py:892] (3/4) Epoch 20, batch 1400, loss[loss=0.2036, simple_loss=0.2881, pruned_loss=0.05951, over 19687.00 frames. ], tot_loss[loss=0.1849, simple_loss=0.2561, pruned_loss=0.05692, over 3947607.03 frames. ], batch size: 55, lr: 7.82e-03, grad_scale: 16.0
+2023-03-28 13:15:03,578 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:15:09,216 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:15:16,949 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36673.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:15:49,481 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9017, 4.5765, 4.5613, 4.8949, 4.7359, 5.2263, 4.8868, 5.0529],
+       device='cuda:3'), covar=tensor([0.0768, 0.0415, 0.0616, 0.0390, 0.0616, 0.0375, 0.0661, 0.0607],
+       device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0162, 0.0185, 0.0160, 0.0159, 0.0145, 0.0142, 0.0181],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 13:16:06,755 INFO [train.py:892] (3/4) Epoch 20, batch 1450, loss[loss=0.194, simple_loss=0.2628, pruned_loss=0.06257, over 19827.00 frames. ], tot_loss[loss=0.1856, simple_loss=0.257, pruned_loss=0.05708, over 3949539.80 frames. ], batch size: 103, lr: 7.82e-03, grad_scale: 16.0
+2023-03-28 13:16:17,618 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36700.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 13:16:29,631 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2087, 3.8402, 3.9435, 4.1494, 3.8523, 4.2169, 4.3591, 4.5046],
+       device='cuda:3'), covar=tensor([0.0636, 0.0385, 0.0514, 0.0336, 0.0734, 0.0478, 0.0371, 0.0281],
+       device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0163, 0.0186, 0.0160, 0.0160, 0.0145, 0.0142, 0.0182],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 13:16:54,915 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.742e+02 4.075e+02 4.928e+02 6.205e+02 1.251e+03, threshold=9.857e+02, percent-clipped=1.0
+2023-03-28 13:17:56,074 INFO [train.py:892] (3/4) Epoch 20, batch 1500, loss[loss=0.2667, simple_loss=0.3295, pruned_loss=0.102, over 19610.00 frames. ], tot_loss[loss=0.187, simple_loss=0.258, pruned_loss=0.05797, over 3948709.71 frames. ], batch size: 367, lr: 7.81e-03, grad_scale: 16.0
+2023-03-28 13:18:30,398 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36761.0, num_to_drop=1, layers_to_drop={3}
+2023-03-28 13:19:49,250 INFO [train.py:892] (3/4) Epoch 20, batch 1550, loss[loss=0.1842, simple_loss=0.2624, pruned_loss=0.05296, over 19603.00 frames. ], tot_loss[loss=0.1862, simple_loss=0.2578, pruned_loss=0.0573, over 3949019.66 frames. ], batch size: 50, lr: 7.81e-03, grad_scale: 16.0
+2023-03-28 13:20:10,041 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36805.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 13:20:26,784 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.77 vs. limit=5.0
+2023-03-28 13:20:35,963 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.615e+02 4.163e+02 4.935e+02 6.337e+02 1.233e+03, threshold=9.870e+02, percent-clipped=4.0
+2023-03-28 13:21:03,998 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0
+2023-03-28 13:21:24,671 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36837.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:21:43,528 INFO [train.py:892] (3/4) Epoch 20, batch 1600, loss[loss=0.1816, simple_loss=0.2519, pruned_loss=0.05569, over 19795.00 frames. ], tot_loss[loss=0.1857, simple_loss=0.2572, pruned_loss=0.05714, over 3949648.35 frames. ], batch size: 185, lr: 7.80e-03, grad_scale: 16.0
+2023-03-28 13:22:33,937 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36866.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 13:23:15,929 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36885.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:23:38,877 INFO [train.py:892] (3/4) Epoch 20, batch 1650, loss[loss=0.1869, simple_loss=0.2528, pruned_loss=0.06044, over 19749.00 frames. ], tot_loss[loss=0.186, simple_loss=0.2572, pruned_loss=0.05742, over 3950482.21 frames. ], batch size: 139, lr: 7.80e-03, grad_scale: 16.0
+2023-03-28 13:24:13,797 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36911.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:24:26,324 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.686e+02 3.920e+02 4.713e+02 5.573e+02 1.416e+03, threshold=9.427e+02, percent-clipped=1.0
+2023-03-28 13:24:48,849 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8551, 3.2923, 3.7012, 3.4740, 4.0534, 4.0196, 4.6536, 5.1815],
+       device='cuda:3'), covar=tensor([0.0453, 0.1589, 0.1276, 0.1820, 0.1606, 0.1275, 0.0466, 0.0354],
+       device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0233, 0.0254, 0.0247, 0.0282, 0.0247, 0.0214, 0.0232],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 13:24:55,120 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0
+2023-03-28 13:25:31,479 INFO [train.py:892] (3/4) Epoch 20, batch 1700, loss[loss=0.1667, simple_loss=0.2465, pruned_loss=0.04348, over 19802.00 frames. ], tot_loss[loss=0.1863, simple_loss=0.2577, pruned_loss=0.0574, over 3950767.76 frames. ], batch size: 67, lr: 7.79e-03, grad_scale: 16.0
+2023-03-28 13:25:58,524 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:26:04,539 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36961.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:26:10,893 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:26:21,331 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:26:29,912 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:27:17,392 INFO [train.py:892] (3/4) Epoch 20, batch 1750, loss[loss=0.3493, simple_loss=0.3957, pruned_loss=0.1514, over 19420.00 frames. ], tot_loss[loss=0.1868, simple_loss=0.2579, pruned_loss=0.05784, over 3949907.89 frames. ], batch size: 431, lr: 7.79e-03, grad_scale: 16.0
+2023-03-28 13:27:35,614 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:28:00,088 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.609e+02 4.040e+02 4.922e+02 6.023e+02 1.252e+03, threshold=9.844e+02, percent-clipped=2.0
+2023-03-28 13:28:01,174 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.11 vs. limit=2.0
+2023-03-28 13:28:04,755 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:28:29,564 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.2563, 5.7905, 5.9060, 5.6868, 5.4997, 5.4680, 5.5433, 5.4166],
+       device='cuda:3'), covar=tensor([0.1540, 0.1222, 0.0829, 0.1214, 0.0671, 0.0738, 0.1730, 0.2021],
+       device='cuda:3'), in_proj_covar=tensor([0.0273, 0.0293, 0.0340, 0.0274, 0.0256, 0.0252, 0.0329, 0.0362],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 13:28:55,518 INFO [train.py:892] (3/4) Epoch 20, batch 1800, loss[loss=0.2071, simple_loss=0.2752, pruned_loss=0.06954, over 19764.00 frames. ], tot_loss[loss=0.1864, simple_loss=0.2576, pruned_loss=0.05759, over 3950073.88 frames. ], batch size: 244, lr: 7.78e-03, grad_scale: 16.0
+2023-03-28 13:29:13,216 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37056.0, num_to_drop=1, layers_to_drop={3}
+2023-03-28 13:29:29,218 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:30:21,867 INFO [train.py:892] (3/4) Epoch 20, batch 1850, loss[loss=0.1978, simple_loss=0.2672, pruned_loss=0.0642, over 19689.00 frames. ], tot_loss[loss=0.1864, simple_loss=0.2582, pruned_loss=0.05732, over 3949782.34 frames. ], batch size: 55, lr: 7.78e-03, grad_scale: 16.0
+2023-03-28 13:31:23,899 INFO [train.py:892] (3/4) Epoch 21, batch 0, loss[loss=0.1684, simple_loss=0.241, pruned_loss=0.04794, over 19778.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.241, pruned_loss=0.04794, over 19778.00 frames. ], batch size: 91, lr: 7.59e-03, grad_scale: 16.0
+2023-03-28 13:31:23,900 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-28 13:31:56,332 INFO [train.py:926] (3/4) Epoch 21, validation: loss=0.1717, simple_loss=0.248, pruned_loss=0.04765, over 2883724.00 frames. 
+2023-03-28 13:31:56,333 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-28 13:32:31,700 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.932e+02 4.162e+02 4.969e+02 6.097e+02 9.968e+02, threshold=9.939e+02, percent-clipped=2.0
+2023-03-28 13:33:08,814 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-28 13:33:23,773 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4240, 3.4300, 2.1620, 4.2940, 3.8188, 4.1757, 4.2117, 3.2884],
+       device='cuda:3'), covar=tensor([0.0543, 0.0519, 0.1414, 0.0515, 0.0443, 0.0394, 0.0469, 0.0675],
+       device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0135, 0.0140, 0.0139, 0.0121, 0.0121, 0.0135, 0.0137],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 13:33:48,513 INFO [train.py:892] (3/4) Epoch 21, batch 50, loss[loss=0.2119, simple_loss=0.2808, pruned_loss=0.0715, over 19700.00 frames. ], tot_loss[loss=0.1884, simple_loss=0.2598, pruned_loss=0.05847, over 888157.00 frames. ], batch size: 283, lr: 7.58e-03, grad_scale: 16.0
+2023-03-28 13:34:10,847 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37161.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 13:34:15,175 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0
+2023-03-28 13:35:40,043 INFO [train.py:892] (3/4) Epoch 21, batch 100, loss[loss=0.1685, simple_loss=0.2519, pruned_loss=0.04259, over 19844.00 frames. ], tot_loss[loss=0.1829, simple_loss=0.2549, pruned_loss=0.05543, over 1567110.57 frames. ], batch size: 90, lr: 7.58e-03, grad_scale: 16.0
+2023-03-28 13:36:14,434 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.842e+02 4.106e+02 4.671e+02 5.605e+02 1.435e+03, threshold=9.342e+02, percent-clipped=1.0
+2023-03-28 13:37:33,178 INFO [train.py:892] (3/4) Epoch 21, batch 150, loss[loss=0.2478, simple_loss=0.311, pruned_loss=0.09229, over 19623.00 frames. ], tot_loss[loss=0.1847, simple_loss=0.2564, pruned_loss=0.05649, over 2092272.63 frames. ], batch size: 359, lr: 7.57e-03, grad_scale: 16.0
+2023-03-28 13:37:57,930 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37261.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:38:04,156 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:38:11,750 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37267.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:38:14,028 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:38:14,102 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37268.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:39:25,436 INFO [train.py:892] (3/4) Epoch 21, batch 200, loss[loss=0.1885, simple_loss=0.2691, pruned_loss=0.05397, over 19563.00 frames. ], tot_loss[loss=0.1858, simple_loss=0.2578, pruned_loss=0.05691, over 2504205.96 frames. ], batch size: 53, lr: 7.57e-03, grad_scale: 16.0
+2023-03-28 13:39:28,917 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.13 vs. limit=5.0
+2023-03-28 13:39:45,074 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37309.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:39:49,709 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.11 vs. limit=5.0
+2023-03-28 13:39:51,126 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:39:55,559 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37314.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:40:00,791 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37316.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:40:02,146 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.481e+02 4.044e+02 5.052e+02 5.778e+02 1.190e+03, threshold=1.010e+03, percent-clipped=3.0
+2023-03-28 13:40:32,402 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:41:19,122 INFO [train.py:892] (3/4) Epoch 21, batch 250, loss[loss=0.1531, simple_loss=0.2179, pruned_loss=0.04413, over 19843.00 frames. ], tot_loss[loss=0.1852, simple_loss=0.2575, pruned_loss=0.0565, over 2824569.62 frames. ], batch size: 109, lr: 7.56e-03, grad_scale: 16.0
+2023-03-28 13:41:30,873 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37356.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 13:41:39,260 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:41:47,403 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:42:45,601 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-28 13:43:12,247 INFO [train.py:892] (3/4) Epoch 21, batch 300, loss[loss=0.1708, simple_loss=0.2405, pruned_loss=0.05053, over 19847.00 frames. ], tot_loss[loss=0.1844, simple_loss=0.2565, pruned_loss=0.05615, over 3075148.98 frames. ], batch size: 124, lr: 7.56e-03, grad_scale: 16.0
+2023-03-28 13:43:21,043 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37404.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 13:43:49,249 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.947e+02 4.191e+02 4.785e+02 5.606e+02 9.286e+02, threshold=9.571e+02, percent-clipped=0.0
+2023-03-28 13:44:06,942 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:44:26,650 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0
+2023-03-28 13:45:03,289 INFO [train.py:892] (3/4) Epoch 21, batch 350, loss[loss=0.1724, simple_loss=0.2419, pruned_loss=0.05141, over 19746.00 frames. ], tot_loss[loss=0.1861, simple_loss=0.258, pruned_loss=0.05716, over 3269321.22 frames. ], batch size: 139, lr: 7.55e-03, grad_scale: 16.0
+2023-03-28 13:45:26,288 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37461.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 13:46:57,110 INFO [train.py:892] (3/4) Epoch 21, batch 400, loss[loss=0.1828, simple_loss=0.2509, pruned_loss=0.05735, over 19779.00 frames. ], tot_loss[loss=0.1861, simple_loss=0.2576, pruned_loss=0.0573, over 3420816.33 frames. ], batch size: 217, lr: 7.55e-03, grad_scale: 16.0
+2023-03-28 13:47:14,269 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37509.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 13:47:33,497 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.804e+02 3.892e+02 4.730e+02 5.808e+02 1.803e+03, threshold=9.460e+02, percent-clipped=3.0
+2023-03-28 13:48:49,652 INFO [train.py:892] (3/4) Epoch 21, batch 450, loss[loss=0.1975, simple_loss=0.2828, pruned_loss=0.05613, over 19850.00 frames. ], tot_loss[loss=0.1874, simple_loss=0.2589, pruned_loss=0.05798, over 3538081.66 frames. ], batch size: 56, lr: 7.54e-03, grad_scale: 16.0
+2023-03-28 13:49:05,493 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-28 13:49:28,699 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:50:43,317 INFO [train.py:892] (3/4) Epoch 21, batch 500, loss[loss=0.193, simple_loss=0.2557, pruned_loss=0.06511, over 19767.00 frames. ], tot_loss[loss=0.1862, simple_loss=0.2575, pruned_loss=0.05751, over 3630822.41 frames. ], batch size: 217, lr: 7.54e-03, grad_scale: 16.0
+2023-03-28 13:50:47,159 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.58 vs. limit=5.0
+2023-03-28 13:51:16,914 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:51:19,013 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:51:22,189 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.832e+02 3.951e+02 4.577e+02 5.485e+02 9.092e+02, threshold=9.154e+02, percent-clipped=0.0
+2023-03-28 13:51:37,842 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37624.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:51:51,136 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5593, 2.8685, 2.5378, 2.0602, 2.6096, 2.9155, 2.8113, 2.8289],
+       device='cuda:3'), covar=tensor([0.0262, 0.0311, 0.0259, 0.0546, 0.0320, 0.0188, 0.0210, 0.0182],
+       device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0079, 0.0087, 0.0091, 0.0093, 0.0069, 0.0069, 0.0070],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 13:52:32,611 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0
+2023-03-28 13:52:39,178 INFO [train.py:892] (3/4) Epoch 21, batch 550, loss[loss=0.1699, simple_loss=0.2462, pruned_loss=0.04679, over 19692.00 frames. ], tot_loss[loss=0.1854, simple_loss=0.2567, pruned_loss=0.0571, over 3702721.72 frames. ], batch size: 75, lr: 7.53e-03, grad_scale: 16.0
+2023-03-28 13:52:59,476 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:53:03,605 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37662.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:54:32,546 INFO [train.py:892] (3/4) Epoch 21, batch 600, loss[loss=0.1824, simple_loss=0.2454, pruned_loss=0.05972, over 19790.00 frames. ], tot_loss[loss=0.1859, simple_loss=0.257, pruned_loss=0.05742, over 3757588.82 frames. ], batch size: 172, lr: 7.53e-03, grad_scale: 16.0
+2023-03-28 13:54:49,156 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:55:09,557 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.449e+02 4.150e+02 4.838e+02 5.835e+02 9.838e+02, threshold=9.675e+02, percent-clipped=3.0
+2023-03-28 13:55:17,247 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 13:56:00,202 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0
+2023-03-28 13:56:25,854 INFO [train.py:892] (3/4) Epoch 21, batch 650, loss[loss=0.1629, simple_loss=0.2449, pruned_loss=0.04045, over 19781.00 frames. ], tot_loss[loss=0.1849, simple_loss=0.256, pruned_loss=0.05693, over 3801406.79 frames. ], batch size: 52, lr: 7.52e-03, grad_scale: 16.0
+2023-03-28 13:57:17,252 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0384, 4.1237, 2.5317, 4.3522, 4.5545, 1.9858, 3.7604, 3.3252],
+       device='cuda:3'), covar=tensor([0.0697, 0.0807, 0.2488, 0.0883, 0.0544, 0.2718, 0.1040, 0.0791],
+       device='cuda:3'), in_proj_covar=tensor([0.0221, 0.0242, 0.0223, 0.0255, 0.0226, 0.0200, 0.0231, 0.0182],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 13:58:19,132 INFO [train.py:892] (3/4) Epoch 21, batch 700, loss[loss=0.1589, simple_loss=0.2401, pruned_loss=0.03885, over 19765.00 frames. ], tot_loss[loss=0.1861, simple_loss=0.2576, pruned_loss=0.05728, over 3833989.24 frames. ], batch size: 100, lr: 7.52e-03, grad_scale: 16.0
+2023-03-28 13:58:55,422 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.050e+02 4.218e+02 4.924e+02 5.585e+02 8.798e+02, threshold=9.849e+02, percent-clipped=0.0
+2023-03-28 13:59:31,737 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4913, 4.2308, 4.2757, 4.0904, 4.4598, 3.1372, 3.7868, 2.1608],
+       device='cuda:3'), covar=tensor([0.0165, 0.0213, 0.0147, 0.0175, 0.0121, 0.0868, 0.0656, 0.1414],
+       device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0136, 0.0109, 0.0129, 0.0113, 0.0127, 0.0138, 0.0121],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 14:00:09,464 INFO [train.py:892] (3/4) Epoch 21, batch 750, loss[loss=0.2088, simple_loss=0.2789, pruned_loss=0.06932, over 19636.00 frames. ], tot_loss[loss=0.1853, simple_loss=0.2569, pruned_loss=0.05689, over 3860050.05 frames. ], batch size: 299, lr: 7.51e-03, grad_scale: 16.0
+2023-03-28 14:02:04,456 INFO [train.py:892] (3/4) Epoch 21, batch 800, loss[loss=0.1778, simple_loss=0.2552, pruned_loss=0.05023, over 19799.00 frames. ], tot_loss[loss=0.1863, simple_loss=0.2576, pruned_loss=0.0575, over 3880214.48 frames. ], batch size: 68, lr: 7.51e-03, grad_scale: 16.0
+2023-03-28 14:02:34,401 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6567, 4.5191, 5.0811, 4.5439, 4.2143, 4.8206, 4.6727, 5.1844],
+       device='cuda:3'), covar=tensor([0.0783, 0.0362, 0.0316, 0.0379, 0.0751, 0.0404, 0.0461, 0.0290],
+       device='cuda:3'), in_proj_covar=tensor([0.0271, 0.0211, 0.0209, 0.0219, 0.0199, 0.0219, 0.0220, 0.0203],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 14:02:39,798 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.002e+02 4.314e+02 5.039e+02 6.116e+02 1.150e+03, threshold=1.008e+03, percent-clipped=3.0
+2023-03-28 14:02:56,841 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37924.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:03:57,509 INFO [train.py:892] (3/4) Epoch 21, batch 850, loss[loss=0.1927, simple_loss=0.2768, pruned_loss=0.0543, over 19579.00 frames. ], tot_loss[loss=0.1855, simple_loss=0.2574, pruned_loss=0.05679, over 3895811.54 frames. ], batch size: 49, lr: 7.50e-03, grad_scale: 16.0
+2023-03-28 14:04:45,711 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:05:01,911 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37979.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:05:25,315 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.5235, 1.4725, 1.6747, 1.5722, 1.4075, 1.6300, 1.4105, 1.5835],
+       device='cuda:3'), covar=tensor([0.0275, 0.0257, 0.0240, 0.0242, 0.0391, 0.0217, 0.0391, 0.0233],
+       device='cuda:3'), in_proj_covar=tensor([0.0068, 0.0064, 0.0067, 0.0061, 0.0074, 0.0068, 0.0086, 0.0060],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-28 14:05:53,006 INFO [train.py:892] (3/4) Epoch 21, batch 900, loss[loss=0.173, simple_loss=0.2533, pruned_loss=0.04631, over 19739.00 frames. ], tot_loss[loss=0.1865, simple_loss=0.2584, pruned_loss=0.05727, over 3908485.75 frames. ], batch size: 95, lr: 7.50e-03, grad_scale: 16.0
+2023-03-28 14:06:23,636 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.4412, 1.3736, 1.5476, 1.4507, 1.3699, 1.4556, 1.3248, 1.4808],
+       device='cuda:3'), covar=tensor([0.0326, 0.0279, 0.0273, 0.0267, 0.0409, 0.0255, 0.0483, 0.0276],
+       device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0064, 0.0068, 0.0061, 0.0075, 0.0068, 0.0086, 0.0060],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-28 14:06:31,292 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.842e+02 4.121e+02 4.904e+02 6.115e+02 1.131e+03, threshold=9.807e+02, percent-clipped=1.0
+2023-03-28 14:06:38,767 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:07:21,124 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:07:46,388 INFO [train.py:892] (3/4) Epoch 21, batch 950, loss[loss=0.1992, simple_loss=0.2636, pruned_loss=0.06738, over 19755.00 frames. ], tot_loss[loss=0.1892, simple_loss=0.2606, pruned_loss=0.05886, over 3915578.77 frames. ], batch size: 259, lr: 7.49e-03, grad_scale: 16.0
+2023-03-28 14:08:23,053 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=38068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:09:27,763 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4066, 2.6218, 3.6012, 2.9910, 3.0904, 3.0342, 2.1088, 2.2378],
+       device='cuda:3'), covar=tensor([0.1058, 0.2704, 0.0641, 0.0911, 0.1643, 0.1311, 0.2356, 0.2546],
+       device='cuda:3'), in_proj_covar=tensor([0.0338, 0.0368, 0.0322, 0.0262, 0.0362, 0.0341, 0.0345, 0.0316],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 14:09:34,130 INFO [train.py:892] (3/4) Epoch 21, batch 1000, loss[loss=0.1645, simple_loss=0.245, pruned_loss=0.04194, over 19707.00 frames. ], tot_loss[loss=0.1885, simple_loss=0.26, pruned_loss=0.05846, over 3923735.75 frames. ], batch size: 85, lr: 7.49e-03, grad_scale: 16.0
+2023-03-28 14:10:09,403 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.902e+02 4.380e+02 5.195e+02 6.424e+02 1.121e+03, threshold=1.039e+03, percent-clipped=5.0
+2023-03-28 14:11:25,425 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.41 vs. limit=5.0
+2023-03-28 14:11:25,970 INFO [train.py:892] (3/4) Epoch 21, batch 1050, loss[loss=0.2272, simple_loss=0.2856, pruned_loss=0.08441, over 19769.00 frames. ], tot_loss[loss=0.188, simple_loss=0.2597, pruned_loss=0.05816, over 3929396.94 frames. ], batch size: 280, lr: 7.48e-03, grad_scale: 16.0
+2023-03-28 14:13:22,199 INFO [train.py:892] (3/4) Epoch 21, batch 1100, loss[loss=0.1528, simple_loss=0.2227, pruned_loss=0.04146, over 19875.00 frames. ], tot_loss[loss=0.1864, simple_loss=0.2584, pruned_loss=0.05716, over 3934154.75 frames. ], batch size: 77, lr: 7.48e-03, grad_scale: 16.0
+2023-03-28 14:13:26,189 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0
+2023-03-28 14:13:55,404 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.714e+02 3.883e+02 4.794e+02 6.298e+02 1.135e+03, threshold=9.588e+02, percent-clipped=2.0
+2023-03-28 14:15:12,032 INFO [train.py:892] (3/4) Epoch 21, batch 1150, loss[loss=0.1981, simple_loss=0.2659, pruned_loss=0.06516, over 19804.00 frames. ], tot_loss[loss=0.1867, simple_loss=0.2587, pruned_loss=0.05735, over 3935956.53 frames. ], batch size: 195, lr: 7.47e-03, grad_scale: 16.0
+2023-03-28 14:17:02,226 INFO [train.py:892] (3/4) Epoch 21, batch 1200, loss[loss=0.1729, simple_loss=0.2442, pruned_loss=0.0508, over 19767.00 frames. ], tot_loss[loss=0.1858, simple_loss=0.258, pruned_loss=0.05682, over 3940051.04 frames. ], batch size: 113, lr: 7.47e-03, grad_scale: 32.0
+2023-03-28 14:17:37,167 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.715e+02 4.012e+02 4.628e+02 5.373e+02 1.374e+03, threshold=9.255e+02, percent-clipped=1.0
+2023-03-28 14:18:18,196 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38335.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:18:53,058 INFO [train.py:892] (3/4) Epoch 21, batch 1250, loss[loss=0.1745, simple_loss=0.2453, pruned_loss=0.0518, over 19865.00 frames. ], tot_loss[loss=0.1849, simple_loss=0.2572, pruned_loss=0.05627, over 3942395.78 frames. ], batch size: 46, lr: 7.46e-03, grad_scale: 32.0
+2023-03-28 14:19:13,158 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:20:46,462 INFO [train.py:892] (3/4) Epoch 21, batch 1300, loss[loss=0.1903, simple_loss=0.2654, pruned_loss=0.05756, over 19584.00 frames. ], tot_loss[loss=0.1851, simple_loss=0.2574, pruned_loss=0.05638, over 3943972.21 frames. ], batch size: 42, lr: 7.46e-03, grad_scale: 32.0
+2023-03-28 14:21:20,981 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.582e+02 4.064e+02 4.748e+02 5.705e+02 1.088e+03, threshold=9.497e+02, percent-clipped=1.0
+2023-03-28 14:21:32,026 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:22:38,964 INFO [train.py:892] (3/4) Epoch 21, batch 1350, loss[loss=0.1576, simple_loss=0.2294, pruned_loss=0.04294, over 19886.00 frames. ], tot_loss[loss=0.1839, simple_loss=0.2565, pruned_loss=0.05565, over 3945803.32 frames. ], batch size: 87, lr: 7.45e-03, grad_scale: 32.0
+2023-03-28 14:23:53,558 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.6123, 5.8789, 5.9306, 5.7956, 5.6170, 5.8757, 5.2351, 5.3662],
+       device='cuda:3'), covar=tensor([0.0342, 0.0390, 0.0444, 0.0377, 0.0497, 0.0541, 0.0669, 0.0856],
+       device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0257, 0.0278, 0.0241, 0.0242, 0.0231, 0.0250, 0.0291],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 14:24:34,314 INFO [train.py:892] (3/4) Epoch 21, batch 1400, loss[loss=0.1924, simple_loss=0.2609, pruned_loss=0.0619, over 19890.00 frames. ], tot_loss[loss=0.184, simple_loss=0.2563, pruned_loss=0.05579, over 3944608.42 frames. ], batch size: 63, lr: 7.45e-03, grad_scale: 32.0
+2023-03-28 14:25:09,934 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.194e+02 4.264e+02 4.951e+02 5.984e+02 1.387e+03, threshold=9.901e+02, percent-clipped=2.0
+2023-03-28 14:26:05,378 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0
+2023-03-28 14:26:26,395 INFO [train.py:892] (3/4) Epoch 21, batch 1450, loss[loss=0.1597, simple_loss=0.2371, pruned_loss=0.04115, over 19748.00 frames. ], tot_loss[loss=0.1836, simple_loss=0.2561, pruned_loss=0.05553, over 3944454.56 frames. ], batch size: 110, lr: 7.45e-03, grad_scale: 32.0
+2023-03-28 14:28:22,033 INFO [train.py:892] (3/4) Epoch 21, batch 1500, loss[loss=0.1737, simple_loss=0.2441, pruned_loss=0.05162, over 19840.00 frames. ], tot_loss[loss=0.1837, simple_loss=0.2556, pruned_loss=0.05593, over 3946469.17 frames. ], batch size: 43, lr: 7.44e-03, grad_scale: 32.0
+2023-03-28 14:28:58,331 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.556e+02 4.180e+02 4.793e+02 6.063e+02 1.172e+03, threshold=9.587e+02, percent-clipped=1.0
+2023-03-28 14:29:12,583 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6155, 2.9130, 2.4025, 1.9712, 2.5616, 2.7552, 2.7747, 2.8197],
+       device='cuda:3'), covar=tensor([0.0349, 0.0320, 0.0357, 0.0679, 0.0398, 0.0307, 0.0271, 0.0239],
+       device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0081, 0.0088, 0.0093, 0.0095, 0.0072, 0.0070, 0.0072],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 14:29:40,697 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38635.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:29:44,906 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1202, 3.1923, 4.5570, 3.4147, 3.8344, 3.6235, 2.5881, 2.6804],
+       device='cuda:3'), covar=tensor([0.0807, 0.2795, 0.0465, 0.0918, 0.1502, 0.1306, 0.2248, 0.2527],
+       device='cuda:3'), in_proj_covar=tensor([0.0336, 0.0368, 0.0322, 0.0260, 0.0361, 0.0337, 0.0344, 0.0313],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 14:30:15,083 INFO [train.py:892] (3/4) Epoch 21, batch 1550, loss[loss=0.172, simple_loss=0.2506, pruned_loss=0.04673, over 19804.00 frames. ], tot_loss[loss=0.1832, simple_loss=0.2556, pruned_loss=0.05539, over 3947566.22 frames. ], batch size: 82, lr: 7.44e-03, grad_scale: 32.0
+2023-03-28 14:31:00,290 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9124, 2.9342, 1.7096, 3.4563, 3.2040, 3.5136, 3.5277, 2.7391],
+       device='cuda:3'), covar=tensor([0.0650, 0.0592, 0.1701, 0.0652, 0.0526, 0.0408, 0.0477, 0.0807],
+       device='cuda:3'), in_proj_covar=tensor([0.0135, 0.0133, 0.0138, 0.0138, 0.0121, 0.0123, 0.0134, 0.0138],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 14:31:30,389 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=38683.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:32:09,919 INFO [train.py:892] (3/4) Epoch 21, batch 1600, loss[loss=0.1653, simple_loss=0.2484, pruned_loss=0.04109, over 19800.00 frames. ], tot_loss[loss=0.1839, simple_loss=0.2564, pruned_loss=0.05572, over 3947152.64 frames. ], batch size: 79, lr: 7.43e-03, grad_scale: 32.0
+2023-03-28 14:32:47,592 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38716.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:32:48,617 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.983e+02 3.919e+02 4.752e+02 6.022e+02 1.186e+03, threshold=9.505e+02, percent-clipped=3.0
+2023-03-28 14:34:02,101 INFO [train.py:892] (3/4) Epoch 21, batch 1650, loss[loss=0.1806, simple_loss=0.2543, pruned_loss=0.0535, over 19713.00 frames. ], tot_loss[loss=0.185, simple_loss=0.2577, pruned_loss=0.05611, over 3945506.62 frames. ], batch size: 109, lr: 7.43e-03, grad_scale: 32.0
+2023-03-28 14:35:57,103 INFO [train.py:892] (3/4) Epoch 21, batch 1700, loss[loss=0.1999, simple_loss=0.2654, pruned_loss=0.06719, over 19822.00 frames. ], tot_loss[loss=0.1858, simple_loss=0.2583, pruned_loss=0.05669, over 3947341.35 frames. ], batch size: 187, lr: 7.42e-03, grad_scale: 32.0
+2023-03-28 14:36:16,195 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.2506, 5.5301, 5.5592, 5.4465, 5.1987, 5.5469, 4.8899, 5.0229],
+       device='cuda:3'), covar=tensor([0.0397, 0.0439, 0.0498, 0.0420, 0.0571, 0.0509, 0.0769, 0.0959],
+       device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0252, 0.0273, 0.0236, 0.0236, 0.0227, 0.0244, 0.0286],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 14:36:32,828 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.773e+02 3.919e+02 4.516e+02 5.441e+02 1.197e+03, threshold=9.032e+02, percent-clipped=2.0
+2023-03-28 14:36:48,304 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:37:23,794 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.2077, 5.5426, 5.7549, 5.5502, 5.4308, 5.1050, 5.4202, 5.2426],
+       device='cuda:3'), covar=tensor([0.1414, 0.1380, 0.0866, 0.1154, 0.0680, 0.0843, 0.1884, 0.1899],
+       device='cuda:3'), in_proj_covar=tensor([0.0275, 0.0296, 0.0341, 0.0272, 0.0253, 0.0257, 0.0334, 0.0360],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 14:37:43,275 INFO [train.py:892] (3/4) Epoch 21, batch 1750, loss[loss=0.1663, simple_loss=0.2424, pruned_loss=0.04513, over 19837.00 frames. ], tot_loss[loss=0.1849, simple_loss=0.2578, pruned_loss=0.05598, over 3946855.82 frames. ], batch size: 101, lr: 7.42e-03, grad_scale: 32.0
+2023-03-28 14:38:01,118 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.7653, 1.8084, 1.9138, 1.7606, 1.7154, 1.8605, 1.7535, 1.8559],
+       device='cuda:3'), covar=tensor([0.0259, 0.0245, 0.0237, 0.0257, 0.0378, 0.0247, 0.0371, 0.0231],
+       device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0065, 0.0068, 0.0062, 0.0076, 0.0070, 0.0086, 0.0061],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-28 14:38:47,597 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:39:18,405 INFO [train.py:892] (3/4) Epoch 21, batch 1800, loss[loss=0.1673, simple_loss=0.254, pruned_loss=0.0403, over 19603.00 frames. ], tot_loss[loss=0.1844, simple_loss=0.2569, pruned_loss=0.05593, over 3948899.46 frames. ], batch size: 48, lr: 7.41e-03, grad_scale: 32.0
+2023-03-28 14:39:48,600 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.425e+02 4.042e+02 5.128e+02 6.021e+02 1.055e+03, threshold=1.026e+03, percent-clipped=2.0
+2023-03-28 14:40:34,108 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8781, 3.1596, 2.6543, 2.2793, 2.8387, 3.1004, 3.1754, 3.0189],
+       device='cuda:3'), covar=tensor([0.0249, 0.0265, 0.0294, 0.0555, 0.0317, 0.0284, 0.0166, 0.0259],
+       device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0081, 0.0086, 0.0091, 0.0093, 0.0070, 0.0069, 0.0071],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 14:40:47,459 INFO [train.py:892] (3/4) Epoch 21, batch 1850, loss[loss=0.1706, simple_loss=0.2518, pruned_loss=0.04474, over 19572.00 frames. ], tot_loss[loss=0.1839, simple_loss=0.2574, pruned_loss=0.05525, over 3948780.26 frames. ], batch size: 53, lr: 7.41e-03, grad_scale: 32.0
+2023-03-28 14:41:51,760 INFO [train.py:892] (3/4) Epoch 22, batch 0, loss[loss=0.1536, simple_loss=0.2242, pruned_loss=0.04149, over 19802.00 frames. ], tot_loss[loss=0.1536, simple_loss=0.2242, pruned_loss=0.04149, over 19802.00 frames. ], batch size: 107, lr: 7.23e-03, grad_scale: 32.0
+2023-03-28 14:41:51,761 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-28 14:42:30,100 INFO [train.py:926] (3/4) Epoch 22, validation: loss=0.1727, simple_loss=0.2482, pruned_loss=0.04859, over 2883724.00 frames. 
+2023-03-28 14:42:30,102 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-28 14:42:36,146 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9828, 3.3283, 2.7472, 2.3537, 2.7964, 3.1698, 3.1926, 3.2200],
+       device='cuda:3'), covar=tensor([0.0238, 0.0234, 0.0264, 0.0480, 0.0346, 0.0294, 0.0205, 0.0189],
+       device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0081, 0.0087, 0.0091, 0.0094, 0.0071, 0.0069, 0.0071],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 14:44:26,229 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9439, 3.7918, 3.7371, 3.4583, 3.9272, 2.8091, 3.2149, 1.8968],
+       device='cuda:3'), covar=tensor([0.0219, 0.0242, 0.0176, 0.0238, 0.0149, 0.1036, 0.0763, 0.1642],
+       device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0138, 0.0110, 0.0130, 0.0115, 0.0129, 0.0140, 0.0123],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 14:44:29,533 INFO [train.py:892] (3/4) Epoch 22, batch 50, loss[loss=0.1705, simple_loss=0.2444, pruned_loss=0.04829, over 19628.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2466, pruned_loss=0.04985, over 891820.14 frames. ], batch size: 68, lr: 7.23e-03, grad_scale: 32.0
+2023-03-28 14:44:52,600 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:44:53,817 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.550e+02 3.535e+02 4.354e+02 5.558e+02 1.145e+03, threshold=8.708e+02, percent-clipped=3.0
+2023-03-28 14:46:23,800 INFO [train.py:892] (3/4) Epoch 22, batch 100, loss[loss=0.1619, simple_loss=0.2339, pruned_loss=0.045, over 19710.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.251, pruned_loss=0.05155, over 1571150.14 frames. ], batch size: 101, lr: 7.22e-03, grad_scale: 32.0
+2023-03-28 14:46:43,236 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:48:04,922 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6195, 2.6865, 2.6489, 2.1746, 2.8283, 2.2747, 2.7011, 2.8579],
+       device='cuda:3'), covar=tensor([0.0452, 0.0399, 0.0572, 0.0799, 0.0324, 0.0492, 0.0454, 0.0249],
+       device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0077, 0.0075, 0.0105, 0.0072, 0.0072, 0.0070, 0.0062],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 14:48:19,884 INFO [train.py:892] (3/4) Epoch 22, batch 150, loss[loss=0.1724, simple_loss=0.2383, pruned_loss=0.0532, over 19875.00 frames. ], tot_loss[loss=0.1797, simple_loss=0.2533, pruned_loss=0.05306, over 2098436.94 frames. ], batch size: 159, lr: 7.22e-03, grad_scale: 32.0
+2023-03-28 14:48:32,996 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1151, 2.2456, 2.5014, 2.2309, 2.6488, 2.6999, 3.0183, 3.2368],
+       device='cuda:3'), covar=tensor([0.0740, 0.1722, 0.1568, 0.2130, 0.1390, 0.1333, 0.0675, 0.0571],
+       device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0235, 0.0256, 0.0247, 0.0285, 0.0248, 0.0215, 0.0237],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 14:48:48,469 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.956e+02 3.952e+02 4.652e+02 5.506e+02 8.622e+02, threshold=9.304e+02, percent-clipped=0.0
+2023-03-28 14:50:14,953 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0603, 2.9652, 3.3445, 2.4364, 3.3960, 2.7462, 3.0600, 3.3517],
+       device='cuda:3'), covar=tensor([0.0525, 0.0449, 0.0398, 0.0813, 0.0363, 0.0440, 0.0472, 0.0270],
+       device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0077, 0.0075, 0.0105, 0.0071, 0.0071, 0.0069, 0.0062],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 14:50:18,183 INFO [train.py:892] (3/4) Epoch 22, batch 200, loss[loss=0.1798, simple_loss=0.255, pruned_loss=0.05229, over 19851.00 frames. ], tot_loss[loss=0.1831, simple_loss=0.2564, pruned_loss=0.05492, over 2508363.43 frames. ], batch size: 78, lr: 7.22e-03, grad_scale: 32.0
+2023-03-28 14:51:07,147 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39179.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:51:23,402 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39185.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:52:10,106 INFO [train.py:892] (3/4) Epoch 22, batch 250, loss[loss=0.1681, simple_loss=0.2358, pruned_loss=0.05022, over 19847.00 frames. ], tot_loss[loss=0.1824, simple_loss=0.2554, pruned_loss=0.0547, over 2828599.72 frames. ], batch size: 137, lr: 7.21e-03, grad_scale: 32.0
+2023-03-28 14:52:16,032 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0544, 3.8836, 3.8760, 3.6441, 4.0299, 2.8351, 3.3193, 1.9231],
+       device='cuda:3'), covar=tensor([0.0203, 0.0227, 0.0154, 0.0211, 0.0140, 0.1022, 0.0777, 0.1555],
+       device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0137, 0.0109, 0.0129, 0.0114, 0.0128, 0.0139, 0.0121],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 14:52:34,876 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.591e+02 3.931e+02 4.730e+02 5.720e+02 9.531e+02, threshold=9.460e+02, percent-clipped=1.0
+2023-03-28 14:53:42,685 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:54:03,345 INFO [train.py:892] (3/4) Epoch 22, batch 300, loss[loss=0.1902, simple_loss=0.2517, pruned_loss=0.06432, over 19854.00 frames. ], tot_loss[loss=0.1811, simple_loss=0.2543, pruned_loss=0.05398, over 3074914.88 frames. ], batch size: 165, lr: 7.21e-03, grad_scale: 32.0
+2023-03-28 14:55:08,568 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:55:59,302 INFO [train.py:892] (3/4) Epoch 22, batch 350, loss[loss=0.1697, simple_loss=0.2329, pruned_loss=0.05329, over 19803.00 frames. ], tot_loss[loss=0.1806, simple_loss=0.2538, pruned_loss=0.0537, over 3269666.62 frames. ], batch size: 132, lr: 7.20e-03, grad_scale: 32.0
+2023-03-28 14:56:21,664 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.483e+02 3.997e+02 4.728e+02 5.910e+02 1.079e+03, threshold=9.457e+02, percent-clipped=2.0
+2023-03-28 14:56:47,609 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3985, 4.0422, 4.1458, 4.3844, 4.0726, 4.4536, 4.4695, 4.6148],
+       device='cuda:3'), covar=tensor([0.0642, 0.0410, 0.0444, 0.0344, 0.0744, 0.0412, 0.0416, 0.0327],
+       device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0165, 0.0190, 0.0164, 0.0163, 0.0148, 0.0144, 0.0188],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 14:57:26,616 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.86 vs. limit=5.0
+2023-03-28 14:57:27,968 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 14:57:49,325 INFO [train.py:892] (3/4) Epoch 22, batch 400, loss[loss=0.2055, simple_loss=0.2793, pruned_loss=0.06581, over 19707.00 frames. ], tot_loss[loss=0.1804, simple_loss=0.254, pruned_loss=0.0534, over 3419405.08 frames. ], batch size: 295, lr: 7.20e-03, grad_scale: 32.0
+2023-03-28 14:58:58,020 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1296, 3.3759, 2.8855, 2.5368, 2.8815, 3.4075, 3.1958, 3.2323],
+       device='cuda:3'), covar=tensor([0.0254, 0.0253, 0.0274, 0.0481, 0.0350, 0.0211, 0.0205, 0.0189],
+       device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0082, 0.0088, 0.0093, 0.0095, 0.0071, 0.0071, 0.0073],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 14:59:41,760 INFO [train.py:892] (3/4) Epoch 22, batch 450, loss[loss=0.1735, simple_loss=0.2454, pruned_loss=0.05082, over 19843.00 frames. ], tot_loss[loss=0.1813, simple_loss=0.2552, pruned_loss=0.05372, over 3534742.39 frames. ], batch size: 115, lr: 7.19e-03, grad_scale: 32.0
+2023-03-28 15:00:08,521 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.964e+02 3.895e+02 4.549e+02 5.283e+02 8.975e+02, threshold=9.097e+02, percent-clipped=0.0
+2023-03-28 15:01:37,219 INFO [train.py:892] (3/4) Epoch 22, batch 500, loss[loss=0.1531, simple_loss=0.2228, pruned_loss=0.04175, over 19861.00 frames. ], tot_loss[loss=0.1822, simple_loss=0.2555, pruned_loss=0.05446, over 3626712.84 frames. ], batch size: 106, lr: 7.19e-03, grad_scale: 32.0
+2023-03-28 15:01:53,550 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:02:31,950 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:03:33,881 INFO [train.py:892] (3/4) Epoch 22, batch 550, loss[loss=0.1778, simple_loss=0.2498, pruned_loss=0.05292, over 19841.00 frames. ], tot_loss[loss=0.182, simple_loss=0.2551, pruned_loss=0.05444, over 3699589.66 frames. ], batch size: 60, lr: 7.18e-03, grad_scale: 32.0
+2023-03-28 15:03:58,569 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.760e+02 4.104e+02 4.987e+02 6.113e+02 1.669e+03, threshold=9.973e+02, percent-clipped=4.0
+2023-03-28 15:04:16,120 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:04:22,114 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:04:51,801 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39541.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:04:56,482 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0750, 2.2460, 2.8989, 3.1223, 3.6578, 4.1772, 3.9565, 4.0713],
+       device='cuda:3'), covar=tensor([0.0780, 0.1812, 0.1255, 0.0615, 0.0316, 0.0201, 0.0323, 0.0404],
+       device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0168, 0.0171, 0.0141, 0.0124, 0.0120, 0.0114, 0.0108],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 15:05:16,539 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0
+2023-03-28 15:05:26,080 INFO [train.py:892] (3/4) Epoch 22, batch 600, loss[loss=0.1607, simple_loss=0.2321, pruned_loss=0.04462, over 19884.00 frames. ], tot_loss[loss=0.1815, simple_loss=0.2548, pruned_loss=0.05412, over 3753881.70 frames. ], batch size: 77, lr: 7.18e-03, grad_scale: 32.0
+2023-03-28 15:06:15,340 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5592, 3.4503, 3.4538, 3.7417, 3.5646, 3.8891, 3.6831, 3.7279],
+       device='cuda:3'), covar=tensor([0.1044, 0.0670, 0.0779, 0.0539, 0.0894, 0.0601, 0.0666, 0.0792],
+       device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0165, 0.0190, 0.0163, 0.0163, 0.0147, 0.0143, 0.0187],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 15:07:20,938 INFO [train.py:892] (3/4) Epoch 22, batch 650, loss[loss=0.1641, simple_loss=0.2386, pruned_loss=0.04482, over 19793.00 frames. ], tot_loss[loss=0.1807, simple_loss=0.2537, pruned_loss=0.05384, over 3797368.35 frames. ], batch size: 120, lr: 7.17e-03, grad_scale: 32.0
+2023-03-28 15:07:45,277 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.674e+02 3.926e+02 4.968e+02 5.717e+02 7.985e+02, threshold=9.935e+02, percent-clipped=0.0
+2023-03-28 15:08:37,626 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39639.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 15:08:39,447 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:09:12,689 INFO [train.py:892] (3/4) Epoch 22, batch 700, loss[loss=0.1588, simple_loss=0.2285, pruned_loss=0.04449, over 19843.00 frames. ], tot_loss[loss=0.1809, simple_loss=0.2538, pruned_loss=0.05396, over 3831181.29 frames. ], batch size: 43, lr: 7.17e-03, grad_scale: 32.0
+2023-03-28 15:10:10,908 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8405, 2.8629, 4.2174, 3.3162, 3.6123, 3.3117, 2.3616, 2.3680],
+       device='cuda:3'), covar=tensor([0.0887, 0.2832, 0.0538, 0.0856, 0.1497, 0.1252, 0.2182, 0.2785],
+       device='cuda:3'), in_proj_covar=tensor([0.0338, 0.0369, 0.0327, 0.0263, 0.0364, 0.0344, 0.0348, 0.0316],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 15:10:53,866 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39700.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 15:11:05,435 INFO [train.py:892] (3/4) Epoch 22, batch 750, loss[loss=0.2629, simple_loss=0.3262, pruned_loss=0.09978, over 19470.00 frames. ], tot_loss[loss=0.1814, simple_loss=0.2543, pruned_loss=0.05426, over 3857789.99 frames. ], batch size: 396, lr: 7.17e-03, grad_scale: 16.0
+2023-03-28 15:11:32,104 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.919e+02 4.119e+02 5.270e+02 6.689e+02 1.138e+03, threshold=1.054e+03, percent-clipped=1.0
+2023-03-28 15:13:01,456 INFO [train.py:892] (3/4) Epoch 22, batch 800, loss[loss=0.1593, simple_loss=0.2417, pruned_loss=0.03842, over 19661.00 frames. ], tot_loss[loss=0.1819, simple_loss=0.255, pruned_loss=0.05446, over 3877347.06 frames. ], batch size: 67, lr: 7.16e-03, grad_scale: 16.0
+2023-03-28 15:13:24,947 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39767.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:13:39,443 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1553, 3.0961, 4.6317, 3.3983, 3.8438, 3.5600, 2.4126, 2.5747],
+       device='cuda:3'), covar=tensor([0.0850, 0.3054, 0.0438, 0.0983, 0.1591, 0.1351, 0.2521, 0.2792],
+       device='cuda:3'), in_proj_covar=tensor([0.0340, 0.0372, 0.0328, 0.0264, 0.0365, 0.0345, 0.0350, 0.0318],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 15:14:37,275 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9911, 3.3706, 3.4053, 3.9970, 2.8049, 3.1343, 2.3813, 2.4950],
+       device='cuda:3'), covar=tensor([0.0517, 0.1995, 0.0987, 0.0383, 0.1918, 0.0900, 0.1467, 0.1778],
+       device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0333, 0.0243, 0.0186, 0.0242, 0.0200, 0.0212, 0.0212],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 15:14:51,999 INFO [train.py:892] (3/4) Epoch 22, batch 850, loss[loss=0.1775, simple_loss=0.2584, pruned_loss=0.04835, over 19881.00 frames. ], tot_loss[loss=0.1817, simple_loss=0.255, pruned_loss=0.05416, over 3893493.25 frames. ], batch size: 84, lr: 7.16e-03, grad_scale: 16.0
+2023-03-28 15:15:12,699 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39815.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:15:18,342 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.581e+02 3.870e+02 4.647e+02 5.944e+02 1.250e+03, threshold=9.293e+02, percent-clipped=1.0
+2023-03-28 15:15:21,360 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:15:34,140 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0
+2023-03-28 15:15:42,478 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39828.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:16:09,968 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39841.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:16:42,172 INFO [train.py:892] (3/4) Epoch 22, batch 900, loss[loss=0.1962, simple_loss=0.2598, pruned_loss=0.06626, over 19773.00 frames. ], tot_loss[loss=0.1824, simple_loss=0.2557, pruned_loss=0.05462, over 3906191.44 frames. ], batch size: 163, lr: 7.15e-03, grad_scale: 16.0
+2023-03-28 15:16:57,105 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0833, 4.6727, 4.7346, 5.0109, 4.7132, 5.2992, 5.2063, 5.3519],
+       device='cuda:3'), covar=tensor([0.0633, 0.0372, 0.0506, 0.0338, 0.0651, 0.0362, 0.0408, 0.0326],
+       device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0164, 0.0189, 0.0162, 0.0162, 0.0146, 0.0141, 0.0186],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 15:17:25,263 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-28 15:17:30,883 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39876.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:17:39,170 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0
+2023-03-28 15:17:55,225 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7356, 2.7440, 2.8798, 2.3514, 3.0208, 2.5459, 2.9232, 2.9822],
+       device='cuda:3'), covar=tensor([0.0513, 0.0437, 0.0522, 0.0760, 0.0317, 0.0419, 0.0400, 0.0284],
+       device='cuda:3'), in_proj_covar=tensor([0.0070, 0.0079, 0.0076, 0.0105, 0.0072, 0.0072, 0.0070, 0.0064],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 15:18:01,292 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:18:37,556 INFO [train.py:892] (3/4) Epoch 22, batch 950, loss[loss=0.2131, simple_loss=0.2867, pruned_loss=0.06974, over 19740.00 frames. ], tot_loss[loss=0.183, simple_loss=0.2562, pruned_loss=0.05491, over 3916299.81 frames. ], batch size: 291, lr: 7.15e-03, grad_scale: 16.0
+2023-03-28 15:19:06,559 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.773e+02 4.111e+02 4.981e+02 6.379e+02 1.110e+03, threshold=9.962e+02, percent-clipped=1.0
+2023-03-28 15:19:24,529 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.0469, 1.9949, 2.1373, 2.0135, 2.1138, 2.1618, 2.0089, 2.1273],
+       device='cuda:3'), covar=tensor([0.0319, 0.0312, 0.0292, 0.0283, 0.0386, 0.0288, 0.0434, 0.0346],
+       device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0065, 0.0070, 0.0063, 0.0075, 0.0070, 0.0088, 0.0062],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-28 15:19:56,024 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:20:17,260 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4618, 5.7197, 5.7433, 5.6278, 5.3720, 5.7558, 5.0675, 5.2112],
+       device='cuda:3'), covar=tensor([0.0410, 0.0421, 0.0496, 0.0376, 0.0610, 0.0481, 0.0698, 0.0846],
+       device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0254, 0.0273, 0.0240, 0.0238, 0.0228, 0.0246, 0.0289],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 15:20:28,356 INFO [train.py:892] (3/4) Epoch 22, batch 1000, loss[loss=0.1752, simple_loss=0.244, pruned_loss=0.05315, over 19736.00 frames. ], tot_loss[loss=0.1827, simple_loss=0.2557, pruned_loss=0.05483, over 3925093.78 frames. ], batch size: 106, lr: 7.14e-03, grad_scale: 16.0
+2023-03-28 15:21:42,654 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:21:59,367 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39995.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 15:22:27,065 INFO [train.py:892] (3/4) Epoch 22, batch 1050, loss[loss=0.1598, simple_loss=0.2306, pruned_loss=0.04448, over 19875.00 frames. ], tot_loss[loss=0.1823, simple_loss=0.255, pruned_loss=0.05479, over 3932415.09 frames. ], batch size: 125, lr: 7.14e-03, grad_scale: 16.0
+2023-03-28 15:22:53,662 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.755e+02 4.195e+02 4.862e+02 6.171e+02 1.256e+03, threshold=9.724e+02, percent-clipped=2.0
+2023-03-28 15:23:21,917 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9396, 2.8516, 3.0641, 2.3967, 3.1610, 2.6289, 3.0034, 3.1824],
+       device='cuda:3'), covar=tensor([0.0481, 0.0434, 0.0763, 0.0796, 0.0357, 0.0486, 0.0384, 0.0304],
+       device='cuda:3'), in_proj_covar=tensor([0.0069, 0.0078, 0.0075, 0.0104, 0.0071, 0.0072, 0.0070, 0.0063],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 15:23:22,481 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.66 vs. limit=5.0
+2023-03-28 15:23:23,734 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3717, 4.0717, 4.1851, 4.4216, 4.0440, 4.3784, 4.4967, 4.7003],
+       device='cuda:3'), covar=tensor([0.0708, 0.0444, 0.0553, 0.0362, 0.0767, 0.0572, 0.0446, 0.0326],
+       device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0165, 0.0190, 0.0164, 0.0162, 0.0147, 0.0143, 0.0187],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 15:24:16,341 INFO [train.py:892] (3/4) Epoch 22, batch 1100, loss[loss=0.181, simple_loss=0.2546, pruned_loss=0.05369, over 19643.00 frames. ], tot_loss[loss=0.1821, simple_loss=0.2547, pruned_loss=0.05473, over 3935598.86 frames. ], batch size: 69, lr: 7.13e-03, grad_scale: 16.0
+2023-03-28 15:24:38,207 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7054, 2.8536, 4.1792, 3.2486, 3.6107, 3.3094, 2.3169, 2.4549],
+       device='cuda:3'), covar=tensor([0.1060, 0.2821, 0.0513, 0.0927, 0.1471, 0.1395, 0.2361, 0.2726],
+       device='cuda:3'), in_proj_covar=tensor([0.0340, 0.0371, 0.0327, 0.0264, 0.0365, 0.0347, 0.0348, 0.0318],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 15:26:09,785 INFO [train.py:892] (3/4) Epoch 22, batch 1150, loss[loss=0.1811, simple_loss=0.2442, pruned_loss=0.05902, over 19684.00 frames. ], tot_loss[loss=0.1827, simple_loss=0.2554, pruned_loss=0.05501, over 3937706.52 frames. ], batch size: 64, lr: 7.13e-03, grad_scale: 16.0
+2023-03-28 15:26:33,615 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.627e+02 4.168e+02 4.947e+02 6.030e+02 1.277e+03, threshold=9.894e+02, percent-clipped=1.0
+2023-03-28 15:26:37,684 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:26:46,721 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:26:52,903 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.3285, 1.8657, 1.9784, 2.6481, 2.9138, 3.0268, 2.9632, 3.0653],
+       device='cuda:3'), covar=tensor([0.1131, 0.1786, 0.1635, 0.0674, 0.0543, 0.0388, 0.0426, 0.0395],
+       device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0169, 0.0171, 0.0143, 0.0124, 0.0120, 0.0114, 0.0109],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 15:28:02,033 INFO [train.py:892] (3/4) Epoch 22, batch 1200, loss[loss=0.2084, simple_loss=0.2834, pruned_loss=0.06667, over 19764.00 frames. ], tot_loss[loss=0.1843, simple_loss=0.2566, pruned_loss=0.05601, over 3940357.49 frames. ], batch size: 244, lr: 7.13e-03, grad_scale: 16.0
+2023-03-28 15:28:26,263 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:28:35,095 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:29:32,203 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:29:54,914 INFO [train.py:892] (3/4) Epoch 22, batch 1250, loss[loss=0.1685, simple_loss=0.2531, pruned_loss=0.04196, over 19715.00 frames. ], tot_loss[loss=0.1837, simple_loss=0.2564, pruned_loss=0.05556, over 3941764.98 frames. ], batch size: 85, lr: 7.12e-03, grad_scale: 16.0
+2023-03-28 15:30:21,936 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.511e+02 3.858e+02 4.574e+02 5.518e+02 1.243e+03, threshold=9.148e+02, percent-clipped=1.0
+2023-03-28 15:30:23,093 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:31:39,888 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7247, 2.2291, 2.5290, 3.0027, 3.3390, 3.5920, 3.4793, 3.5357],
+       device='cuda:3'), covar=tensor([0.0943, 0.1701, 0.1331, 0.0617, 0.0424, 0.0257, 0.0380, 0.0358],
+       device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0170, 0.0172, 0.0143, 0.0125, 0.0121, 0.0115, 0.0109],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 15:31:44,904 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0
+2023-03-28 15:31:49,164 INFO [train.py:892] (3/4) Epoch 22, batch 1300, loss[loss=0.1811, simple_loss=0.2595, pruned_loss=0.05134, over 19773.00 frames. ], tot_loss[loss=0.1843, simple_loss=0.2568, pruned_loss=0.0559, over 3943689.57 frames. ], batch size: 66, lr: 7.12e-03, grad_scale: 16.0
+2023-03-28 15:31:50,214 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40256.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:32:19,605 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6166, 3.7869, 3.9567, 4.7651, 3.2218, 3.3815, 3.0006, 2.8261],
+       device='cuda:3'), covar=tensor([0.0480, 0.2453, 0.0968, 0.0350, 0.2189, 0.1175, 0.1330, 0.1838],
+       device='cuda:3'), in_proj_covar=tensor([0.0228, 0.0329, 0.0241, 0.0184, 0.0241, 0.0198, 0.0209, 0.0211],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 15:32:25,974 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:32:41,636 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40279.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:33:20,621 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40295.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 15:33:43,551 INFO [train.py:892] (3/4) Epoch 22, batch 1350, loss[loss=0.2981, simple_loss=0.3665, pruned_loss=0.1149, over 19265.00 frames. ], tot_loss[loss=0.1849, simple_loss=0.2577, pruned_loss=0.05601, over 3943101.03 frames. ], batch size: 483, lr: 7.11e-03, grad_scale: 16.0
+2023-03-28 15:34:11,584 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.667e+02 4.340e+02 4.978e+02 5.730e+02 1.244e+03, threshold=9.955e+02, percent-clipped=1.0
+2023-03-28 15:34:39,858 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40330.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:34:46,785 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:34:55,999 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5492, 3.0411, 3.4257, 3.0854, 3.7739, 3.7238, 4.3383, 4.7691],
+       device='cuda:3'), covar=tensor([0.0496, 0.1665, 0.1504, 0.2087, 0.1712, 0.1273, 0.0585, 0.0508],
+       device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0231, 0.0255, 0.0246, 0.0284, 0.0246, 0.0214, 0.0238],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 15:35:09,378 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40343.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 15:35:13,390 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6835, 4.4404, 4.4753, 4.2109, 4.6949, 3.1771, 3.9229, 2.2869],
+       device='cuda:3'), covar=tensor([0.0187, 0.0195, 0.0136, 0.0205, 0.0123, 0.0928, 0.0739, 0.1409],
+       device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0137, 0.0109, 0.0130, 0.0113, 0.0130, 0.0139, 0.0122],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 15:35:15,553 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2508, 4.3683, 2.5396, 4.6595, 4.8268, 1.9797, 3.9782, 3.4652],
+       device='cuda:3'), covar=tensor([0.0637, 0.0766, 0.2638, 0.0603, 0.0413, 0.2787, 0.0930, 0.0812],
+       device='cuda:3'), in_proj_covar=tensor([0.0219, 0.0244, 0.0223, 0.0254, 0.0231, 0.0197, 0.0231, 0.0185],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 15:35:29,652 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-28 15:35:35,541 INFO [train.py:892] (3/4) Epoch 22, batch 1400, loss[loss=0.1694, simple_loss=0.2493, pruned_loss=0.04475, over 19933.00 frames. ], tot_loss[loss=0.1853, simple_loss=0.2584, pruned_loss=0.05608, over 3944872.84 frames. ], batch size: 51, lr: 7.11e-03, grad_scale: 16.0
+2023-03-28 15:35:42,386 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.1547, 2.4808, 2.1202, 1.6149, 2.2447, 2.4106, 2.3405, 2.4877],
+       device='cuda:3'), covar=tensor([0.0344, 0.0284, 0.0337, 0.0680, 0.0415, 0.0289, 0.0278, 0.0241],
+       device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0084, 0.0091, 0.0095, 0.0097, 0.0074, 0.0073, 0.0074],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 15:35:53,512 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.8365, 6.1628, 6.2197, 6.0224, 5.9467, 6.1504, 5.4432, 5.5079],
+       device='cuda:3'), covar=tensor([0.0369, 0.0428, 0.0468, 0.0383, 0.0532, 0.0526, 0.0765, 0.0905],
+       device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0257, 0.0276, 0.0240, 0.0240, 0.0230, 0.0248, 0.0291],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 15:36:58,958 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:37:29,878 INFO [train.py:892] (3/4) Epoch 22, batch 1450, loss[loss=0.1798, simple_loss=0.2514, pruned_loss=0.05411, over 19733.00 frames. ], tot_loss[loss=0.1836, simple_loss=0.2569, pruned_loss=0.05517, over 3946217.82 frames. ], batch size: 106, lr: 7.10e-03, grad_scale: 16.0
+2023-03-28 15:37:57,983 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.955e+02 4.197e+02 4.850e+02 6.324e+02 1.307e+03, threshold=9.700e+02, percent-clipped=2.0
+2023-03-28 15:38:11,760 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:39:20,787 INFO [train.py:892] (3/4) Epoch 22, batch 1500, loss[loss=0.1654, simple_loss=0.2256, pruned_loss=0.05259, over 19839.00 frames. ], tot_loss[loss=0.1829, simple_loss=0.2562, pruned_loss=0.05475, over 3946240.08 frames. ], batch size: 143, lr: 7.10e-03, grad_scale: 16.0
+2023-03-28 15:39:34,816 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0
+2023-03-28 15:39:52,097 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0
+2023-03-28 15:39:55,916 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:39:56,028 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:40:16,745 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.82 vs. limit=5.0
+2023-03-28 15:40:41,529 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:41:05,768 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.15 vs. limit=5.0
+2023-03-28 15:41:12,875 INFO [train.py:892] (3/4) Epoch 22, batch 1550, loss[loss=0.1857, simple_loss=0.2559, pruned_loss=0.05773, over 19795.00 frames. ], tot_loss[loss=0.184, simple_loss=0.2568, pruned_loss=0.05553, over 3947310.12 frames. ], batch size: 185, lr: 7.10e-03, grad_scale: 16.0
+2023-03-28 15:41:39,827 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.411e+02 3.822e+02 4.612e+02 5.584e+02 1.112e+03, threshold=9.223e+02, percent-clipped=2.0
+2023-03-28 15:41:43,099 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40519.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:42:55,117 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40551.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:42:59,750 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40553.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:43:05,446 INFO [train.py:892] (3/4) Epoch 22, batch 1600, loss[loss=0.175, simple_loss=0.2478, pruned_loss=0.05113, over 19778.00 frames. ], tot_loss[loss=0.1835, simple_loss=0.2563, pruned_loss=0.05536, over 3947527.63 frames. ], batch size: 91, lr: 7.09e-03, grad_scale: 16.0
+2023-03-28 15:43:42,579 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0592, 3.1832, 2.0043, 3.2387, 3.3359, 1.6017, 2.6907, 2.5251],
+       device='cuda:3'), covar=tensor([0.0840, 0.0858, 0.2667, 0.0791, 0.0593, 0.2519, 0.1188, 0.0947],
+       device='cuda:3'), in_proj_covar=tensor([0.0223, 0.0248, 0.0225, 0.0258, 0.0233, 0.0199, 0.0233, 0.0186],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 15:43:46,492 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40574.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:44:55,479 INFO [train.py:892] (3/4) Epoch 22, batch 1650, loss[loss=0.1674, simple_loss=0.2492, pruned_loss=0.04278, over 19881.00 frames. ], tot_loss[loss=0.1824, simple_loss=0.2557, pruned_loss=0.05458, over 3945963.35 frames. ], batch size: 61, lr: 7.09e-03, grad_scale: 16.0
+2023-03-28 15:45:24,157 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.699e+02 4.293e+02 4.863e+02 5.850e+02 9.887e+02, threshold=9.726e+02, percent-clipped=2.0
+2023-03-28 15:45:27,328 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3573, 3.3813, 4.8542, 3.5689, 3.9919, 3.9093, 2.6485, 2.8959],
+       device='cuda:3'), covar=tensor([0.0799, 0.2861, 0.0429, 0.0985, 0.1471, 0.1130, 0.2354, 0.2375],
+       device='cuda:3'), in_proj_covar=tensor([0.0340, 0.0370, 0.0329, 0.0264, 0.0362, 0.0347, 0.0349, 0.0317],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 15:45:46,686 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40628.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:45:56,747 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8238, 2.7373, 1.7946, 3.2790, 3.0284, 3.2146, 3.3651, 2.6237],
+       device='cuda:3'), covar=tensor([0.0621, 0.0683, 0.1610, 0.0583, 0.0580, 0.0537, 0.0508, 0.0796],
+       device='cuda:3'), in_proj_covar=tensor([0.0136, 0.0134, 0.0139, 0.0139, 0.0123, 0.0124, 0.0136, 0.0138],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 15:46:34,550 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:46:46,989 INFO [train.py:892] (3/4) Epoch 22, batch 1700, loss[loss=0.2227, simple_loss=0.2919, pruned_loss=0.07672, over 19772.00 frames. ], tot_loss[loss=0.1822, simple_loss=0.2554, pruned_loss=0.05451, over 3948420.16 frames. ], batch size: 256, lr: 7.08e-03, grad_scale: 16.0
+2023-03-28 15:46:58,757 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6052, 4.6384, 5.0281, 4.7610, 4.8750, 4.2513, 4.7220, 4.5406],
+       device='cuda:3'), covar=tensor([0.1466, 0.1521, 0.0933, 0.1219, 0.0751, 0.1141, 0.1961, 0.1984],
+       device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0306, 0.0355, 0.0284, 0.0264, 0.0266, 0.0340, 0.0370],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 15:47:20,085 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8989, 3.0096, 3.1012, 3.0831, 2.8885, 2.9230, 2.9278, 3.1024],
+       device='cuda:3'), covar=tensor([0.0206, 0.0265, 0.0235, 0.0205, 0.0309, 0.0267, 0.0333, 0.0300],
+       device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0067, 0.0070, 0.0064, 0.0076, 0.0071, 0.0088, 0.0062],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-28 15:47:45,945 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0
+2023-03-28 15:47:52,431 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:48:31,901 INFO [train.py:892] (3/4) Epoch 22, batch 1750, loss[loss=0.1819, simple_loss=0.2466, pruned_loss=0.0586, over 19750.00 frames. ], tot_loss[loss=0.183, simple_loss=0.2559, pruned_loss=0.05508, over 3947724.17 frames. ], batch size: 140, lr: 7.08e-03, grad_scale: 16.0
+2023-03-28 15:48:44,975 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:48:56,707 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.471e+02 3.805e+02 4.645e+02 5.444e+02 1.360e+03, threshold=9.290e+02, percent-clipped=2.0
+2023-03-28 15:50:07,461 INFO [train.py:892] (3/4) Epoch 22, batch 1800, loss[loss=0.1911, simple_loss=0.2825, pruned_loss=0.04987, over 19841.00 frames. ], tot_loss[loss=0.1847, simple_loss=0.2573, pruned_loss=0.05603, over 3946757.14 frames. ], batch size: 58, lr: 7.07e-03, grad_scale: 16.0
+2023-03-28 15:50:17,843 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-28 15:51:24,883 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0657, 4.7839, 4.7644, 5.1098, 4.8591, 5.2723, 5.2061, 5.4450],
+       device='cuda:3'), covar=tensor([0.0619, 0.0358, 0.0373, 0.0276, 0.0522, 0.0322, 0.0399, 0.0266],
+       device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0165, 0.0189, 0.0163, 0.0162, 0.0147, 0.0142, 0.0187],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 15:51:37,480 INFO [train.py:892] (3/4) Epoch 22, batch 1850, loss[loss=0.1929, simple_loss=0.2835, pruned_loss=0.05114, over 19862.00 frames. ], tot_loss[loss=0.1856, simple_loss=0.2592, pruned_loss=0.05601, over 3945668.32 frames. ], batch size: 58, lr: 7.07e-03, grad_scale: 16.0
+2023-03-28 15:52:42,422 INFO [train.py:892] (3/4) Epoch 23, batch 0, loss[loss=0.1569, simple_loss=0.2393, pruned_loss=0.03728, over 19848.00 frames. ], tot_loss[loss=0.1569, simple_loss=0.2393, pruned_loss=0.03728, over 19848.00 frames. ], batch size: 58, lr: 6.91e-03, grad_scale: 16.0
+2023-03-28 15:52:42,422 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-28 15:53:13,265 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8665, 3.7147, 3.7277, 3.9567, 3.8040, 3.6895, 3.9745, 4.1494],
+       device='cuda:3'), covar=tensor([0.0605, 0.0391, 0.0454, 0.0306, 0.0552, 0.0593, 0.0403, 0.0260],
+       device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0166, 0.0190, 0.0164, 0.0163, 0.0148, 0.0143, 0.0188],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 15:53:16,618 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0548, 2.7901, 3.4646, 3.3535, 3.7864, 4.1839, 4.1638, 4.1138],
+       device='cuda:3'), covar=tensor([0.0755, 0.1493, 0.0926, 0.0583, 0.0309, 0.0204, 0.0260, 0.0540],
+       device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0168, 0.0173, 0.0144, 0.0125, 0.0121, 0.0115, 0.0109],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 15:53:21,306 INFO [train.py:926] (3/4) Epoch 23, validation: loss=0.1723, simple_loss=0.2475, pruned_loss=0.04853, over 2883724.00 frames. 
+2023-03-28 15:53:21,309 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-28 15:53:38,426 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.877e+02 3.829e+02 4.262e+02 4.921e+02 1.071e+03, threshold=8.525e+02, percent-clipped=1.0
+2023-03-28 15:53:57,381 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.59 vs. limit=5.0
+2023-03-28 15:54:46,245 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0
+2023-03-28 15:54:49,703 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:54:52,531 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2279, 2.4163, 2.6101, 2.3710, 2.7119, 2.7381, 3.1384, 3.3385],
+       device='cuda:3'), covar=tensor([0.0687, 0.1619, 0.1681, 0.2117, 0.1466, 0.1430, 0.0669, 0.0653],
+       device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0237, 0.0260, 0.0249, 0.0290, 0.0250, 0.0218, 0.0242],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 15:54:56,761 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:55:18,634 INFO [train.py:892] (3/4) Epoch 23, batch 50, loss[loss=0.1804, simple_loss=0.2459, pruned_loss=0.05743, over 19810.00 frames. ], tot_loss[loss=0.1808, simple_loss=0.2529, pruned_loss=0.0544, over 890889.30 frames. ], batch size: 147, lr: 6.91e-03, grad_scale: 16.0
+2023-03-28 15:55:28,659 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40865.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:55:48,778 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40874.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:56:49,857 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:56:52,083 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:57:16,972 INFO [train.py:892] (3/4) Epoch 23, batch 100, loss[loss=0.1851, simple_loss=0.268, pruned_loss=0.05105, over 19893.00 frames. ], tot_loss[loss=0.18, simple_loss=0.2529, pruned_loss=0.05351, over 1568061.93 frames. ], batch size: 91, lr: 6.90e-03, grad_scale: 16.0
+2023-03-28 15:57:32,517 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.530e+02 4.075e+02 4.882e+02 5.777e+02 1.089e+03, threshold=9.764e+02, percent-clipped=3.0
+2023-03-28 15:57:44,679 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40922.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:57:54,330 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40926.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:57:58,499 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40928.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:59:13,799 INFO [train.py:892] (3/4) Epoch 23, batch 150, loss[loss=0.1689, simple_loss=0.2437, pruned_loss=0.04706, over 19535.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2508, pruned_loss=0.0523, over 2097159.70 frames. ], batch size: 54, lr: 6.90e-03, grad_scale: 16.0
+2023-03-28 15:59:14,907 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40961.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 15:59:45,048 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9963, 4.8050, 4.7628, 5.0716, 4.8854, 5.4774, 5.0642, 5.1448],
+       device='cuda:3'), covar=tensor([0.0883, 0.0481, 0.0515, 0.0410, 0.0664, 0.0370, 0.0578, 0.0782],
+       device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0166, 0.0188, 0.0164, 0.0163, 0.0148, 0.0142, 0.0188],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 15:59:46,969 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 15:59:47,526 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.29 vs. limit=5.0
+2023-03-28 16:00:11,716 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40986.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:00:14,740 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.82 vs. limit=5.0
+2023-03-28 16:00:40,897 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:01:02,021 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:01:09,791 INFO [train.py:892] (3/4) Epoch 23, batch 200, loss[loss=0.1694, simple_loss=0.2425, pruned_loss=0.04818, over 19676.00 frames. ], tot_loss[loss=0.1792, simple_loss=0.2528, pruned_loss=0.05277, over 2507466.58 frames. ], batch size: 73, lr: 6.89e-03, grad_scale: 16.0
+2023-03-28 16:01:24,976 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.565e+02 4.412e+02 5.143e+02 6.492e+02 1.088e+03, threshold=1.029e+03, percent-clipped=2.0
+2023-03-28 16:01:35,325 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8961, 3.3388, 3.7045, 3.4438, 4.1624, 4.0635, 4.6047, 5.2352],
+       device='cuda:3'), covar=tensor([0.0356, 0.1522, 0.1298, 0.1916, 0.1363, 0.1220, 0.0540, 0.0376],
+       device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0234, 0.0257, 0.0247, 0.0286, 0.0249, 0.0216, 0.0239],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 16:02:08,403 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41034.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:02:30,318 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41045.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:03:00,142 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:03:05,879 INFO [train.py:892] (3/4) Epoch 23, batch 250, loss[loss=0.1671, simple_loss=0.2386, pruned_loss=0.0478, over 19564.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.2527, pruned_loss=0.053, over 2826511.09 frames. ], batch size: 41, lr: 6.89e-03, grad_scale: 16.0
+2023-03-28 16:04:44,505 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:04:55,357 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41106.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:05:07,151 INFO [train.py:892] (3/4) Epoch 23, batch 300, loss[loss=0.1674, simple_loss=0.2504, pruned_loss=0.04215, over 19864.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2519, pruned_loss=0.05217, over 3077138.56 frames. ], batch size: 46, lr: 6.89e-03, grad_scale: 16.0
+2023-03-28 16:05:08,114 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9029, 3.8809, 4.2944, 4.0495, 4.1854, 3.7581, 3.9976, 3.7927],
+       device='cuda:3'), covar=tensor([0.1426, 0.1691, 0.1017, 0.1303, 0.1106, 0.1221, 0.1957, 0.2146],
+       device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0305, 0.0353, 0.0284, 0.0262, 0.0262, 0.0340, 0.0367],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 16:05:23,592 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.930e+02 3.981e+02 4.919e+02 5.993e+02 1.063e+03, threshold=9.839e+02, percent-clipped=1.0
+2023-03-28 16:06:37,259 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:07:05,676 INFO [train.py:892] (3/4) Epoch 23, batch 350, loss[loss=0.1568, simple_loss=0.2327, pruned_loss=0.04039, over 19801.00 frames. ], tot_loss[loss=0.1813, simple_loss=0.2549, pruned_loss=0.05383, over 3268070.90 frames. ], batch size: 45, lr: 6.88e-03, grad_scale: 16.0
+2023-03-28 16:07:11,174 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41163.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:08:27,634 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:09:03,419 INFO [train.py:892] (3/4) Epoch 23, batch 400, loss[loss=0.18, simple_loss=0.2577, pruned_loss=0.05116, over 19813.00 frames. ], tot_loss[loss=0.1802, simple_loss=0.254, pruned_loss=0.05324, over 3419647.97 frames. ], batch size: 67, lr: 6.88e-03, grad_scale: 16.0
+2023-03-28 16:09:22,951 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.551e+02 4.181e+02 5.035e+02 6.093e+02 9.382e+02, threshold=1.007e+03, percent-clipped=0.0
+2023-03-28 16:09:31,482 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:09:40,424 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0
+2023-03-28 16:09:42,002 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6916, 2.8877, 4.7775, 4.0201, 4.5340, 4.6665, 4.5719, 4.3293],
+       device='cuda:3'), covar=tensor([0.0422, 0.0815, 0.0087, 0.0896, 0.0117, 0.0166, 0.0137, 0.0150],
+       device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0098, 0.0081, 0.0148, 0.0078, 0.0091, 0.0086, 0.0079],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0002, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 16:10:46,706 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3225, 4.0055, 4.1571, 4.3284, 4.0420, 4.2770, 4.4530, 4.5943],
+       device='cuda:3'), covar=tensor([0.0613, 0.0428, 0.0493, 0.0361, 0.0663, 0.0563, 0.0422, 0.0297],
+       device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0164, 0.0186, 0.0162, 0.0161, 0.0146, 0.0140, 0.0184],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 16:10:58,846 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41256.0, num_to_drop=1, layers_to_drop={3}
+2023-03-28 16:11:01,204 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.80 vs. limit=5.0
+2023-03-28 16:11:08,155 INFO [train.py:892] (3/4) Epoch 23, batch 450, loss[loss=0.173, simple_loss=0.2431, pruned_loss=0.05146, over 19793.00 frames. ], tot_loss[loss=0.1819, simple_loss=0.2554, pruned_loss=0.05425, over 3537286.33 frames. ], batch size: 65, lr: 6.87e-03, grad_scale: 16.0
+2023-03-28 16:12:57,597 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:13:05,207 INFO [train.py:892] (3/4) Epoch 23, batch 500, loss[loss=0.1681, simple_loss=0.2445, pruned_loss=0.04587, over 19710.00 frames. ], tot_loss[loss=0.1798, simple_loss=0.253, pruned_loss=0.05329, over 3629580.39 frames. ], batch size: 61, lr: 6.87e-03, grad_scale: 16.0
+2023-03-28 16:13:24,682 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.561e+02 3.916e+02 4.619e+02 5.416e+02 1.072e+03, threshold=9.239e+02, percent-clipped=2.0
+2023-03-28 16:13:59,659 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=5.11 vs. limit=5.0
+2023-03-28 16:14:51,540 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:14:56,409 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:15:11,420 INFO [train.py:892] (3/4) Epoch 23, batch 550, loss[loss=0.1697, simple_loss=0.2617, pruned_loss=0.03882, over 19667.00 frames. ], tot_loss[loss=0.1809, simple_loss=0.2542, pruned_loss=0.05378, over 3700691.68 frames. ], batch size: 55, lr: 6.87e-03, grad_scale: 16.0
+2023-03-28 16:15:42,189 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2344, 2.5160, 3.5312, 2.8846, 3.0537, 2.9788, 2.1699, 2.3029],
+       device='cuda:3'), covar=tensor([0.1163, 0.2932, 0.0651, 0.0969, 0.1679, 0.1403, 0.2300, 0.2590],
+       device='cuda:3'), in_proj_covar=tensor([0.0341, 0.0370, 0.0329, 0.0265, 0.0362, 0.0347, 0.0350, 0.0316],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 16:16:06,641 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-28 16:16:51,556 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:17:14,962 INFO [train.py:892] (3/4) Epoch 23, batch 600, loss[loss=0.1512, simple_loss=0.2243, pruned_loss=0.03905, over 19764.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.2529, pruned_loss=0.05291, over 3756654.12 frames. ], batch size: 152, lr: 6.86e-03, grad_scale: 16.0
+2023-03-28 16:17:30,507 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.341e+02 3.739e+02 4.752e+02 5.799e+02 9.691e+02, threshold=9.504e+02, percent-clipped=1.0
+2023-03-28 16:19:02,771 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:19:09,254 INFO [train.py:892] (3/4) Epoch 23, batch 650, loss[loss=0.2494, simple_loss=0.3008, pruned_loss=0.09902, over 19736.00 frames. ], tot_loss[loss=0.1788, simple_loss=0.2519, pruned_loss=0.0528, over 3800553.25 frames. ], batch size: 291, lr: 6.86e-03, grad_scale: 16.0
+2023-03-28 16:21:07,577 INFO [train.py:892] (3/4) Epoch 23, batch 700, loss[loss=0.2034, simple_loss=0.275, pruned_loss=0.0659, over 19902.00 frames. ], tot_loss[loss=0.1791, simple_loss=0.2528, pruned_loss=0.05272, over 3832308.06 frames. ], batch size: 50, lr: 6.85e-03, grad_scale: 16.0
+2023-03-28 16:21:25,096 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.559e+02 3.930e+02 4.652e+02 5.850e+02 9.862e+02, threshold=9.304e+02, percent-clipped=2.0
+2023-03-28 16:21:32,038 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:22:57,186 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41556.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 16:23:06,386 INFO [train.py:892] (3/4) Epoch 23, batch 750, loss[loss=0.1869, simple_loss=0.2618, pruned_loss=0.056, over 19818.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2516, pruned_loss=0.05228, over 3860607.11 frames. ], batch size: 93, lr: 6.85e-03, grad_scale: 16.0
+2023-03-28 16:23:24,560 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:23:31,119 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.77 vs. limit=5.0
+2023-03-28 16:24:46,084 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0
+2023-03-28 16:24:47,300 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41604.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:25:00,932 INFO [train.py:892] (3/4) Epoch 23, batch 800, loss[loss=0.1772, simple_loss=0.2592, pruned_loss=0.04755, over 19826.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2511, pruned_loss=0.05204, over 3881710.26 frames. ], batch size: 57, lr: 6.85e-03, grad_scale: 16.0
+2023-03-28 16:25:18,879 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.557e+02 3.999e+02 4.823e+02 6.128e+02 1.113e+03, threshold=9.646e+02, percent-clipped=2.0
+2023-03-28 16:26:43,454 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:26:45,852 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41654.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:27:02,103 INFO [train.py:892] (3/4) Epoch 23, batch 850, loss[loss=0.1712, simple_loss=0.2432, pruned_loss=0.04965, over 19773.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2512, pruned_loss=0.05175, over 3895230.28 frames. ], batch size: 169, lr: 6.84e-03, grad_scale: 16.0
+2023-03-28 16:28:35,327 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:28:35,442 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:28:59,776 INFO [train.py:892] (3/4) Epoch 23, batch 900, loss[loss=0.1796, simple_loss=0.252, pruned_loss=0.05357, over 19656.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2509, pruned_loss=0.05149, over 3907800.13 frames. ], batch size: 67, lr: 6.84e-03, grad_scale: 16.0
+2023-03-28 16:29:10,802 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41715.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 16:29:18,873 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.536e+02 3.998e+02 4.733e+02 5.736e+02 9.757e+02, threshold=9.466e+02, percent-clipped=2.0
+2023-03-28 16:30:08,573 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-28 16:30:31,739 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:30:54,498 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41758.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:31:00,102 INFO [train.py:892] (3/4) Epoch 23, batch 950, loss[loss=0.1902, simple_loss=0.2637, pruned_loss=0.05833, over 19662.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2512, pruned_loss=0.05164, over 3916556.61 frames. ], batch size: 55, lr: 6.83e-03, grad_scale: 16.0
+2023-03-28 16:32:13,771 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0
+2023-03-28 16:32:53,107 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:33:02,647 INFO [train.py:892] (3/4) Epoch 23, batch 1000, loss[loss=0.217, simple_loss=0.2858, pruned_loss=0.07413, over 19759.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2517, pruned_loss=0.05203, over 3924080.40 frames. ], batch size: 256, lr: 6.83e-03, grad_scale: 16.0
+2023-03-28 16:33:19,754 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.497e+02 4.077e+02 4.736e+02 5.828e+02 1.008e+03, threshold=9.473e+02, percent-clipped=1.0
+2023-03-28 16:33:44,038 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9196, 4.5172, 4.6566, 4.4490, 4.8614, 3.1574, 4.0878, 2.5577],
+       device='cuda:3'), covar=tensor([0.0152, 0.0206, 0.0116, 0.0175, 0.0118, 0.0904, 0.0700, 0.1236],
+       device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0137, 0.0108, 0.0128, 0.0112, 0.0128, 0.0138, 0.0122],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 16:34:38,001 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2540, 3.4773, 3.6435, 4.3065, 2.8061, 3.3148, 2.6857, 2.6877],
+       device='cuda:3'), covar=tensor([0.0548, 0.2123, 0.1031, 0.0340, 0.2138, 0.0964, 0.1318, 0.1707],
+       device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0334, 0.0242, 0.0187, 0.0244, 0.0200, 0.0211, 0.0212],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 16:35:00,312 INFO [train.py:892] (3/4) Epoch 23, batch 1050, loss[loss=0.1778, simple_loss=0.261, pruned_loss=0.04732, over 19941.00 frames. ], tot_loss[loss=0.179, simple_loss=0.2529, pruned_loss=0.05258, over 3929722.41 frames. ], batch size: 52, lr: 6.83e-03, grad_scale: 16.0
+2023-03-28 16:36:36,204 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0
+2023-03-28 16:36:49,531 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:37:01,116 INFO [train.py:892] (3/4) Epoch 23, batch 1100, loss[loss=0.1631, simple_loss=0.2428, pruned_loss=0.04167, over 19802.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.2526, pruned_loss=0.05244, over 3933048.55 frames. ], batch size: 167, lr: 6.82e-03, grad_scale: 16.0
+2023-03-28 16:37:20,555 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.656e+02 4.250e+02 5.076e+02 6.187e+02 1.225e+03, threshold=1.015e+03, percent-clipped=1.0
+2023-03-28 16:37:31,549 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5259, 4.4564, 4.9274, 4.7285, 4.8058, 4.2666, 4.6672, 4.4500],
+       device='cuda:3'), covar=tensor([0.1497, 0.1452, 0.0936, 0.1223, 0.0772, 0.0979, 0.1694, 0.1962],
+       device='cuda:3'), in_proj_covar=tensor([0.0279, 0.0304, 0.0351, 0.0282, 0.0262, 0.0262, 0.0336, 0.0364],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 16:37:53,861 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.1370, 2.0838, 2.2688, 2.1080, 2.1900, 2.2173, 2.1342, 2.3217],
+       device='cuda:3'), covar=tensor([0.0289, 0.0287, 0.0243, 0.0288, 0.0365, 0.0286, 0.0388, 0.0256],
+       device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0067, 0.0070, 0.0064, 0.0076, 0.0071, 0.0088, 0.0063],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-28 16:38:31,206 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41948.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 16:38:58,877 INFO [train.py:892] (3/4) Epoch 23, batch 1150, loss[loss=0.2709, simple_loss=0.3407, pruned_loss=0.1005, over 19411.00 frames. ], tot_loss[loss=0.1799, simple_loss=0.2532, pruned_loss=0.05323, over 3936418.80 frames. ], batch size: 412, lr: 6.82e-03, grad_scale: 16.0
+2023-03-28 16:39:13,911 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:40:59,209 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42009.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 16:41:01,797 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42010.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 16:41:03,127 INFO [train.py:892] (3/4) Epoch 23, batch 1200, loss[loss=0.1777, simple_loss=0.2527, pruned_loss=0.05138, over 19922.00 frames. ], tot_loss[loss=0.1796, simple_loss=0.253, pruned_loss=0.05312, over 3939305.89 frames. ], batch size: 45, lr: 6.81e-03, grad_scale: 16.0
+2023-03-28 16:41:23,895 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.867e+02 3.967e+02 4.687e+02 5.246e+02 9.664e+02, threshold=9.374e+02, percent-clipped=0.0
+2023-03-28 16:41:34,036 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6403, 3.4809, 3.9020, 3.0086, 4.0172, 3.2287, 3.4800, 3.9888],
+       device='cuda:3'), covar=tensor([0.0512, 0.0439, 0.0453, 0.0736, 0.0308, 0.0478, 0.0391, 0.0274],
+       device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0079, 0.0077, 0.0106, 0.0073, 0.0074, 0.0071, 0.0064],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 16:43:04,807 INFO [train.py:892] (3/4) Epoch 23, batch 1250, loss[loss=0.163, simple_loss=0.2478, pruned_loss=0.03915, over 19781.00 frames. ], tot_loss[loss=0.1789, simple_loss=0.2523, pruned_loss=0.05276, over 3941509.79 frames. ], batch size: 53, lr: 6.81e-03, grad_scale: 16.0
+2023-03-28 16:44:12,123 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.58 vs. limit=5.0
+2023-03-28 16:44:59,386 INFO [train.py:892] (3/4) Epoch 23, batch 1300, loss[loss=0.1795, simple_loss=0.242, pruned_loss=0.05849, over 19797.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.2516, pruned_loss=0.05239, over 3943417.22 frames. ], batch size: 126, lr: 6.81e-03, grad_scale: 16.0
+2023-03-28 16:45:10,071 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0
+2023-03-28 16:45:16,101 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.506e+02 3.507e+02 4.427e+02 5.595e+02 1.023e+03, threshold=8.855e+02, percent-clipped=1.0
+2023-03-28 16:46:11,597 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6652, 3.9317, 4.0663, 4.7789, 3.2042, 3.5557, 2.8791, 2.7551],
+       device='cuda:3'), covar=tensor([0.0439, 0.1939, 0.0772, 0.0274, 0.1870, 0.0954, 0.1317, 0.1709],
+       device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0330, 0.0240, 0.0185, 0.0240, 0.0198, 0.0211, 0.0211],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 16:46:38,020 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-28 16:46:59,036 INFO [train.py:892] (3/4) Epoch 23, batch 1350, loss[loss=0.1705, simple_loss=0.25, pruned_loss=0.04549, over 19833.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2513, pruned_loss=0.05227, over 3945772.47 frames. ], batch size: 52, lr: 6.80e-03, grad_scale: 16.0
+2023-03-28 16:47:44,666 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.1739, 2.4915, 2.2394, 1.5896, 2.2668, 2.4490, 2.3642, 2.4648],
+       device='cuda:3'), covar=tensor([0.0377, 0.0277, 0.0293, 0.0572, 0.0376, 0.0320, 0.0243, 0.0225],
+       device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0085, 0.0091, 0.0095, 0.0098, 0.0075, 0.0074, 0.0075],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 16:48:06,535 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7388, 2.3001, 2.5069, 2.9217, 3.3713, 3.5458, 3.6083, 3.6043],
+       device='cuda:3'), covar=tensor([0.1003, 0.1710, 0.1482, 0.0730, 0.0488, 0.0359, 0.0384, 0.0366],
+       device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0168, 0.0172, 0.0143, 0.0125, 0.0122, 0.0116, 0.0109],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 16:48:58,507 INFO [train.py:892] (3/4) Epoch 23, batch 1400, loss[loss=0.1719, simple_loss=0.252, pruned_loss=0.04586, over 19535.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2517, pruned_loss=0.0523, over 3946195.93 frames. ], batch size: 54, lr: 6.80e-03, grad_scale: 16.0
+2023-03-28 16:49:10,515 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1289, 3.1510, 4.9637, 4.2824, 4.5222, 4.8949, 4.7024, 4.5374],
+       device='cuda:3'), covar=tensor([0.0351, 0.0773, 0.0086, 0.0819, 0.0131, 0.0192, 0.0155, 0.0135],
+       device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0101, 0.0082, 0.0150, 0.0080, 0.0094, 0.0087, 0.0081],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0002, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 16:49:18,203 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.596e+02 4.046e+02 4.840e+02 5.531e+02 1.167e+03, threshold=9.681e+02, percent-clipped=2.0
+2023-03-28 16:50:56,034 INFO [train.py:892] (3/4) Epoch 23, batch 1450, loss[loss=0.2142, simple_loss=0.2776, pruned_loss=0.07541, over 19756.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2511, pruned_loss=0.05211, over 3947374.00 frames. ], batch size: 253, lr: 6.79e-03, grad_scale: 16.0
+2023-03-28 16:50:59,086 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:52:13,370 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4876, 2.5328, 2.7710, 2.5660, 2.9113, 2.8832, 3.3518, 3.6510],
+       device='cuda:3'), covar=tensor([0.0718, 0.1743, 0.1771, 0.2029, 0.1626, 0.1530, 0.0676, 0.0684],
+       device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0236, 0.0259, 0.0248, 0.0288, 0.0250, 0.0219, 0.0242],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 16:52:39,757 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42304.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 16:52:55,169 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42310.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 16:52:56,186 INFO [train.py:892] (3/4) Epoch 23, batch 1500, loss[loss=0.1643, simple_loss=0.2416, pruned_loss=0.0435, over 19749.00 frames. ], tot_loss[loss=0.1781, simple_loss=0.2516, pruned_loss=0.05232, over 3948406.39 frames. ], batch size: 89, lr: 6.79e-03, grad_scale: 16.0
+2023-03-28 16:53:12,435 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.820e+02 3.834e+02 4.491e+02 5.475e+02 9.229e+02, threshold=8.983e+02, percent-clipped=0.0
+2023-03-28 16:54:27,760 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.3540, 4.8674, 4.9393, 5.2980, 4.9496, 5.5823, 5.4114, 5.6659],
+       device='cuda:3'), covar=tensor([0.0566, 0.0300, 0.0436, 0.0300, 0.0571, 0.0251, 0.0379, 0.0251],
+       device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0168, 0.0191, 0.0164, 0.0165, 0.0147, 0.0142, 0.0188],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 16:54:46,331 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:54:51,544 INFO [train.py:892] (3/4) Epoch 23, batch 1550, loss[loss=0.1652, simple_loss=0.2395, pruned_loss=0.04545, over 19774.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2511, pruned_loss=0.05163, over 3950136.41 frames. ], batch size: 198, lr: 6.79e-03, grad_scale: 16.0
+2023-03-28 16:55:38,165 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:56:55,995 INFO [train.py:892] (3/4) Epoch 23, batch 1600, loss[loss=0.1693, simple_loss=0.245, pruned_loss=0.04675, over 19531.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2512, pruned_loss=0.05164, over 3950339.98 frames. ], batch size: 46, lr: 6.78e-03, grad_scale: 16.0
+2023-03-28 16:57:13,628 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.225e+02 3.903e+02 4.499e+02 5.665e+02 1.044e+03, threshold=8.998e+02, percent-clipped=1.0
+2023-03-28 16:57:38,685 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0
+2023-03-28 16:58:07,323 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 16:58:53,569 INFO [train.py:892] (3/4) Epoch 23, batch 1650, loss[loss=0.1529, simple_loss=0.2265, pruned_loss=0.03965, over 19951.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2505, pruned_loss=0.05139, over 3951109.98 frames. ], batch size: 46, lr: 6.78e-03, grad_scale: 16.0
+2023-03-28 16:59:44,623 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.83 vs. limit=5.0
+2023-03-28 17:00:37,937 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42506.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:00:47,860 INFO [train.py:892] (3/4) Epoch 23, batch 1700, loss[loss=0.1619, simple_loss=0.2321, pruned_loss=0.04585, over 19791.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2517, pruned_loss=0.05215, over 3951246.88 frames. ], batch size: 111, lr: 6.77e-03, grad_scale: 16.0
+2023-03-28 17:01:09,675 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.489e+02 4.052e+02 4.477e+02 5.328e+02 1.019e+03, threshold=8.953e+02, percent-clipped=3.0
+2023-03-28 17:02:11,797 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9724, 4.9295, 5.4695, 4.9851, 4.3414, 5.2009, 5.1167, 5.6133],
+       device='cuda:3'), covar=tensor([0.0825, 0.0316, 0.0319, 0.0300, 0.0734, 0.0397, 0.0380, 0.0258],
+       device='cuda:3'), in_proj_covar=tensor([0.0271, 0.0213, 0.0212, 0.0221, 0.0199, 0.0226, 0.0222, 0.0205],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 17:02:36,517 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.2090, 4.7815, 4.8902, 4.6460, 5.1966, 3.0882, 3.9796, 2.5645],
+       device='cuda:3'), covar=tensor([0.0251, 0.0222, 0.0178, 0.0217, 0.0185, 0.1076, 0.1043, 0.1675],
+       device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0139, 0.0110, 0.0129, 0.0114, 0.0130, 0.0139, 0.0123],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 17:02:39,531 INFO [train.py:892] (3/4) Epoch 23, batch 1750, loss[loss=0.162, simple_loss=0.2388, pruned_loss=0.04265, over 19818.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.251, pruned_loss=0.05165, over 3950583.85 frames. ], batch size: 67, lr: 6.77e-03, grad_scale: 16.0
+2023-03-28 17:02:40,708 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0
+2023-03-28 17:02:43,401 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:02:52,571 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2408, 3.9524, 4.0476, 4.2516, 4.0376, 4.3514, 4.3818, 4.5402],
+       device='cuda:3'), covar=tensor([0.0710, 0.0380, 0.0553, 0.0364, 0.0719, 0.0440, 0.0429, 0.0301],
+       device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0169, 0.0193, 0.0166, 0.0166, 0.0149, 0.0144, 0.0189],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 17:02:52,698 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:03:21,520 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4171, 3.1438, 3.2044, 3.3919, 3.2253, 3.3074, 3.5264, 3.6884],
+       device='cuda:3'), covar=tensor([0.0761, 0.0521, 0.0641, 0.0458, 0.0828, 0.0749, 0.0525, 0.0387],
+       device='cuda:3'), in_proj_covar=tensor([0.0146, 0.0170, 0.0194, 0.0167, 0.0167, 0.0150, 0.0145, 0.0190],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 17:04:07,905 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42604.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 17:04:20,174 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:04:21,417 INFO [train.py:892] (3/4) Epoch 23, batch 1800, loss[loss=0.1433, simple_loss=0.224, pruned_loss=0.03135, over 19775.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2513, pruned_loss=0.05174, over 3950105.55 frames. ], batch size: 52, lr: 6.77e-03, grad_scale: 16.0
+2023-03-28 17:04:37,546 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.634e+02 3.841e+02 4.565e+02 5.704e+02 1.447e+03, threshold=9.129e+02, percent-clipped=3.0
+2023-03-28 17:05:05,691 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3507, 4.5057, 5.0192, 4.5075, 4.2584, 4.9322, 4.7881, 5.2722],
+       device='cuda:3'), covar=tensor([0.1421, 0.0457, 0.0573, 0.0439, 0.0754, 0.0493, 0.0442, 0.0392],
+       device='cuda:3'), in_proj_covar=tensor([0.0274, 0.0214, 0.0214, 0.0223, 0.0201, 0.0228, 0.0224, 0.0206],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 17:05:16,595 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.3324, 5.5407, 5.8495, 5.7554, 5.5899, 5.3210, 5.4887, 5.4664],
+       device='cuda:3'), covar=tensor([0.1500, 0.1360, 0.0877, 0.1053, 0.0705, 0.0861, 0.2085, 0.1957],
+       device='cuda:3'), in_proj_covar=tensor([0.0280, 0.0307, 0.0352, 0.0281, 0.0264, 0.0262, 0.0336, 0.0366],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 17:05:32,201 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:05:37,678 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42652.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 17:05:49,617 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.3252, 4.7993, 4.9351, 4.7242, 5.2247, 3.2471, 4.1065, 2.4775],
+       device='cuda:3'), covar=tensor([0.0161, 0.0180, 0.0128, 0.0189, 0.0118, 0.0892, 0.0981, 0.1484],
+       device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0140, 0.0110, 0.0129, 0.0115, 0.0130, 0.0140, 0.0123],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 17:05:54,213 INFO [train.py:892] (3/4) Epoch 23, batch 1850, loss[loss=0.1687, simple_loss=0.2482, pruned_loss=0.04461, over 19845.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2531, pruned_loss=0.05144, over 3948169.00 frames. ], batch size: 57, lr: 6.76e-03, grad_scale: 16.0
+2023-03-28 17:06:55,120 INFO [train.py:892] (3/4) Epoch 24, batch 0, loss[loss=0.1583, simple_loss=0.2419, pruned_loss=0.03737, over 19746.00 frames. ], tot_loss[loss=0.1583, simple_loss=0.2419, pruned_loss=0.03737, over 19746.00 frames. ], batch size: 97, lr: 6.62e-03, grad_scale: 16.0
+2023-03-28 17:06:55,120 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-28 17:07:24,186 INFO [train.py:926] (3/4) Epoch 24, validation: loss=0.1738, simple_loss=0.2478, pruned_loss=0.0499, over 2883724.00 frames. 
+2023-03-28 17:07:24,187 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-28 17:07:25,334 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5185, 4.4309, 4.9485, 4.4786, 4.0889, 4.7409, 4.6105, 5.0906],
+       device='cuda:3'), covar=tensor([0.0942, 0.0373, 0.0342, 0.0369, 0.0906, 0.0504, 0.0460, 0.0292],
+       device='cuda:3'), in_proj_covar=tensor([0.0275, 0.0215, 0.0214, 0.0224, 0.0201, 0.0229, 0.0224, 0.0206],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 17:09:11,978 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42710.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:09:24,500 INFO [train.py:892] (3/4) Epoch 24, batch 50, loss[loss=0.1766, simple_loss=0.2525, pruned_loss=0.05037, over 19926.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2447, pruned_loss=0.04777, over 891720.02 frames. ], batch size: 51, lr: 6.61e-03, grad_scale: 16.0
+2023-03-28 17:09:30,668 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.706e+02 3.984e+02 4.809e+02 5.646e+02 9.126e+02, threshold=9.617e+02, percent-clipped=0.0
+2023-03-28 17:10:09,791 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:11:17,035 INFO [train.py:892] (3/4) Epoch 24, batch 100, loss[loss=0.1692, simple_loss=0.2475, pruned_loss=0.04542, over 19788.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2458, pruned_loss=0.04804, over 1570229.09 frames. ], batch size: 83, lr: 6.61e-03, grad_scale: 16.0
+2023-03-28 17:11:32,394 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7267, 3.9681, 4.0724, 4.8296, 3.1795, 3.4596, 3.0251, 2.9775],
+       device='cuda:3'), covar=tensor([0.0418, 0.1929, 0.0854, 0.0306, 0.1925, 0.0939, 0.1201, 0.1572],
+       device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0332, 0.0242, 0.0189, 0.0243, 0.0199, 0.0212, 0.0210],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 17:12:45,591 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6661, 3.7179, 2.2073, 3.8759, 3.9849, 1.7090, 3.2761, 3.0740],
+       device='cuda:3'), covar=tensor([0.0728, 0.0861, 0.2863, 0.0840, 0.0584, 0.2894, 0.1176, 0.0868],
+       device='cuda:3'), in_proj_covar=tensor([0.0224, 0.0248, 0.0225, 0.0258, 0.0236, 0.0199, 0.0233, 0.0187],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 17:13:12,214 INFO [train.py:892] (3/4) Epoch 24, batch 150, loss[loss=0.1649, simple_loss=0.2323, pruned_loss=0.04877, over 19770.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2484, pruned_loss=0.0491, over 2096959.37 frames. ], batch size: 182, lr: 6.60e-03, grad_scale: 16.0
+2023-03-28 17:13:19,941 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.303e+02 3.674e+02 4.384e+02 5.229e+02 7.320e+02, threshold=8.767e+02, percent-clipped=0.0
+2023-03-28 17:14:17,385 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.8494, 2.1050, 1.9288, 1.3539, 1.9833, 2.0741, 1.9589, 2.0208],
+       device='cuda:3'), covar=tensor([0.0382, 0.0270, 0.0310, 0.0563, 0.0393, 0.0284, 0.0266, 0.0265],
+       device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0083, 0.0089, 0.0093, 0.0096, 0.0074, 0.0073, 0.0074],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 17:14:33,824 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9373, 2.8607, 1.8623, 3.4786, 3.2432, 3.4412, 3.5375, 2.7123],
+       device='cuda:3'), covar=tensor([0.0638, 0.0716, 0.1657, 0.0704, 0.0600, 0.0470, 0.0606, 0.0827],
+       device='cuda:3'), in_proj_covar=tensor([0.0138, 0.0136, 0.0140, 0.0142, 0.0125, 0.0124, 0.0138, 0.0140],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 17:14:41,030 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4085, 3.3302, 3.7073, 3.3634, 3.2007, 3.6801, 3.5066, 3.7766],
+       device='cuda:3'), covar=tensor([0.1015, 0.0422, 0.0428, 0.0459, 0.1472, 0.0547, 0.0470, 0.0413],
+       device='cuda:3'), in_proj_covar=tensor([0.0274, 0.0215, 0.0214, 0.0224, 0.0202, 0.0228, 0.0223, 0.0207],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 17:14:53,515 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.7423, 6.0975, 6.1046, 5.9593, 5.8651, 6.0842, 5.4361, 5.5229],
+       device='cuda:3'), covar=tensor([0.0351, 0.0366, 0.0384, 0.0354, 0.0431, 0.0415, 0.0616, 0.0850],
+       device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0261, 0.0280, 0.0240, 0.0244, 0.0232, 0.0250, 0.0297],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 17:14:53,598 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:15:01,980 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:15:09,274 INFO [train.py:892] (3/4) Epoch 24, batch 200, loss[loss=0.2237, simple_loss=0.3262, pruned_loss=0.06061, over 18933.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2499, pruned_loss=0.0498, over 2507996.61 frames. ], batch size: 514, lr: 6.60e-03, grad_scale: 16.0
+2023-03-28 17:15:39,890 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42879.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:16:57,894 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.3594, 2.7997, 2.4193, 1.9184, 2.4968, 2.6187, 2.6863, 2.7035],
+       device='cuda:3'), covar=tensor([0.0367, 0.0257, 0.0313, 0.0622, 0.0396, 0.0291, 0.0232, 0.0230],
+       device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0084, 0.0090, 0.0094, 0.0097, 0.0074, 0.0074, 0.0075],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 17:17:07,545 INFO [train.py:892] (3/4) Epoch 24, batch 250, loss[loss=0.1595, simple_loss=0.2285, pruned_loss=0.04527, over 19840.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2508, pruned_loss=0.05036, over 2827725.00 frames. ], batch size: 160, lr: 6.60e-03, grad_scale: 16.0
+2023-03-28 17:17:15,134 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.878e+02 4.086e+02 4.768e+02 5.713e+02 1.218e+03, threshold=9.536e+02, percent-clipped=1.0
+2023-03-28 17:17:18,167 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:18:08,750 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:19:11,295 INFO [train.py:892] (3/4) Epoch 24, batch 300, loss[loss=0.1764, simple_loss=0.2466, pruned_loss=0.05304, over 19868.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2512, pruned_loss=0.05077, over 3077295.43 frames. ], batch size: 104, lr: 6.59e-03, grad_scale: 16.0
+2023-03-28 17:19:28,758 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0717, 4.9182, 5.4457, 4.9997, 4.2673, 5.1968, 5.0858, 5.6142],
+       device='cuda:3'), covar=tensor([0.0827, 0.0345, 0.0333, 0.0312, 0.0872, 0.0424, 0.0372, 0.0294],
+       device='cuda:3'), in_proj_covar=tensor([0.0274, 0.0215, 0.0214, 0.0225, 0.0202, 0.0229, 0.0224, 0.0208],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 17:20:47,294 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:21:09,365 INFO [train.py:892] (3/4) Epoch 24, batch 350, loss[loss=0.1634, simple_loss=0.2457, pruned_loss=0.0406, over 19871.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2498, pruned_loss=0.05032, over 3271676.59 frames. ], batch size: 48, lr: 6.59e-03, grad_scale: 16.0
+2023-03-28 17:21:15,743 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.627e+02 4.222e+02 4.669e+02 5.474e+02 9.136e+02, threshold=9.338e+02, percent-clipped=0.0
+2023-03-28 17:21:58,421 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:23:09,366 INFO [train.py:892] (3/4) Epoch 24, batch 400, loss[loss=0.1525, simple_loss=0.229, pruned_loss=0.03805, over 19715.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2499, pruned_loss=0.05083, over 3422916.92 frames. ], batch size: 60, lr: 6.59e-03, grad_scale: 16.0
+2023-03-28 17:23:55,616 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:25:13,723 INFO [train.py:892] (3/4) Epoch 24, batch 450, loss[loss=0.2362, simple_loss=0.3054, pruned_loss=0.08347, over 19561.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2513, pruned_loss=0.05169, over 3538769.60 frames. ], batch size: 376, lr: 6.58e-03, grad_scale: 16.0
+2023-03-28 17:25:20,845 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.554e+02 4.148e+02 4.850e+02 5.814e+02 7.870e+02, threshold=9.701e+02, percent-clipped=0.0
+2023-03-28 17:25:21,868 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5249, 3.8760, 4.0005, 4.6863, 3.0381, 3.4723, 3.0245, 2.9436],
+       device='cuda:3'), covar=tensor([0.0521, 0.1890, 0.0877, 0.0354, 0.2191, 0.0980, 0.1164, 0.1632],
+       device='cuda:3'), in_proj_covar=tensor([0.0231, 0.0329, 0.0239, 0.0188, 0.0240, 0.0196, 0.0208, 0.0209],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 17:25:28,141 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4108, 5.7317, 5.7444, 5.6673, 5.3814, 5.7153, 5.0866, 5.1759],
+       device='cuda:3'), covar=tensor([0.0392, 0.0375, 0.0483, 0.0375, 0.0566, 0.0507, 0.0689, 0.0913],
+       device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0261, 0.0278, 0.0238, 0.0243, 0.0232, 0.0249, 0.0296],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 17:27:01,487 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:27:08,760 INFO [train.py:892] (3/4) Epoch 24, batch 500, loss[loss=0.1893, simple_loss=0.2668, pruned_loss=0.05589, over 19572.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2504, pruned_loss=0.05112, over 3630626.99 frames. ], batch size: 60, lr: 6.58e-03, grad_scale: 16.0
+2023-03-28 17:28:57,140 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43210.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:29:09,464 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:29:10,742 INFO [train.py:892] (3/4) Epoch 24, batch 550, loss[loss=0.2463, simple_loss=0.3143, pruned_loss=0.08909, over 19608.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2508, pruned_loss=0.051, over 3699254.13 frames. ], batch size: 387, lr: 6.57e-03, grad_scale: 16.0
+2023-03-28 17:29:18,533 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.622e+02 3.772e+02 4.624e+02 5.545e+02 8.960e+02, threshold=9.249e+02, percent-clipped=0.0
+2023-03-28 17:29:59,349 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43235.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:30:14,263 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.79 vs. limit=5.0
+2023-03-28 17:31:16,245 INFO [train.py:892] (3/4) Epoch 24, batch 600, loss[loss=0.1596, simple_loss=0.2384, pruned_loss=0.04038, over 19739.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2514, pruned_loss=0.05094, over 3754052.70 frames. ], batch size: 44, lr: 6.57e-03, grad_scale: 16.0
+2023-03-28 17:32:50,201 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:33:14,102 INFO [train.py:892] (3/4) Epoch 24, batch 650, loss[loss=0.2056, simple_loss=0.2777, pruned_loss=0.06677, over 19713.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2498, pruned_loss=0.05071, over 3798502.91 frames. ], batch size: 295, lr: 6.57e-03, grad_scale: 16.0
+2023-03-28 17:33:20,466 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.750e+02 3.990e+02 4.608e+02 6.131e+02 1.046e+03, threshold=9.216e+02, percent-clipped=2.0
+2023-03-28 17:34:34,766 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43353.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:35:00,960 INFO [train.py:892] (3/4) Epoch 24, batch 700, loss[loss=0.1423, simple_loss=0.2131, pruned_loss=0.03578, over 19810.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2494, pruned_loss=0.05045, over 3831064.73 frames. ], batch size: 47, lr: 6.56e-03, grad_scale: 16.0
+2023-03-28 17:37:05,684 INFO [train.py:892] (3/4) Epoch 24, batch 750, loss[loss=0.1582, simple_loss=0.2351, pruned_loss=0.04066, over 19815.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2504, pruned_loss=0.05086, over 3857082.67 frames. ], batch size: 96, lr: 6.56e-03, grad_scale: 16.0
+2023-03-28 17:37:13,200 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.432e+02 3.941e+02 4.702e+02 5.595e+02 1.048e+03, threshold=9.403e+02, percent-clipped=2.0
+2023-03-28 17:37:46,030 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4637, 3.7421, 3.9391, 4.5641, 3.0581, 3.3200, 2.7746, 2.7439],
+       device='cuda:3'), covar=tensor([0.0507, 0.2128, 0.0887, 0.0354, 0.1983, 0.1066, 0.1329, 0.1795],
+       device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0330, 0.0242, 0.0189, 0.0242, 0.0199, 0.0211, 0.0212],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 17:39:06,728 INFO [train.py:892] (3/4) Epoch 24, batch 800, loss[loss=0.1666, simple_loss=0.2396, pruned_loss=0.04683, over 19640.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2499, pruned_loss=0.05022, over 3878723.49 frames. ], batch size: 68, lr: 6.56e-03, grad_scale: 16.0
+2023-03-28 17:39:31,923 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6252, 2.8399, 2.9257, 2.8852, 2.5734, 2.7290, 2.5894, 2.9642],
+       device='cuda:3'), covar=tensor([0.0343, 0.0242, 0.0236, 0.0206, 0.0393, 0.0278, 0.0364, 0.0256],
+       device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0069, 0.0072, 0.0066, 0.0079, 0.0074, 0.0090, 0.0065],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001],
+       device='cuda:3')
+2023-03-28 17:41:03,957 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:41:05,990 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:41:06,996 INFO [train.py:892] (3/4) Epoch 24, batch 850, loss[loss=0.2034, simple_loss=0.2728, pruned_loss=0.06699, over 19690.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2504, pruned_loss=0.05022, over 3893215.23 frames. ], batch size: 265, lr: 6.55e-03, grad_scale: 16.0
+2023-03-28 17:41:14,146 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.358e+02 3.878e+02 4.715e+02 5.531e+02 7.871e+02, threshold=9.429e+02, percent-clipped=0.0
+2023-03-28 17:41:30,252 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-28 17:41:52,212 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43535.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:42:23,907 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0260, 2.4546, 3.1939, 2.7210, 2.7697, 2.8195, 1.9714, 2.1485],
+       device='cuda:3'), covar=tensor([0.1047, 0.2135, 0.0662, 0.0907, 0.1523, 0.1226, 0.2209, 0.2297],
+       device='cuda:3'), in_proj_covar=tensor([0.0341, 0.0371, 0.0331, 0.0266, 0.0362, 0.0350, 0.0351, 0.0320],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 17:43:00,894 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:43:06,611 INFO [train.py:892] (3/4) Epoch 24, batch 900, loss[loss=0.1778, simple_loss=0.2392, pruned_loss=0.05819, over 19888.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2494, pruned_loss=0.05009, over 3905894.57 frames. ], batch size: 176, lr: 6.55e-03, grad_scale: 16.0
+2023-03-28 17:43:27,534 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:43:30,343 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0
+2023-03-28 17:43:45,945 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43583.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:45:04,179 INFO [train.py:892] (3/4) Epoch 24, batch 950, loss[loss=0.1637, simple_loss=0.2425, pruned_loss=0.04244, over 19840.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2495, pruned_loss=0.0497, over 3916588.89 frames. ], batch size: 90, lr: 6.54e-03, grad_scale: 16.0
+2023-03-28 17:45:11,472 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.960e+02 4.203e+02 4.948e+02 5.601e+02 1.021e+03, threshold=9.897e+02, percent-clipped=1.0
+2023-03-28 17:45:20,246 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0
+2023-03-28 17:47:04,241 INFO [train.py:892] (3/4) Epoch 24, batch 1000, loss[loss=0.1661, simple_loss=0.2438, pruned_loss=0.04419, over 19868.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2495, pruned_loss=0.0498, over 3923938.90 frames. ], batch size: 48, lr: 6.54e-03, grad_scale: 16.0
+2023-03-28 17:47:07,612 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3165, 4.5567, 2.5812, 4.8037, 5.0480, 2.0004, 4.1947, 3.5253],
+       device='cuda:3'), covar=tensor([0.0617, 0.0615, 0.2519, 0.0550, 0.0360, 0.2739, 0.0877, 0.0799],
+       device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0250, 0.0227, 0.0261, 0.0241, 0.0201, 0.0235, 0.0189],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 17:47:52,873 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2415, 3.0450, 3.2220, 2.8585, 3.4699, 3.5605, 4.0661, 4.4628],
+       device='cuda:3'), covar=tensor([0.0565, 0.1671, 0.1536, 0.2341, 0.1705, 0.1366, 0.0589, 0.0524],
+       device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0238, 0.0261, 0.0249, 0.0291, 0.0251, 0.0220, 0.0243],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 17:49:06,778 INFO [train.py:892] (3/4) Epoch 24, batch 1050, loss[loss=0.1925, simple_loss=0.269, pruned_loss=0.05805, over 19763.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.249, pruned_loss=0.04974, over 3930749.89 frames. ], batch size: 253, lr: 6.54e-03, grad_scale: 32.0
+2023-03-28 17:49:14,107 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.754e+02 4.065e+02 4.703e+02 5.528e+02 1.039e+03, threshold=9.406e+02, percent-clipped=2.0
+2023-03-28 17:50:19,534 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0
+2023-03-28 17:50:46,500 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1060, 4.0768, 4.4135, 4.0443, 3.7356, 4.2546, 4.0682, 4.4523],
+       device='cuda:3'), covar=tensor([0.0810, 0.0318, 0.0332, 0.0397, 0.1012, 0.0546, 0.0479, 0.0353],
+       device='cuda:3'), in_proj_covar=tensor([0.0273, 0.0213, 0.0214, 0.0224, 0.0201, 0.0227, 0.0222, 0.0206],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 17:50:50,293 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4789, 1.9066, 2.2790, 2.7532, 3.0965, 3.1501, 3.0688, 3.2148],
+       device='cuda:3'), covar=tensor([0.0976, 0.1719, 0.1350, 0.0715, 0.0450, 0.0367, 0.0459, 0.0367],
+       device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0167, 0.0172, 0.0146, 0.0127, 0.0123, 0.0117, 0.0109],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 17:51:07,568 INFO [train.py:892] (3/4) Epoch 24, batch 1100, loss[loss=0.1943, simple_loss=0.2573, pruned_loss=0.06562, over 19888.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2498, pruned_loss=0.05054, over 3935449.04 frames. ], batch size: 176, lr: 6.53e-03, grad_scale: 16.0
+2023-03-28 17:51:11,194 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7700, 4.4113, 4.5625, 4.3440, 4.7498, 3.2888, 3.9644, 2.4770],
+       device='cuda:3'), covar=tensor([0.0163, 0.0209, 0.0130, 0.0167, 0.0126, 0.0811, 0.0703, 0.1355],
+       device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0142, 0.0112, 0.0131, 0.0116, 0.0132, 0.0142, 0.0125],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 17:51:57,103 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:52:07,239 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-28 17:52:11,128 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6367, 2.7387, 4.1066, 3.0761, 3.4949, 3.1935, 2.3505, 2.3653],
+       device='cuda:3'), covar=tensor([0.1183, 0.3074, 0.0548, 0.1008, 0.1508, 0.1438, 0.2334, 0.2984],
+       device='cuda:3'), in_proj_covar=tensor([0.0344, 0.0376, 0.0334, 0.0270, 0.0367, 0.0354, 0.0356, 0.0323],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 17:52:16,088 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7737, 4.9155, 5.1674, 4.9802, 5.0774, 4.6626, 4.8506, 4.6929],
+       device='cuda:3'), covar=tensor([0.1521, 0.1337, 0.0901, 0.1255, 0.0726, 0.0830, 0.1912, 0.2048],
+       device='cuda:3'), in_proj_covar=tensor([0.0278, 0.0305, 0.0350, 0.0278, 0.0259, 0.0256, 0.0332, 0.0361],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 17:52:16,284 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9282, 3.0810, 3.3103, 2.4880, 3.2751, 2.7976, 3.0637, 3.2272],
+       device='cuda:3'), covar=tensor([0.0534, 0.0444, 0.0453, 0.0795, 0.0369, 0.0462, 0.0438, 0.0311],
+       device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0080, 0.0077, 0.0107, 0.0074, 0.0075, 0.0073, 0.0065],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 17:52:18,667 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.21 vs. limit=5.0
+2023-03-28 17:52:39,844 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-28 17:53:09,088 INFO [train.py:892] (3/4) Epoch 24, batch 1150, loss[loss=0.1705, simple_loss=0.2462, pruned_loss=0.0474, over 19677.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2499, pruned_loss=0.05064, over 3939417.73 frames. ], batch size: 59, lr: 6.53e-03, grad_scale: 16.0
+2023-03-28 17:53:19,343 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.514e+02 3.854e+02 4.753e+02 5.981e+02 1.175e+03, threshold=9.505e+02, percent-clipped=4.0
+2023-03-28 17:53:22,184 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7561, 4.4209, 4.5274, 4.2689, 4.7120, 3.2278, 3.9290, 2.5451],
+       device='cuda:3'), covar=tensor([0.0171, 0.0211, 0.0136, 0.0178, 0.0134, 0.0885, 0.0744, 0.1295],
+       device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0141, 0.0111, 0.0130, 0.0115, 0.0131, 0.0141, 0.0125],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 17:53:24,296 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8005, 3.8533, 2.3717, 4.0551, 4.2179, 1.7858, 3.3961, 3.2039],
+       device='cuda:3'), covar=tensor([0.0711, 0.0872, 0.2705, 0.0840, 0.0580, 0.2993, 0.1109, 0.0841],
+       device='cuda:3'), in_proj_covar=tensor([0.0227, 0.0251, 0.0228, 0.0262, 0.0241, 0.0202, 0.0235, 0.0189],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 17:54:27,302 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:54:50,132 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0
+2023-03-28 17:55:06,386 INFO [train.py:892] (3/4) Epoch 24, batch 1200, loss[loss=0.1718, simple_loss=0.2436, pruned_loss=0.05004, over 19758.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2507, pruned_loss=0.05084, over 3942139.09 frames. ], batch size: 198, lr: 6.53e-03, grad_scale: 16.0
+2023-03-28 17:55:15,709 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43870.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:55:24,617 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2140, 2.9303, 3.2345, 2.9113, 3.4419, 3.3659, 4.0307, 4.4235],
+       device='cuda:3'), covar=tensor([0.0535, 0.1634, 0.1500, 0.2051, 0.1543, 0.1375, 0.0593, 0.0605],
+       device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0236, 0.0259, 0.0248, 0.0288, 0.0250, 0.0219, 0.0242],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 17:55:53,671 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0394, 3.6197, 3.7925, 4.0476, 3.7946, 3.9915, 4.1469, 4.3004],
+       device='cuda:3'), covar=tensor([0.0679, 0.0455, 0.0488, 0.0345, 0.0658, 0.0540, 0.0400, 0.0300],
+       device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0170, 0.0193, 0.0166, 0.0165, 0.0147, 0.0143, 0.0188],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 17:56:08,848 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9998, 5.2989, 5.3080, 5.2271, 4.9438, 5.2745, 4.7586, 4.8491],
+       device='cuda:3'), covar=tensor([0.0435, 0.0460, 0.0483, 0.0431, 0.0541, 0.0533, 0.0704, 0.0902],
+       device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0262, 0.0280, 0.0243, 0.0244, 0.0234, 0.0251, 0.0296],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 17:56:47,459 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:56:51,825 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5894, 4.0394, 4.2377, 4.7336, 2.9442, 3.4306, 2.8887, 2.8685],
+       device='cuda:3'), covar=tensor([0.0519, 0.1633, 0.0719, 0.0338, 0.2101, 0.1026, 0.1169, 0.1578],
+       device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0329, 0.0242, 0.0190, 0.0241, 0.0199, 0.0210, 0.0211],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 17:56:58,255 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1177, 3.4217, 3.5760, 4.0930, 2.7641, 3.1356, 2.5469, 2.5509],
+       device='cuda:3'), covar=tensor([0.0490, 0.2085, 0.0985, 0.0375, 0.2015, 0.0932, 0.1341, 0.1761],
+       device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0329, 0.0242, 0.0190, 0.0241, 0.0199, 0.0210, 0.0211],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 17:57:04,532 INFO [train.py:892] (3/4) Epoch 24, batch 1250, loss[loss=0.1588, simple_loss=0.2328, pruned_loss=0.04239, over 19758.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2502, pruned_loss=0.05096, over 3944598.69 frames. ], batch size: 88, lr: 6.52e-03, grad_scale: 8.0
+2023-03-28 17:57:16,430 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.338e+02 3.868e+02 4.714e+02 5.719e+02 9.399e+02, threshold=9.429e+02, percent-clipped=0.0
+2023-03-28 17:58:45,993 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43959.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 17:58:51,840 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1979, 4.0853, 4.4843, 4.1503, 3.8240, 4.3392, 4.1634, 4.5600],
+       device='cuda:3'), covar=tensor([0.0840, 0.0358, 0.0351, 0.0354, 0.1042, 0.0490, 0.0440, 0.0344],
+       device='cuda:3'), in_proj_covar=tensor([0.0273, 0.0214, 0.0213, 0.0225, 0.0201, 0.0227, 0.0222, 0.0207],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 17:59:01,816 INFO [train.py:892] (3/4) Epoch 24, batch 1300, loss[loss=0.1522, simple_loss=0.2343, pruned_loss=0.03506, over 19836.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2485, pruned_loss=0.04998, over 3946596.33 frames. ], batch size: 76, lr: 6.52e-03, grad_scale: 8.0
+2023-03-28 17:59:13,372 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:00:59,570 INFO [train.py:892] (3/4) Epoch 24, batch 1350, loss[loss=0.1732, simple_loss=0.253, pruned_loss=0.04663, over 19795.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2492, pruned_loss=0.04994, over 3948334.56 frames. ], batch size: 51, lr: 6.52e-03, grad_scale: 8.0
+2023-03-28 18:01:08,908 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:01:09,758 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.475e+02 3.867e+02 4.545e+02 5.331e+02 8.729e+02, threshold=9.091e+02, percent-clipped=0.0
+2023-03-28 18:02:45,805 INFO [train.py:892] (3/4) Epoch 24, batch 1400, loss[loss=0.1835, simple_loss=0.2583, pruned_loss=0.05437, over 19682.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2498, pruned_loss=0.05006, over 3949307.49 frames. ], batch size: 52, lr: 6.51e-03, grad_scale: 8.0
+2023-03-28 18:04:35,919 INFO [train.py:892] (3/4) Epoch 24, batch 1450, loss[loss=0.1496, simple_loss=0.2206, pruned_loss=0.03927, over 19863.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2512, pruned_loss=0.0506, over 3948440.13 frames. ], batch size: 46, lr: 6.51e-03, grad_scale: 8.0
+2023-03-28 18:04:46,672 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.585e+02 3.832e+02 4.632e+02 5.399e+02 1.168e+03, threshold=9.265e+02, percent-clipped=4.0
+2023-03-28 18:05:40,674 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44143.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:06:31,972 INFO [train.py:892] (3/4) Epoch 24, batch 1500, loss[loss=0.1535, simple_loss=0.2299, pruned_loss=0.03854, over 19777.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2518, pruned_loss=0.05116, over 3948120.16 frames. ], batch size: 108, lr: 6.50e-03, grad_scale: 8.0
+2023-03-28 18:06:41,438 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:06:49,203 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0976, 4.1762, 2.5610, 4.4537, 4.5878, 2.0237, 3.7999, 3.3105],
+       device='cuda:3'), covar=tensor([0.0662, 0.0896, 0.2591, 0.0717, 0.0613, 0.2821, 0.1113, 0.0879],
+       device='cuda:3'), in_proj_covar=tensor([0.0225, 0.0250, 0.0227, 0.0262, 0.0239, 0.0200, 0.0234, 0.0188],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 18:07:36,379 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8019, 3.1806, 3.6539, 3.3528, 3.9357, 3.8643, 4.5749, 5.0602],
+       device='cuda:3'), covar=tensor([0.0422, 0.1653, 0.1377, 0.1999, 0.1556, 0.1345, 0.0541, 0.0452],
+       device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0238, 0.0261, 0.0250, 0.0289, 0.0252, 0.0223, 0.0245],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 18:07:54,235 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.9919, 6.2840, 6.2554, 6.1448, 5.9379, 6.2698, 5.5845, 5.5933],
+       device='cuda:3'), covar=tensor([0.0342, 0.0373, 0.0489, 0.0391, 0.0521, 0.0470, 0.0613, 0.0903],
+       device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0260, 0.0280, 0.0241, 0.0245, 0.0235, 0.0251, 0.0297],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 18:08:30,502 INFO [train.py:892] (3/4) Epoch 24, batch 1550, loss[loss=0.1705, simple_loss=0.2429, pruned_loss=0.04898, over 19839.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.251, pruned_loss=0.05067, over 3948701.57 frames. ], batch size: 208, lr: 6.50e-03, grad_scale: 8.0
+2023-03-28 18:08:35,925 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44218.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:08:41,921 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.584e+02 4.103e+02 5.047e+02 5.818e+02 1.108e+03, threshold=1.009e+03, percent-clipped=1.0
+2023-03-28 18:10:29,375 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:10:30,465 INFO [train.py:892] (3/4) Epoch 24, batch 1600, loss[loss=0.1992, simple_loss=0.2821, pruned_loss=0.05813, over 19735.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2518, pruned_loss=0.05108, over 3947608.63 frames. ], batch size: 77, lr: 6.50e-03, grad_scale: 8.0
+2023-03-28 18:10:47,890 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0
+2023-03-28 18:11:32,112 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.9647, 2.1773, 2.1764, 1.7925, 2.2172, 1.9410, 2.1346, 2.2253],
+       device='cuda:3'), covar=tensor([0.0488, 0.0481, 0.0483, 0.0947, 0.0415, 0.0471, 0.0453, 0.0328],
+       device='cuda:3'), in_proj_covar=tensor([0.0073, 0.0080, 0.0078, 0.0107, 0.0074, 0.0076, 0.0073, 0.0066],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 18:12:24,997 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:12:26,304 INFO [train.py:892] (3/4) Epoch 24, batch 1650, loss[loss=0.1798, simple_loss=0.2633, pruned_loss=0.0482, over 19602.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2528, pruned_loss=0.05164, over 3946878.88 frames. ], batch size: 50, lr: 6.49e-03, grad_scale: 8.0
+2023-03-28 18:12:36,791 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.188e+02 4.041e+02 4.652e+02 5.558e+02 9.682e+02, threshold=9.304e+02, percent-clipped=0.0
+2023-03-28 18:13:44,374 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8630, 2.6673, 4.6065, 3.9060, 4.4527, 4.5868, 4.3719, 4.3400],
+       device='cuda:3'), covar=tensor([0.0402, 0.0975, 0.0093, 0.0812, 0.0124, 0.0167, 0.0157, 0.0131],
+       device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0100, 0.0082, 0.0150, 0.0080, 0.0094, 0.0087, 0.0082],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0002, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 18:14:21,543 INFO [train.py:892] (3/4) Epoch 24, batch 1700, loss[loss=0.1609, simple_loss=0.2347, pruned_loss=0.04359, over 19742.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2521, pruned_loss=0.05121, over 3948694.04 frames. ], batch size: 44, lr: 6.49e-03, grad_scale: 8.0
+2023-03-28 18:16:13,225 INFO [train.py:892] (3/4) Epoch 24, batch 1750, loss[loss=0.1492, simple_loss=0.228, pruned_loss=0.03519, over 19735.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.2518, pruned_loss=0.05122, over 3948068.09 frames. ], batch size: 47, lr: 6.49e-03, grad_scale: 8.0
+2023-03-28 18:16:22,192 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.732e+02 4.086e+02 4.781e+02 5.963e+02 1.155e+03, threshold=9.562e+02, percent-clipped=1.0
+2023-03-28 18:16:42,425 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7942, 2.8982, 4.1733, 3.1810, 3.5241, 3.3406, 2.2746, 2.5209],
+       device='cuda:3'), covar=tensor([0.1032, 0.2693, 0.0497, 0.0960, 0.1511, 0.1359, 0.2410, 0.2442],
+       device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0374, 0.0335, 0.0270, 0.0365, 0.0356, 0.0355, 0.0324],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 18:17:12,434 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:17:22,409 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7254, 3.0517, 2.6894, 2.2242, 2.7040, 3.0003, 2.8823, 2.9401],
+       device='cuda:3'), covar=tensor([0.0314, 0.0260, 0.0283, 0.0531, 0.0344, 0.0240, 0.0214, 0.0237],
+       device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0087, 0.0092, 0.0095, 0.0098, 0.0077, 0.0075, 0.0077],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 18:17:56,857 INFO [train.py:892] (3/4) Epoch 24, batch 1800, loss[loss=0.1722, simple_loss=0.2384, pruned_loss=0.05299, over 19875.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.2526, pruned_loss=0.05187, over 3945503.43 frames. ], batch size: 125, lr: 6.48e-03, grad_scale: 8.0
+2023-03-28 18:18:42,771 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44491.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:19:13,695 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0166, 3.3385, 2.8872, 2.4706, 2.9339, 3.3086, 3.1646, 3.2302],
+       device='cuda:3'), covar=tensor([0.0277, 0.0247, 0.0252, 0.0506, 0.0307, 0.0226, 0.0207, 0.0183],
+       device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0087, 0.0093, 0.0096, 0.0098, 0.0077, 0.0076, 0.0077],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 18:19:30,405 INFO [train.py:892] (3/4) Epoch 24, batch 1850, loss[loss=0.1703, simple_loss=0.2493, pruned_loss=0.0457, over 19672.00 frames. ], tot_loss[loss=0.1788, simple_loss=0.2542, pruned_loss=0.05173, over 3944596.79 frames. ], batch size: 55, lr: 6.48e-03, grad_scale: 8.0
+2023-03-28 18:20:24,349 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.697e+02 4.242e+02 4.967e+02 6.010e+02 1.010e+03, threshold=9.934e+02, percent-clipped=1.0
+2023-03-28 18:20:24,380 INFO [train.py:892] (3/4) Epoch 25, batch 0, loss[loss=0.1585, simple_loss=0.2298, pruned_loss=0.04359, over 19835.00 frames. ], tot_loss[loss=0.1585, simple_loss=0.2298, pruned_loss=0.04359, over 19835.00 frames. ], batch size: 161, lr: 6.35e-03, grad_scale: 8.0
+2023-03-28 18:20:24,380 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-28 18:20:47,980 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5553, 4.0186, 3.8772, 3.9080, 3.9415, 3.9075, 3.8267, 3.5843],
+       device='cuda:3'), covar=tensor([0.2184, 0.1246, 0.1593, 0.1311, 0.1237, 0.0907, 0.1747, 0.2412],
+       device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0310, 0.0354, 0.0283, 0.0264, 0.0261, 0.0340, 0.0370],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 18:20:53,334 INFO [train.py:926] (3/4) Epoch 25, validation: loss=0.1751, simple_loss=0.2485, pruned_loss=0.05079, over 2883724.00 frames. 
+2023-03-28 18:20:53,335 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-28 18:21:07,816 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44527.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:21:57,662 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:22:06,455 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.49 vs. limit=5.0
+2023-03-28 18:22:33,042 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:22:45,069 INFO [train.py:892] (3/4) Epoch 25, batch 50, loss[loss=0.1862, simple_loss=0.2567, pruned_loss=0.05787, over 19754.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2421, pruned_loss=0.04636, over 889940.65 frames. ], batch size: 213, lr: 6.34e-03, grad_scale: 8.0
+2023-03-28 18:23:23,793 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:23:57,112 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7001, 4.7832, 5.0806, 4.8549, 4.9787, 4.5614, 4.7283, 4.6477],
+       device='cuda:3'), covar=tensor([0.1424, 0.1555, 0.0943, 0.1212, 0.0759, 0.1029, 0.2047, 0.2009],
+       device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0311, 0.0356, 0.0284, 0.0266, 0.0263, 0.0340, 0.0371],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 18:24:19,422 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44611.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:24:24,368 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:24:29,426 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44615.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:24:45,064 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.512e+02 3.985e+02 4.536e+02 5.355e+02 8.886e+02, threshold=9.072e+02, percent-clipped=0.0
+2023-03-28 18:24:45,090 INFO [train.py:892] (3/4) Epoch 25, batch 100, loss[loss=0.2056, simple_loss=0.2801, pruned_loss=0.06558, over 19560.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2446, pruned_loss=0.0476, over 1570006.50 frames. ], batch size: 41, lr: 6.34e-03, grad_scale: 8.0
+2023-03-28 18:26:30,247 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:26:47,957 INFO [train.py:892] (3/4) Epoch 25, batch 150, loss[loss=0.1538, simple_loss=0.2337, pruned_loss=0.03694, over 19852.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2462, pruned_loss=0.0475, over 2097916.67 frames. ], batch size: 112, lr: 6.33e-03, grad_scale: 8.0
+2023-03-28 18:26:51,049 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5320, 3.5251, 3.8915, 3.5347, 3.3957, 3.8343, 3.5913, 3.9618],
+       device='cuda:3'), covar=tensor([0.1161, 0.0486, 0.0517, 0.0525, 0.1373, 0.0625, 0.0621, 0.0485],
+       device='cuda:3'), in_proj_covar=tensor([0.0277, 0.0218, 0.0218, 0.0228, 0.0205, 0.0231, 0.0227, 0.0209],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 18:28:52,095 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.576e+02 4.007e+02 4.679e+02 6.121e+02 1.422e+03, threshold=9.359e+02, percent-clipped=5.0
+2023-03-28 18:28:52,205 INFO [train.py:892] (3/4) Epoch 25, batch 200, loss[loss=0.1548, simple_loss=0.2216, pruned_loss=0.04397, over 19819.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2477, pruned_loss=0.04823, over 2508968.98 frames. ], batch size: 121, lr: 6.33e-03, grad_scale: 8.0
+2023-03-28 18:29:36,132 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.1817, 2.3131, 2.3513, 2.2658, 2.2790, 2.4035, 2.3440, 2.3501],
+       device='cuda:3'), covar=tensor([0.0402, 0.0311, 0.0332, 0.0309, 0.0406, 0.0315, 0.0387, 0.0330],
+       device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0071, 0.0073, 0.0068, 0.0081, 0.0074, 0.0091, 0.0066],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 18:30:54,062 INFO [train.py:892] (3/4) Epoch 25, batch 250, loss[loss=0.1736, simple_loss=0.2483, pruned_loss=0.04942, over 19811.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2467, pruned_loss=0.04766, over 2829761.84 frames. ], batch size: 167, lr: 6.33e-03, grad_scale: 8.0
+2023-03-28 18:31:09,369 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0191, 4.6528, 4.6711, 4.9634, 4.5022, 5.1339, 5.1069, 5.2861],
+       device='cuda:3'), covar=tensor([0.0586, 0.0334, 0.0382, 0.0280, 0.0690, 0.0335, 0.0338, 0.0239],
+       device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0170, 0.0193, 0.0168, 0.0168, 0.0149, 0.0146, 0.0189],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 18:31:40,998 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5851, 2.0515, 2.3077, 2.7699, 3.0583, 3.1717, 3.1252, 3.2140],
+       device='cuda:3'), covar=tensor([0.0928, 0.1690, 0.1347, 0.0754, 0.0501, 0.0385, 0.0444, 0.0438],
+       device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0167, 0.0172, 0.0146, 0.0128, 0.0125, 0.0118, 0.0110],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 18:32:57,530 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.592e+02 3.844e+02 4.514e+02 5.418e+02 9.178e+02, threshold=9.028e+02, percent-clipped=0.0
+2023-03-28 18:32:57,584 INFO [train.py:892] (3/4) Epoch 25, batch 300, loss[loss=0.2235, simple_loss=0.2832, pruned_loss=0.08188, over 19738.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2473, pruned_loss=0.04813, over 3078425.18 frames. ], batch size: 269, lr: 6.32e-03, grad_scale: 8.0
+2023-03-28 18:33:48,341 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4610, 4.1023, 4.2126, 4.4684, 4.0910, 4.5307, 4.5957, 4.7453],
+       device='cuda:3'), covar=tensor([0.0652, 0.0421, 0.0518, 0.0361, 0.0809, 0.0470, 0.0426, 0.0305],
+       device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0171, 0.0194, 0.0169, 0.0169, 0.0150, 0.0146, 0.0191],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 18:35:01,721 INFO [train.py:892] (3/4) Epoch 25, batch 350, loss[loss=0.1951, simple_loss=0.2677, pruned_loss=0.06125, over 19705.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2489, pruned_loss=0.04855, over 3270598.38 frames. ], batch size: 54, lr: 6.32e-03, grad_scale: 8.0
+2023-03-28 18:35:28,994 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44883.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:36:26,536 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:36:54,288 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7866, 3.8333, 4.1674, 3.9645, 4.0849, 3.7351, 3.8790, 3.7189],
+       device='cuda:3'), covar=tensor([0.1695, 0.1635, 0.0978, 0.1370, 0.1228, 0.1094, 0.1898, 0.2346],
+       device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0309, 0.0354, 0.0285, 0.0264, 0.0262, 0.0339, 0.0370],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 18:36:58,895 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0
+2023-03-28 18:37:01,662 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.886e+02 4.144e+02 4.807e+02 5.961e+02 1.159e+03, threshold=9.615e+02, percent-clipped=3.0
+2023-03-28 18:37:01,726 INFO [train.py:892] (3/4) Epoch 25, batch 400, loss[loss=0.2106, simple_loss=0.2862, pruned_loss=0.06744, over 19708.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2509, pruned_loss=0.04996, over 3419962.73 frames. ], batch size: 315, lr: 6.32e-03, grad_scale: 8.0
+2023-03-28 18:38:49,499 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5115, 2.5764, 2.6852, 2.6560, 2.5456, 2.8126, 2.4638, 2.6611],
+       device='cuda:3'), covar=tensor([0.0290, 0.0321, 0.0288, 0.0244, 0.0353, 0.0236, 0.0415, 0.0388],
+       device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0071, 0.0073, 0.0067, 0.0081, 0.0074, 0.0092, 0.0066],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 18:39:03,927 INFO [train.py:892] (3/4) Epoch 25, batch 450, loss[loss=0.1702, simple_loss=0.2456, pruned_loss=0.04738, over 19893.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2505, pruned_loss=0.05001, over 3538270.57 frames. ], batch size: 63, lr: 6.31e-03, grad_scale: 8.0
+2023-03-28 18:40:59,618 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.643e+02 4.015e+02 4.807e+02 5.728e+02 9.598e+02, threshold=9.613e+02, percent-clipped=0.0
+2023-03-28 18:40:59,644 INFO [train.py:892] (3/4) Epoch 25, batch 500, loss[loss=0.1896, simple_loss=0.2586, pruned_loss=0.06034, over 19780.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2499, pruned_loss=0.04988, over 3631022.36 frames. ], batch size: 191, lr: 6.31e-03, grad_scale: 8.0
+2023-03-28 18:41:11,550 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4535, 4.2167, 4.2820, 4.0274, 4.4281, 3.1239, 3.6852, 2.0859],
+       device='cuda:3'), covar=tensor([0.0207, 0.0221, 0.0138, 0.0176, 0.0138, 0.0937, 0.0725, 0.1552],
+       device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0144, 0.0111, 0.0132, 0.0117, 0.0132, 0.0142, 0.0125],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 18:42:20,400 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:42:52,495 INFO [train.py:892] (3/4) Epoch 25, batch 550, loss[loss=0.1654, simple_loss=0.2338, pruned_loss=0.04853, over 19837.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2494, pruned_loss=0.04985, over 3702563.60 frames. ], batch size: 208, lr: 6.31e-03, grad_scale: 8.0
+2023-03-28 18:43:07,344 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2429, 3.1443, 3.3802, 2.8030, 3.5291, 2.9646, 3.1411, 3.4017],
+       device='cuda:3'), covar=tensor([0.0742, 0.0394, 0.0831, 0.0703, 0.0338, 0.0437, 0.0561, 0.0428],
+       device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0079, 0.0078, 0.0106, 0.0074, 0.0075, 0.0073, 0.0066],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 18:43:21,313 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45081.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:44:44,850 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45117.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 18:44:51,745 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.460e+02 3.947e+02 4.547e+02 5.526e+02 8.636e+02, threshold=9.094e+02, percent-clipped=0.0
+2023-03-28 18:44:51,773 INFO [train.py:892] (3/4) Epoch 25, batch 600, loss[loss=0.1487, simple_loss=0.229, pruned_loss=0.03424, over 19753.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2485, pruned_loss=0.04924, over 3758705.08 frames. ], batch size: 84, lr: 6.30e-03, grad_scale: 8.0
+2023-03-28 18:45:04,123 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7239, 5.0413, 5.0704, 4.9431, 4.6809, 5.0401, 4.5689, 4.5675],
+       device='cuda:3'), covar=tensor([0.0456, 0.0445, 0.0499, 0.0478, 0.0623, 0.0545, 0.0659, 0.0978],
+       device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0263, 0.0283, 0.0244, 0.0248, 0.0237, 0.0253, 0.0299],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 18:45:04,227 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6699, 2.0270, 2.3820, 2.8470, 3.2306, 3.3576, 3.3224, 3.2948],
+       device='cuda:3'), covar=tensor([0.0963, 0.1770, 0.1406, 0.0763, 0.0501, 0.0342, 0.0441, 0.0473],
+       device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0170, 0.0175, 0.0148, 0.0129, 0.0126, 0.0119, 0.0113],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 18:45:41,173 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.4967, 1.3265, 1.5374, 1.5019, 1.3496, 1.5321, 1.2712, 1.4710],
+       device='cuda:3'), covar=tensor([0.0339, 0.0338, 0.0308, 0.0279, 0.0487, 0.0295, 0.0527, 0.0300],
+       device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0070, 0.0073, 0.0067, 0.0081, 0.0074, 0.0091, 0.0066],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 18:45:46,551 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45142.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:46:55,688 INFO [train.py:892] (3/4) Epoch 25, batch 650, loss[loss=0.1667, simple_loss=0.2319, pruned_loss=0.0508, over 19888.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2488, pruned_loss=0.04922, over 3800390.93 frames. ], batch size: 176, lr: 6.30e-03, grad_scale: 8.0
+2023-03-28 18:47:21,547 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4356, 2.1581, 3.2120, 2.7533, 3.3070, 3.4156, 3.1162, 3.2542],
+       device='cuda:3'), covar=tensor([0.0702, 0.1025, 0.0127, 0.0452, 0.0142, 0.0219, 0.0208, 0.0182],
+       device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0102, 0.0084, 0.0153, 0.0081, 0.0095, 0.0088, 0.0083],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0002, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 18:47:25,377 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:48:24,984 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:48:58,111 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.622e+02 4.071e+02 4.790e+02 5.596e+02 1.238e+03, threshold=9.579e+02, percent-clipped=2.0
+2023-03-28 18:48:58,137 INFO [train.py:892] (3/4) Epoch 25, batch 700, loss[loss=0.1647, simple_loss=0.2457, pruned_loss=0.04184, over 19793.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2488, pruned_loss=0.04907, over 3834182.03 frames. ], batch size: 45, lr: 6.30e-03, grad_scale: 8.0
+2023-03-28 18:49:21,634 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:50:14,564 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:50:54,527 INFO [train.py:892] (3/4) Epoch 25, batch 750, loss[loss=0.1607, simple_loss=0.2384, pruned_loss=0.04148, over 19812.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2477, pruned_loss=0.04879, over 3861004.35 frames. ], batch size: 96, lr: 6.29e-03, grad_scale: 8.0
+2023-03-28 18:51:38,468 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0
+2023-03-28 18:52:49,364 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.500e+02 3.772e+02 4.659e+02 5.559e+02 8.961e+02, threshold=9.318e+02, percent-clipped=0.0
+2023-03-28 18:52:49,387 INFO [train.py:892] (3/4) Epoch 25, batch 800, loss[loss=0.1538, simple_loss=0.229, pruned_loss=0.03935, over 19708.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2474, pruned_loss=0.04888, over 3881711.76 frames. ], batch size: 81, lr: 6.29e-03, grad_scale: 8.0
+2023-03-28 18:53:27,642 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0
+2023-03-28 18:53:54,190 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45348.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 18:54:43,365 INFO [train.py:892] (3/4) Epoch 25, batch 850, loss[loss=0.1654, simple_loss=0.2378, pruned_loss=0.04646, over 19759.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.247, pruned_loss=0.0482, over 3896395.46 frames. ], batch size: 205, lr: 6.29e-03, grad_scale: 8.0
+2023-03-28 18:56:15,208 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45409.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 18:56:22,375 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45412.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 18:56:43,426 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.654e+02 4.187e+02 5.025e+02 5.805e+02 1.375e+03, threshold=1.005e+03, percent-clipped=1.0
+2023-03-28 18:56:43,477 INFO [train.py:892] (3/4) Epoch 25, batch 900, loss[loss=0.1644, simple_loss=0.2359, pruned_loss=0.04642, over 19765.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2468, pruned_loss=0.04795, over 3908613.85 frames. ], batch size: 217, lr: 6.28e-03, grad_scale: 8.0
+2023-03-28 18:57:24,634 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 18:58:28,587 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.88 vs. limit=5.0
+2023-03-28 18:58:45,923 INFO [train.py:892] (3/4) Epoch 25, batch 950, loss[loss=0.1609, simple_loss=0.2349, pruned_loss=0.04345, over 19772.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2474, pruned_loss=0.04819, over 3917629.18 frames. ], batch size: 113, lr: 6.28e-03, grad_scale: 8.0
+2023-03-28 18:59:05,336 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6886, 4.8079, 5.0738, 4.9273, 4.9820, 4.5238, 4.8346, 4.6721],
+       device='cuda:3'), covar=tensor([0.1514, 0.1588, 0.0984, 0.1244, 0.0814, 0.1089, 0.1971, 0.2149],
+       device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0311, 0.0353, 0.0286, 0.0264, 0.0264, 0.0338, 0.0370],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 19:00:40,925 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.498e+02 3.760e+02 4.893e+02 5.936e+02 1.409e+03, threshold=9.787e+02, percent-clipped=1.0
+2023-03-28 19:00:40,959 INFO [train.py:892] (3/4) Epoch 25, batch 1000, loss[loss=0.1514, simple_loss=0.2281, pruned_loss=0.03731, over 19859.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2478, pruned_loss=0.04859, over 3925323.11 frames. ], batch size: 104, lr: 6.28e-03, grad_scale: 8.0
+2023-03-28 19:01:55,571 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6408, 4.5479, 5.0266, 4.6130, 4.2198, 4.8673, 4.6926, 5.1703],
+       device='cuda:3'), covar=tensor([0.0826, 0.0369, 0.0359, 0.0354, 0.0756, 0.0419, 0.0383, 0.0295],
+       device='cuda:3'), in_proj_covar=tensor([0.0277, 0.0218, 0.0218, 0.0228, 0.0204, 0.0230, 0.0227, 0.0209],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 19:02:32,343 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4610, 2.6513, 2.7031, 2.5436, 2.5322, 2.6558, 2.4915, 2.6941],
+       device='cuda:3'), covar=tensor([0.0298, 0.0298, 0.0248, 0.0266, 0.0401, 0.0282, 0.0388, 0.0302],
+       device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0071, 0.0074, 0.0068, 0.0081, 0.0074, 0.0091, 0.0066],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 19:02:41,462 INFO [train.py:892] (3/4) Epoch 25, batch 1050, loss[loss=0.1882, simple_loss=0.2757, pruned_loss=0.05035, over 19669.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2487, pruned_loss=0.04874, over 3931137.58 frames. ], batch size: 57, lr: 6.27e-03, grad_scale: 8.0
+2023-03-28 19:03:11,458 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2904, 3.4136, 2.0989, 3.4916, 3.5848, 1.7269, 3.0137, 2.7553],
+       device='cuda:3'), covar=tensor([0.0902, 0.0957, 0.2815, 0.0904, 0.0727, 0.2657, 0.1128, 0.0924],
+       device='cuda:3'), in_proj_covar=tensor([0.0227, 0.0251, 0.0228, 0.0265, 0.0243, 0.0200, 0.0233, 0.0191],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 19:03:57,493 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-28 19:04:40,794 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.489e+02 3.919e+02 4.482e+02 5.377e+02 1.345e+03, threshold=8.964e+02, percent-clipped=5.0
+2023-03-28 19:04:40,821 INFO [train.py:892] (3/4) Epoch 25, batch 1100, loss[loss=0.1665, simple_loss=0.2408, pruned_loss=0.04614, over 19837.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2491, pruned_loss=0.0488, over 3935893.02 frames. ], batch size: 128, lr: 6.27e-03, grad_scale: 8.0
+2023-03-28 19:06:36,327 INFO [train.py:892] (3/4) Epoch 25, batch 1150, loss[loss=0.1458, simple_loss=0.2208, pruned_loss=0.03537, over 19743.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.248, pruned_loss=0.04855, over 3940149.17 frames. ], batch size: 110, lr: 6.27e-03, grad_scale: 8.0
+2023-03-28 19:07:52,552 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:07:58,020 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45704.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 19:08:19,599 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45712.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 19:08:39,148 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.049e+02 4.389e+02 4.900e+02 5.957e+02 9.190e+02, threshold=9.801e+02, percent-clipped=1.0
+2023-03-28 19:08:39,205 INFO [train.py:892] (3/4) Epoch 25, batch 1200, loss[loss=0.1677, simple_loss=0.2511, pruned_loss=0.04213, over 19781.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.249, pruned_loss=0.04925, over 3943139.56 frames. ], batch size: 94, lr: 6.26e-03, grad_scale: 8.0
+2023-03-28 19:09:16,242 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45737.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:10:08,299 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:10:11,924 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.4668, 1.3650, 1.5317, 1.4560, 1.3836, 1.5230, 1.3221, 1.5138],
+       device='cuda:3'), covar=tensor([0.0314, 0.0321, 0.0310, 0.0284, 0.0465, 0.0300, 0.0508, 0.0303],
+       device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0072, 0.0074, 0.0069, 0.0082, 0.0076, 0.0093, 0.0066],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 19:10:13,999 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:10:31,139 INFO [train.py:892] (3/4) Epoch 25, batch 1250, loss[loss=0.1905, simple_loss=0.2577, pruned_loss=0.06163, over 19781.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2499, pruned_loss=0.04969, over 3944133.95 frames. ], batch size: 193, lr: 6.26e-03, grad_scale: 8.0
+2023-03-28 19:10:38,169 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45774.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:11:03,342 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:12:12,174 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6587, 3.7787, 2.3227, 3.9309, 4.0063, 1.8693, 3.2932, 3.2193],
+       device='cuda:3'), covar=tensor([0.0768, 0.0839, 0.2675, 0.0776, 0.0632, 0.2678, 0.1105, 0.0764],
+       device='cuda:3'), in_proj_covar=tensor([0.0226, 0.0250, 0.0227, 0.0263, 0.0241, 0.0200, 0.0233, 0.0190],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 19:12:24,907 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.740e+02 4.060e+02 4.817e+02 5.850e+02 1.056e+03, threshold=9.634e+02, percent-clipped=3.0
+2023-03-28 19:12:24,933 INFO [train.py:892] (3/4) Epoch 25, batch 1300, loss[loss=0.1636, simple_loss=0.2326, pruned_loss=0.04733, over 19768.00 frames. ], tot_loss[loss=0.1748, simple_loss=0.2498, pruned_loss=0.0499, over 3946141.19 frames. ], batch size: 233, lr: 6.26e-03, grad_scale: 8.0
+2023-03-28 19:13:01,693 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45835.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:14:00,830 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45859.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:14:08,824 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0
+2023-03-28 19:14:29,560 INFO [train.py:892] (3/4) Epoch 25, batch 1350, loss[loss=0.2638, simple_loss=0.3276, pruned_loss=0.1001, over 19465.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2502, pruned_loss=0.04999, over 3945321.71 frames. ], batch size: 396, lr: 6.25e-03, grad_scale: 8.0
+2023-03-28 19:14:43,596 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7457, 2.6572, 2.9125, 2.6603, 3.0908, 3.0280, 3.5898, 3.8900],
+       device='cuda:3'), covar=tensor([0.0613, 0.1723, 0.1564, 0.2117, 0.1667, 0.1468, 0.0651, 0.0652],
+       device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0238, 0.0261, 0.0250, 0.0290, 0.0252, 0.0224, 0.0247],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 19:16:24,423 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45920.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:16:25,389 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.758e+02 4.142e+02 4.668e+02 5.534e+02 8.223e+02, threshold=9.336e+02, percent-clipped=0.0
+2023-03-28 19:16:25,414 INFO [train.py:892] (3/4) Epoch 25, batch 1400, loss[loss=0.2033, simple_loss=0.2876, pruned_loss=0.05953, over 19665.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2501, pruned_loss=0.04951, over 3946537.08 frames. ], batch size: 55, lr: 6.25e-03, grad_scale: 16.0
+2023-03-28 19:18:21,783 INFO [train.py:892] (3/4) Epoch 25, batch 1450, loss[loss=0.196, simple_loss=0.2766, pruned_loss=0.05769, over 19826.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2483, pruned_loss=0.04861, over 3948399.24 frames. ], batch size: 204, lr: 6.25e-03, grad_scale: 16.0
+2023-03-28 19:19:53,715 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46004.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 19:20:30,242 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.575e+02 3.735e+02 4.746e+02 5.527e+02 8.310e+02, threshold=9.492e+02, percent-clipped=0.0
+2023-03-28 19:20:30,330 INFO [train.py:892] (3/4) Epoch 25, batch 1500, loss[loss=0.1901, simple_loss=0.2718, pruned_loss=0.05422, over 19840.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2492, pruned_loss=0.04899, over 3948019.53 frames. ], batch size: 160, lr: 6.24e-03, grad_scale: 16.0
+2023-03-28 19:20:59,294 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9327, 3.9392, 2.3292, 4.2170, 4.4307, 2.0228, 3.5730, 3.4366],
+       device='cuda:3'), covar=tensor([0.0722, 0.1054, 0.2995, 0.0954, 0.0571, 0.2929, 0.1168, 0.0895],
+       device='cuda:3'), in_proj_covar=tensor([0.0227, 0.0251, 0.0228, 0.0265, 0.0244, 0.0202, 0.0236, 0.0192],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 19:21:06,262 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.56 vs. limit=5.0
+2023-03-28 19:21:42,019 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0
+2023-03-28 19:21:43,811 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46052.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 19:21:58,006 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46058.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:22:29,055 INFO [train.py:892] (3/4) Epoch 25, batch 1550, loss[loss=0.1745, simple_loss=0.2522, pruned_loss=0.04836, over 19611.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2495, pruned_loss=0.04903, over 3947633.11 frames. ], batch size: 46, lr: 6.24e-03, grad_scale: 16.0
+2023-03-28 19:24:26,938 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.641e+02 4.209e+02 5.084e+02 6.041e+02 9.841e+02, threshold=1.017e+03, percent-clipped=2.0
+2023-03-28 19:24:26,964 INFO [train.py:892] (3/4) Epoch 25, batch 1600, loss[loss=0.157, simple_loss=0.2365, pruned_loss=0.03872, over 19614.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2492, pruned_loss=0.0488, over 3949010.48 frames. ], batch size: 51, lr: 6.24e-03, grad_scale: 16.0
+2023-03-28 19:24:49,231 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:24:49,441 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1327, 2.9839, 3.4630, 2.6470, 3.5656, 2.9104, 3.1665, 3.5119],
+       device='cuda:3'), covar=tensor([0.0649, 0.0462, 0.0440, 0.0731, 0.0325, 0.0486, 0.0474, 0.0263],
+       device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0080, 0.0077, 0.0106, 0.0073, 0.0076, 0.0073, 0.0066],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 19:26:20,498 INFO [train.py:892] (3/4) Epoch 25, batch 1650, loss[loss=0.1446, simple_loss=0.2277, pruned_loss=0.03074, over 19781.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.25, pruned_loss=0.04933, over 3947467.35 frames. ], batch size: 91, lr: 6.23e-03, grad_scale: 16.0
+2023-03-28 19:28:09,034 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46215.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:28:20,990 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.501e+02 4.063e+02 5.043e+02 6.372e+02 1.431e+03, threshold=1.009e+03, percent-clipped=3.0
+2023-03-28 19:28:21,024 INFO [train.py:892] (3/4) Epoch 25, batch 1700, loss[loss=0.1554, simple_loss=0.2305, pruned_loss=0.04012, over 19652.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2495, pruned_loss=0.0491, over 3947090.66 frames. ], batch size: 67, lr: 6.23e-03, grad_scale: 16.0
+2023-03-28 19:29:25,817 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.57 vs. limit=5.0
+2023-03-28 19:30:18,594 INFO [train.py:892] (3/4) Epoch 25, batch 1750, loss[loss=0.1349, simple_loss=0.2057, pruned_loss=0.03207, over 19857.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2489, pruned_loss=0.04882, over 3948489.92 frames. ], batch size: 118, lr: 6.23e-03, grad_scale: 16.0
+2023-03-28 19:30:29,021 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46276.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:30:46,789 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5460, 2.6509, 2.7687, 2.2584, 2.9395, 2.4402, 2.7887, 2.8993],
+       device='cuda:3'), covar=tensor([0.0694, 0.0514, 0.0569, 0.0788, 0.0373, 0.0455, 0.0461, 0.0288],
+       device='cuda:3'), in_proj_covar=tensor([0.0072, 0.0079, 0.0077, 0.0106, 0.0073, 0.0076, 0.0073, 0.0066],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 19:30:56,035 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0060, 3.3792, 3.5320, 4.0271, 2.8015, 3.2924, 2.6217, 2.6395],
+       device='cuda:3'), covar=tensor([0.0542, 0.1797, 0.0879, 0.0391, 0.1862, 0.0812, 0.1250, 0.1562],
+       device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0334, 0.0243, 0.0196, 0.0244, 0.0205, 0.0214, 0.0215],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 19:31:48,494 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.51 vs. limit=5.0
+2023-03-28 19:32:03,796 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.718e+02 4.184e+02 4.852e+02 5.831e+02 2.262e+03, threshold=9.705e+02, percent-clipped=1.0
+2023-03-28 19:32:03,822 INFO [train.py:892] (3/4) Epoch 25, batch 1800, loss[loss=0.1693, simple_loss=0.2385, pruned_loss=0.05009, over 19853.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2513, pruned_loss=0.05036, over 3947294.86 frames. ], batch size: 197, lr: 6.22e-03, grad_scale: 16.0
+2023-03-28 19:32:32,308 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:32:40,772 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.7012, 5.9880, 6.0180, 5.8667, 5.7690, 6.0413, 5.3047, 5.4122],
+       device='cuda:3'), covar=tensor([0.0452, 0.0473, 0.0459, 0.0414, 0.0548, 0.0427, 0.0621, 0.0912],
+       device='cuda:3'), in_proj_covar=tensor([0.0256, 0.0268, 0.0286, 0.0249, 0.0251, 0.0238, 0.0256, 0.0303],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 19:32:57,095 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5733, 3.7745, 4.1164, 4.7725, 3.1845, 3.3182, 3.0166, 2.8483],
+       device='cuda:3'), covar=tensor([0.0470, 0.2229, 0.0817, 0.0334, 0.1965, 0.1037, 0.1212, 0.1746],
+       device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0335, 0.0244, 0.0196, 0.0244, 0.0205, 0.0214, 0.0215],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 19:33:15,137 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:33:38,585 INFO [train.py:892] (3/4) Epoch 25, batch 1850, loss[loss=0.1662, simple_loss=0.2463, pruned_loss=0.04301, over 19676.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.252, pruned_loss=0.05025, over 3946074.05 frames. ], batch size: 55, lr: 6.22e-03, grad_scale: 16.0
+2023-03-28 19:34:37,786 INFO [train.py:892] (3/4) Epoch 26, batch 0, loss[loss=0.1532, simple_loss=0.2298, pruned_loss=0.03825, over 19856.00 frames. ], tot_loss[loss=0.1532, simple_loss=0.2298, pruned_loss=0.03825, over 19856.00 frames. ], batch size: 99, lr: 6.10e-03, grad_scale: 16.0
+2023-03-28 19:34:37,787 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-28 19:34:56,873 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3707, 3.2529, 3.7193, 3.0592, 3.9960, 3.1936, 3.3099, 3.9028],
+       device='cuda:3'), covar=tensor([0.0879, 0.0420, 0.0640, 0.0702, 0.0290, 0.0427, 0.0538, 0.0298],
+       device='cuda:3'), in_proj_covar=tensor([0.0071, 0.0079, 0.0077, 0.0105, 0.0073, 0.0076, 0.0073, 0.0066],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 19:35:05,753 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0478, 3.1232, 3.1160, 3.0636, 2.9155, 3.0741, 2.9628, 3.2661],
+       device='cuda:3'), covar=tensor([0.0253, 0.0317, 0.0315, 0.0286, 0.0353, 0.0223, 0.0303, 0.0279],
+       device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0073, 0.0075, 0.0070, 0.0083, 0.0076, 0.0094, 0.0067],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 19:35:16,911 INFO [train.py:926] (3/4) Epoch 26, validation: loss=0.176, simple_loss=0.2485, pruned_loss=0.05179, over 2883724.00 frames. 
+2023-03-28 19:35:16,914 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-28 19:35:36,417 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0370, 3.9176, 3.9292, 3.6902, 4.0512, 3.0154, 3.3605, 2.0286],
+       device='cuda:3'), covar=tensor([0.0211, 0.0228, 0.0145, 0.0186, 0.0143, 0.0870, 0.0650, 0.1447],
+       device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0144, 0.0111, 0.0131, 0.0118, 0.0132, 0.0141, 0.0126],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 19:36:01,644 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6951, 2.8327, 3.5446, 3.0144, 3.7920, 3.9005, 4.5755, 5.0290],
+       device='cuda:3'), covar=tensor([0.0480, 0.2035, 0.1542, 0.2340, 0.1798, 0.1286, 0.0524, 0.0410],
+       device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0237, 0.0260, 0.0249, 0.0289, 0.0250, 0.0225, 0.0245],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 19:36:29,153 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:37:07,546 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.563e+02 3.816e+02 4.430e+02 5.059e+02 8.683e+02, threshold=8.861e+02, percent-clipped=0.0
+2023-03-28 19:37:18,538 INFO [train.py:892] (3/4) Epoch 26, batch 50, loss[loss=0.1656, simple_loss=0.2339, pruned_loss=0.04867, over 19853.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.247, pruned_loss=0.04964, over 891334.58 frames. ], batch size: 165, lr: 6.09e-03, grad_scale: 16.0
+2023-03-28 19:37:28,843 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46430.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:37:40,746 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4255, 3.1779, 3.1961, 3.3765, 3.2701, 3.2800, 3.5411, 3.6547],
+       device='cuda:3'), covar=tensor([0.0726, 0.0481, 0.0595, 0.0424, 0.0782, 0.0786, 0.0423, 0.0346],
+       device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0173, 0.0195, 0.0169, 0.0168, 0.0152, 0.0147, 0.0192],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 19:38:22,858 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7994, 3.9553, 2.2934, 4.1683, 4.2895, 1.8674, 3.4833, 3.2038],
+       device='cuda:3'), covar=tensor([0.0728, 0.0776, 0.2814, 0.0739, 0.0481, 0.2847, 0.1147, 0.0879],
+       device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0251, 0.0229, 0.0266, 0.0246, 0.0203, 0.0237, 0.0193],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 19:39:07,495 INFO [train.py:892] (3/4) Epoch 26, batch 100, loss[loss=0.1842, simple_loss=0.2727, pruned_loss=0.04778, over 19647.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2507, pruned_loss=0.05003, over 1568184.04 frames. ], batch size: 57, lr: 6.09e-03, grad_scale: 16.0
+2023-03-28 19:39:10,956 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0
+2023-03-28 19:39:12,164 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:39:19,805 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8162, 3.0553, 2.7250, 2.2919, 2.7270, 2.9619, 2.9020, 2.9950],
+       device='cuda:3'), covar=tensor([0.0325, 0.0309, 0.0296, 0.0589, 0.0362, 0.0365, 0.0240, 0.0246],
+       device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0090, 0.0094, 0.0098, 0.0101, 0.0079, 0.0079, 0.0080],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 19:40:31,586 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46515.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:40:42,635 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.820e+02 4.177e+02 4.877e+02 5.600e+02 1.186e+03, threshold=9.755e+02, percent-clipped=5.0
+2023-03-28 19:40:54,317 INFO [train.py:892] (3/4) Epoch 26, batch 150, loss[loss=0.1563, simple_loss=0.2341, pruned_loss=0.0392, over 19786.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2481, pruned_loss=0.04839, over 2097006.19 frames. ], batch size: 154, lr: 6.09e-03, grad_scale: 16.0
+2023-03-28 19:41:01,805 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8519, 3.1677, 3.0439, 3.0661, 2.9017, 3.0214, 2.8175, 3.2737],
+       device='cuda:3'), covar=tensor([0.0254, 0.0290, 0.0364, 0.0267, 0.0368, 0.0277, 0.0380, 0.0231],
+       device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0073, 0.0075, 0.0070, 0.0083, 0.0076, 0.0094, 0.0067],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 19:42:14,627 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:42:42,420 INFO [train.py:892] (3/4) Epoch 26, batch 200, loss[loss=0.1897, simple_loss=0.2644, pruned_loss=0.05751, over 19764.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.248, pruned_loss=0.04824, over 2508904.65 frames. ], batch size: 244, lr: 6.08e-03, grad_scale: 16.0
+2023-03-28 19:43:25,591 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:43:42,784 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:44:24,557 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.062e+02 3.750e+02 4.526e+02 5.438e+02 1.075e+03, threshold=9.053e+02, percent-clipped=3.0
+2023-03-28 19:44:31,797 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4694, 4.3041, 4.3609, 4.5441, 4.4692, 4.8954, 4.5954, 4.6849],
+       device='cuda:3'), covar=tensor([0.0939, 0.0538, 0.0670, 0.0486, 0.0757, 0.0427, 0.0663, 0.0650],
+       device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0174, 0.0196, 0.0170, 0.0169, 0.0152, 0.0147, 0.0193],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 19:44:34,766 INFO [train.py:892] (3/4) Epoch 26, batch 250, loss[loss=0.144, simple_loss=0.2209, pruned_loss=0.03358, over 19749.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2471, pruned_loss=0.04779, over 2827295.66 frames. ], batch size: 84, lr: 6.08e-03, grad_scale: 16.0
+2023-03-28 19:44:50,086 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:44:53,992 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.3999, 4.8796, 4.9781, 4.7515, 5.2912, 3.2631, 4.1799, 2.4897],
+       device='cuda:3'), covar=tensor([0.0153, 0.0207, 0.0148, 0.0185, 0.0138, 0.0977, 0.0998, 0.1568],
+       device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0142, 0.0111, 0.0130, 0.0117, 0.0132, 0.0140, 0.0125],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 19:45:47,145 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46656.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 19:46:01,409 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46663.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:46:01,557 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5019, 3.1317, 3.4559, 3.0408, 3.7106, 3.7145, 4.3502, 4.8078],
+       device='cuda:3'), covar=tensor([0.0487, 0.1610, 0.1452, 0.2158, 0.1590, 0.1337, 0.0575, 0.0552],
+       device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0237, 0.0261, 0.0251, 0.0290, 0.0251, 0.0225, 0.0245],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 19:46:16,888 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0516, 4.1097, 2.4337, 4.3766, 4.5653, 1.9603, 3.7237, 3.4082],
+       device='cuda:3'), covar=tensor([0.0608, 0.0785, 0.2745, 0.0689, 0.0403, 0.2807, 0.0969, 0.0802],
+       device='cuda:3'), in_proj_covar=tensor([0.0227, 0.0249, 0.0226, 0.0265, 0.0244, 0.0201, 0.0236, 0.0191],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 19:46:31,201 INFO [train.py:892] (3/4) Epoch 26, batch 300, loss[loss=0.1741, simple_loss=0.2505, pruned_loss=0.04884, over 19709.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2473, pruned_loss=0.04761, over 3076229.88 frames. ], batch size: 81, lr: 6.08e-03, grad_scale: 16.0
+2023-03-28 19:46:58,486 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.11 vs. limit=2.0
+2023-03-28 19:48:12,964 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.886e+02 3.778e+02 4.525e+02 5.535e+02 8.157e+02, threshold=9.049e+02, percent-clipped=0.0
+2023-03-28 19:48:23,048 INFO [train.py:892] (3/4) Epoch 26, batch 350, loss[loss=0.1575, simple_loss=0.2311, pruned_loss=0.04194, over 19858.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2476, pruned_loss=0.04758, over 3270975.39 frames. ], batch size: 85, lr: 6.07e-03, grad_scale: 16.0
+2023-03-28 19:50:21,550 INFO [train.py:892] (3/4) Epoch 26, batch 400, loss[loss=0.1501, simple_loss=0.2284, pruned_loss=0.03589, over 19775.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2474, pruned_loss=0.04753, over 3420807.71 frames. ], batch size: 198, lr: 6.07e-03, grad_scale: 16.0
+2023-03-28 19:50:45,594 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.8262, 6.0752, 6.0699, 5.9680, 5.8336, 6.1063, 5.3466, 5.4255],
+       device='cuda:3'), covar=tensor([0.0361, 0.0395, 0.0501, 0.0379, 0.0523, 0.0483, 0.0730, 0.1001],
+       device='cuda:3'), in_proj_covar=tensor([0.0256, 0.0267, 0.0288, 0.0247, 0.0252, 0.0239, 0.0257, 0.0303],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 19:51:41,973 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:51:43,996 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7790, 2.9107, 3.2328, 2.8842, 3.1325, 3.2638, 3.0603, 3.2934],
+       device='cuda:3'), covar=tensor([0.1318, 0.0481, 0.0554, 0.0569, 0.1060, 0.0638, 0.0524, 0.0532],
+       device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0222, 0.0221, 0.0231, 0.0206, 0.0234, 0.0230, 0.0212],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 19:52:09,141 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.488e+02 3.851e+02 4.931e+02 6.030e+02 1.390e+03, threshold=9.862e+02, percent-clipped=3.0
+2023-03-28 19:52:19,752 INFO [train.py:892] (3/4) Epoch 26, batch 450, loss[loss=0.179, simple_loss=0.2491, pruned_loss=0.05445, over 19648.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2484, pruned_loss=0.04805, over 3536550.17 frames. ], batch size: 47, lr: 6.07e-03, grad_scale: 16.0
+2023-03-28 19:52:56,613 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8949, 2.9386, 3.1349, 3.0255, 2.8962, 3.0992, 2.8978, 3.0888],
+       device='cuda:3'), covar=tensor([0.0283, 0.0371, 0.0285, 0.0293, 0.0439, 0.0309, 0.0377, 0.0388],
+       device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0072, 0.0074, 0.0069, 0.0082, 0.0076, 0.0093, 0.0067],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 19:54:06,202 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:54:16,501 INFO [train.py:892] (3/4) Epoch 26, batch 500, loss[loss=0.2019, simple_loss=0.2621, pruned_loss=0.07082, over 19830.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2477, pruned_loss=0.04809, over 3629032.77 frames. ], batch size: 190, lr: 6.06e-03, grad_scale: 16.0
+2023-03-28 19:54:21,964 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5415, 2.5839, 1.4257, 2.9548, 2.6547, 2.8201, 2.9513, 2.3545],
+       device='cuda:3'), covar=tensor([0.0696, 0.0729, 0.1795, 0.0594, 0.0697, 0.0522, 0.0583, 0.0908],
+       device='cuda:3'), in_proj_covar=tensor([0.0140, 0.0140, 0.0141, 0.0146, 0.0130, 0.0129, 0.0142, 0.0142],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 19:56:03,180 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.854e+02 4.348e+02 5.094e+02 6.034e+02 1.006e+03, threshold=1.019e+03, percent-clipped=1.0
+2023-03-28 19:56:14,637 INFO [train.py:892] (3/4) Epoch 26, batch 550, loss[loss=0.162, simple_loss=0.2354, pruned_loss=0.04429, over 19796.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2488, pruned_loss=0.04879, over 3699008.39 frames. ], batch size: 193, lr: 6.06e-03, grad_scale: 16.0
+2023-03-28 19:56:27,621 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:57:13,262 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46951.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 19:57:28,722 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46958.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:57:29,018 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5556, 3.7817, 4.0178, 4.7299, 3.1188, 3.2941, 2.9519, 2.9058],
+       device='cuda:3'), covar=tensor([0.0462, 0.2045, 0.0810, 0.0326, 0.1844, 0.1071, 0.1191, 0.1558],
+       device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0334, 0.0244, 0.0196, 0.0242, 0.0203, 0.0212, 0.0214],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 19:58:10,256 INFO [train.py:892] (3/4) Epoch 26, batch 600, loss[loss=0.1981, simple_loss=0.2666, pruned_loss=0.06485, over 19729.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2482, pruned_loss=0.04875, over 3755370.19 frames. ], batch size: 269, lr: 6.06e-03, grad_scale: 16.0
+2023-03-28 19:58:20,935 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 19:58:51,184 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.17 vs. limit=5.0
+2023-03-28 19:59:54,796 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.697e+02 3.904e+02 4.581e+02 5.783e+02 1.259e+03, threshold=9.163e+02, percent-clipped=1.0
+2023-03-28 20:00:06,800 INFO [train.py:892] (3/4) Epoch 26, batch 650, loss[loss=0.1648, simple_loss=0.2374, pruned_loss=0.04614, over 19882.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2489, pruned_loss=0.04903, over 3798169.52 frames. ], batch size: 134, lr: 6.05e-03, grad_scale: 16.0
+2023-03-28 20:02:09,492 INFO [train.py:892] (3/4) Epoch 26, batch 700, loss[loss=0.1773, simple_loss=0.2515, pruned_loss=0.05151, over 19846.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2487, pruned_loss=0.04848, over 3832888.01 frames. ], batch size: 197, lr: 6.05e-03, grad_scale: 16.0
+2023-03-28 20:03:57,430 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.011e+02 3.881e+02 4.556e+02 5.406e+02 9.543e+02, threshold=9.112e+02, percent-clipped=1.0
+2023-03-28 20:04:09,140 INFO [train.py:892] (3/4) Epoch 26, batch 750, loss[loss=0.1692, simple_loss=0.245, pruned_loss=0.04668, over 19790.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2483, pruned_loss=0.04806, over 3858099.00 frames. ], batch size: 42, lr: 6.05e-03, grad_scale: 16.0
+2023-03-28 20:05:49,786 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:05:50,051 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2446, 4.2718, 2.5458, 4.4950, 4.7150, 2.0122, 3.9765, 3.5184],
+       device='cuda:3'), covar=tensor([0.0582, 0.0753, 0.2647, 0.0867, 0.0501, 0.2807, 0.0884, 0.0794],
+       device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0253, 0.0228, 0.0268, 0.0248, 0.0202, 0.0238, 0.0194],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 20:06:15,246 INFO [train.py:892] (3/4) Epoch 26, batch 800, loss[loss=0.1556, simple_loss=0.2233, pruned_loss=0.04393, over 19866.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2469, pruned_loss=0.04748, over 3878587.28 frames. ], batch size: 129, lr: 6.04e-03, grad_scale: 16.0
+2023-03-28 20:07:12,099 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6793, 4.7473, 5.0662, 4.8895, 4.9804, 4.5498, 4.8036, 4.6277],
+       device='cuda:3'), covar=tensor([0.1460, 0.1585, 0.0921, 0.1218, 0.0784, 0.0971, 0.1934, 0.1976],
+       device='cuda:3'), in_proj_covar=tensor([0.0291, 0.0320, 0.0363, 0.0290, 0.0270, 0.0271, 0.0347, 0.0381],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 20:08:13,346 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.013e+02 4.034e+02 4.451e+02 5.296e+02 1.110e+03, threshold=8.902e+02, percent-clipped=2.0
+2023-03-28 20:08:25,314 INFO [train.py:892] (3/4) Epoch 26, batch 850, loss[loss=0.1593, simple_loss=0.2357, pruned_loss=0.04144, over 19891.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2474, pruned_loss=0.04775, over 3894652.44 frames. ], batch size: 87, lr: 6.04e-03, grad_scale: 16.0
+2023-03-28 20:09:05,014 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:09:28,908 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:09:48,109 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:10:29,776 INFO [train.py:892] (3/4) Epoch 26, batch 900, loss[loss=0.1935, simple_loss=0.2701, pruned_loss=0.05848, over 19775.00 frames. ], tot_loss[loss=0.1713, simple_loss=0.2474, pruned_loss=0.04759, over 3906756.79 frames. ], batch size: 247, lr: 6.04e-03, grad_scale: 16.0
+2023-03-28 20:11:16,019 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0
+2023-03-28 20:11:30,981 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:11:39,072 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:11:48,013 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47306.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:12:26,928 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.526e+02 3.904e+02 4.611e+02 5.515e+02 1.040e+03, threshold=9.223e+02, percent-clipped=2.0
+2023-03-28 20:12:28,115 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0448, 5.1922, 5.4811, 5.3575, 5.3315, 4.9323, 5.2152, 5.1221],
+       device='cuda:3'), covar=tensor([0.1453, 0.1486, 0.0871, 0.1122, 0.0685, 0.0918, 0.1762, 0.1877],
+       device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0318, 0.0362, 0.0289, 0.0268, 0.0269, 0.0346, 0.0379],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 20:12:38,830 INFO [train.py:892] (3/4) Epoch 26, batch 950, loss[loss=0.1902, simple_loss=0.2637, pruned_loss=0.05834, over 19765.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2479, pruned_loss=0.04755, over 3914165.46 frames. ], batch size: 263, lr: 6.03e-03, grad_scale: 16.0
+2023-03-28 20:14:46,326 INFO [train.py:892] (3/4) Epoch 26, batch 1000, loss[loss=0.1341, simple_loss=0.2, pruned_loss=0.03409, over 19754.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2476, pruned_loss=0.04741, over 3920540.82 frames. ], batch size: 129, lr: 6.03e-03, grad_scale: 16.0
+2023-03-28 20:14:55,143 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:16:35,937 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.1640, 4.7952, 4.8615, 5.1886, 4.8799, 5.3981, 5.2616, 5.4620],
+       device='cuda:3'), covar=tensor([0.0636, 0.0369, 0.0407, 0.0293, 0.0590, 0.0320, 0.0424, 0.0275],
+       device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0172, 0.0195, 0.0168, 0.0168, 0.0151, 0.0147, 0.0191],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 20:16:43,931 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.712e+02 3.612e+02 4.402e+02 5.356e+02 9.611e+02, threshold=8.803e+02, percent-clipped=1.0
+2023-03-28 20:16:52,099 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7159, 4.3744, 4.4493, 4.7447, 4.4828, 4.8413, 4.7867, 5.0077],
+       device='cuda:3'), covar=tensor([0.0692, 0.0383, 0.0546, 0.0346, 0.0677, 0.0435, 0.0507, 0.0285],
+       device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0172, 0.0195, 0.0168, 0.0168, 0.0151, 0.0147, 0.0191],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 20:16:55,242 INFO [train.py:892] (3/4) Epoch 26, batch 1050, loss[loss=0.1972, simple_loss=0.2727, pruned_loss=0.06087, over 19722.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.247, pruned_loss=0.04726, over 3928289.93 frames. ], batch size: 295, lr: 6.03e-03, grad_scale: 16.0
+2023-03-28 20:17:31,999 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:18:20,826 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1318, 3.0340, 4.5042, 3.4683, 3.6757, 3.5868, 2.5189, 2.6854],
+       device='cuda:3'), covar=tensor([0.0854, 0.2929, 0.0453, 0.0985, 0.1700, 0.1305, 0.2485, 0.2557],
+       device='cuda:3'), in_proj_covar=tensor([0.0345, 0.0378, 0.0339, 0.0276, 0.0368, 0.0363, 0.0361, 0.0329],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 20:18:34,481 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:18:57,923 INFO [train.py:892] (3/4) Epoch 26, batch 1100, loss[loss=0.1618, simple_loss=0.231, pruned_loss=0.04631, over 19741.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2476, pruned_loss=0.04776, over 3933664.13 frames. ], batch size: 99, lr: 6.03e-03, grad_scale: 16.0
+2023-03-28 20:20:37,059 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:20:53,320 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.834e+02 3.692e+02 4.407e+02 5.502e+02 8.056e+02, threshold=8.815e+02, percent-clipped=0.0
+2023-03-28 20:21:05,816 INFO [train.py:892] (3/4) Epoch 26, batch 1150, loss[loss=0.1645, simple_loss=0.2287, pruned_loss=0.05013, over 19822.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2467, pruned_loss=0.04741, over 3937715.66 frames. ], batch size: 123, lr: 6.02e-03, grad_scale: 16.0
+2023-03-28 20:21:54,157 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9804, 3.3816, 3.3536, 4.0248, 2.7671, 3.1500, 2.4890, 2.4283],
+       device='cuda:3'), covar=tensor([0.0523, 0.1714, 0.0966, 0.0358, 0.1820, 0.0785, 0.1358, 0.1690],
+       device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0332, 0.0244, 0.0193, 0.0241, 0.0202, 0.0211, 0.0214],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 20:23:12,005 INFO [train.py:892] (3/4) Epoch 26, batch 1200, loss[loss=0.1747, simple_loss=0.2569, pruned_loss=0.0462, over 19780.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2483, pruned_loss=0.04852, over 3940768.17 frames. ], batch size: 48, lr: 6.02e-03, grad_scale: 16.0
+2023-03-28 20:24:05,388 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:25:05,918 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.537e+02 4.111e+02 4.755e+02 5.836e+02 8.359e+02, threshold=9.510e+02, percent-clipped=0.0
+2023-03-28 20:25:19,254 INFO [train.py:892] (3/4) Epoch 26, batch 1250, loss[loss=0.1654, simple_loss=0.2383, pruned_loss=0.04621, over 19840.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2474, pruned_loss=0.0484, over 3944302.90 frames. ], batch size: 160, lr: 6.02e-03, grad_scale: 16.0
+2023-03-28 20:27:24,907 INFO [train.py:892] (3/4) Epoch 26, batch 1300, loss[loss=0.2226, simple_loss=0.301, pruned_loss=0.07212, over 19580.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.248, pruned_loss=0.04873, over 3946214.42 frames. ], batch size: 376, lr: 6.01e-03, grad_scale: 16.0
+2023-03-28 20:28:20,002 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5969, 3.0672, 3.4835, 3.0470, 3.8462, 3.8357, 4.4514, 4.9394],
+       device='cuda:3'), covar=tensor([0.0476, 0.1642, 0.1495, 0.2166, 0.1418, 0.1233, 0.0538, 0.0318],
+       device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0238, 0.0263, 0.0250, 0.0292, 0.0252, 0.0226, 0.0248],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 20:29:07,183 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6816, 4.3855, 4.4109, 4.1304, 4.6043, 3.1279, 3.8221, 2.0401],
+       device='cuda:3'), covar=tensor([0.0179, 0.0207, 0.0144, 0.0205, 0.0145, 0.0926, 0.0802, 0.1703],
+       device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0143, 0.0112, 0.0133, 0.0118, 0.0134, 0.0141, 0.0127],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 20:29:08,053 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-28 20:29:19,682 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.334e+02 3.779e+02 4.339e+02 5.421e+02 8.530e+02, threshold=8.679e+02, percent-clipped=0.0
+2023-03-28 20:29:34,042 INFO [train.py:892] (3/4) Epoch 26, batch 1350, loss[loss=0.1668, simple_loss=0.2466, pruned_loss=0.04348, over 19834.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2482, pruned_loss=0.04859, over 3945735.09 frames. ], batch size: 57, lr: 6.01e-03, grad_scale: 16.0
+2023-03-28 20:29:59,673 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47735.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:30:06,628 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:31:42,326 INFO [train.py:892] (3/4) Epoch 26, batch 1400, loss[loss=0.1765, simple_loss=0.2528, pruned_loss=0.05012, over 19652.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2492, pruned_loss=0.04926, over 3944520.05 frames. ], batch size: 47, lr: 6.01e-03, grad_scale: 16.0
+2023-03-28 20:32:14,039 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47788.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:32:43,736 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:32:59,913 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9595, 4.6903, 4.6489, 4.3475, 4.9107, 3.0232, 3.8382, 1.9995],
+       device='cuda:3'), covar=tensor([0.0269, 0.0239, 0.0207, 0.0237, 0.0240, 0.1127, 0.1134, 0.2285],
+       device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0142, 0.0112, 0.0132, 0.0118, 0.0134, 0.0140, 0.0126],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 20:33:05,711 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47809.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 20:33:29,902 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.973e+02 4.160e+02 4.762e+02 6.051e+02 1.589e+03, threshold=9.525e+02, percent-clipped=3.0
+2023-03-28 20:33:39,314 INFO [train.py:892] (3/4) Epoch 26, batch 1450, loss[loss=0.1664, simple_loss=0.2436, pruned_loss=0.04455, over 19683.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2495, pruned_loss=0.04919, over 3945059.08 frames. ], batch size: 82, lr: 6.00e-03, grad_scale: 16.0
+2023-03-28 20:34:39,370 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:35:29,961 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47870.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 20:35:46,893 INFO [train.py:892] (3/4) Epoch 26, batch 1500, loss[loss=0.1538, simple_loss=0.2226, pruned_loss=0.0425, over 19761.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2499, pruned_loss=0.04911, over 3945130.29 frames. ], batch size: 155, lr: 6.00e-03, grad_scale: 16.0
+2023-03-28 20:36:42,771 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:37:43,357 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.594e+02 4.057e+02 4.781e+02 5.680e+02 8.130e+02, threshold=9.563e+02, percent-clipped=0.0
+2023-03-28 20:37:54,769 INFO [train.py:892] (3/4) Epoch 26, batch 1550, loss[loss=0.1556, simple_loss=0.2383, pruned_loss=0.03641, over 19855.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2491, pruned_loss=0.04816, over 3946129.94 frames. ], batch size: 51, lr: 6.00e-03, grad_scale: 32.0
+2023-03-28 20:38:45,545 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:40:00,011 INFO [train.py:892] (3/4) Epoch 26, batch 1600, loss[loss=0.1721, simple_loss=0.2364, pruned_loss=0.05391, over 19795.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2498, pruned_loss=0.04799, over 3945351.23 frames. ], batch size: 174, lr: 5.99e-03, grad_scale: 32.0
+2023-03-28 20:41:53,734 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.602e+02 3.888e+02 4.549e+02 5.335e+02 7.774e+02, threshold=9.097e+02, percent-clipped=0.0
+2023-03-28 20:42:08,505 INFO [train.py:892] (3/4) Epoch 26, batch 1650, loss[loss=0.2498, simple_loss=0.3146, pruned_loss=0.09251, over 19626.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2492, pruned_loss=0.04816, over 3945584.01 frames. ], batch size: 359, lr: 5.99e-03, grad_scale: 32.0
+2023-03-28 20:42:32,504 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48035.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:44:05,496 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:44:15,577 INFO [train.py:892] (3/4) Epoch 26, batch 1700, loss[loss=0.1729, simple_loss=0.2521, pruned_loss=0.04687, over 19872.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2494, pruned_loss=0.04842, over 3945757.13 frames. ], batch size: 108, lr: 5.99e-03, grad_scale: 32.0
+2023-03-28 20:44:34,485 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:45:03,288 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48094.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:45:32,289 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.1062, 5.4410, 5.4576, 5.3862, 5.1237, 5.4599, 4.9254, 4.9398],
+       device='cuda:3'), covar=tensor([0.0446, 0.0434, 0.0461, 0.0387, 0.0596, 0.0507, 0.0610, 0.0941],
+       device='cuda:3'), in_proj_covar=tensor([0.0256, 0.0271, 0.0284, 0.0248, 0.0253, 0.0238, 0.0255, 0.0302],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 20:46:06,342 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.958e+02 3.839e+02 4.571e+02 5.458e+02 1.264e+03, threshold=9.143e+02, percent-clipped=4.0
+2023-03-28 20:46:18,198 INFO [train.py:892] (3/4) Epoch 26, batch 1750, loss[loss=0.1829, simple_loss=0.2675, pruned_loss=0.04918, over 19686.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2494, pruned_loss=0.04856, over 3947012.74 frames. ], batch size: 55, lr: 5.98e-03, grad_scale: 32.0
+2023-03-28 20:46:35,002 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48133.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:46:58,825 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:47:09,932 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7581, 2.8003, 2.8286, 2.3204, 2.9782, 2.4880, 2.8758, 2.8349],
+       device='cuda:3'), covar=tensor([0.0540, 0.0424, 0.0634, 0.0817, 0.0365, 0.0484, 0.0407, 0.0376],
+       device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0081, 0.0079, 0.0107, 0.0076, 0.0077, 0.0075, 0.0068],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 20:47:42,220 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48165.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 20:48:01,798 INFO [train.py:892] (3/4) Epoch 26, batch 1800, loss[loss=0.1664, simple_loss=0.2474, pruned_loss=0.04269, over 19771.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2489, pruned_loss=0.04862, over 3948124.56 frames. ], batch size: 108, lr: 5.98e-03, grad_scale: 16.0
+2023-03-28 20:49:11,116 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48212.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 20:49:30,288 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.294e+02 3.982e+02 4.834e+02 5.879e+02 1.454e+03, threshold=9.667e+02, percent-clipped=6.0
+2023-03-28 20:49:38,263 INFO [train.py:892] (3/4) Epoch 26, batch 1850, loss[loss=0.1916, simple_loss=0.2722, pruned_loss=0.05551, over 19839.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2508, pruned_loss=0.04853, over 3947959.97 frames. ], batch size: 58, lr: 5.98e-03, grad_scale: 16.0
+2023-03-28 20:50:46,038 INFO [train.py:892] (3/4) Epoch 27, batch 0, loss[loss=0.1652, simple_loss=0.2396, pruned_loss=0.04539, over 19469.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2396, pruned_loss=0.04539, over 19469.00 frames. ], batch size: 43, lr: 5.86e-03, grad_scale: 16.0
+2023-03-28 20:50:46,039 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-28 20:51:18,957 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8583, 3.8877, 4.1089, 3.8603, 3.7500, 3.9675, 3.7449, 4.1290],
+       device='cuda:3'), covar=tensor([0.0668, 0.0313, 0.0341, 0.0352, 0.0714, 0.0500, 0.0533, 0.0328],
+       device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0220, 0.0221, 0.0232, 0.0205, 0.0233, 0.0230, 0.0212],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 20:51:20,904 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.7725, 1.5796, 1.7738, 1.6870, 1.7071, 1.7216, 1.7269, 1.7561],
+       device='cuda:3'), covar=tensor([0.0339, 0.0343, 0.0330, 0.0319, 0.0441, 0.0343, 0.0452, 0.0392],
+       device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0074, 0.0077, 0.0070, 0.0086, 0.0078, 0.0095, 0.0069],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 20:51:25,319 INFO [train.py:926] (3/4) Epoch 27, validation: loss=0.1767, simple_loss=0.2485, pruned_loss=0.05248, over 2883724.00 frames. 
+2023-03-28 20:51:25,320 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-28 20:51:34,720 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3817, 4.7475, 5.1462, 4.6224, 4.2340, 4.9996, 4.8372, 5.3730],
+       device='cuda:3'), covar=tensor([0.1336, 0.0448, 0.0601, 0.0483, 0.0892, 0.0461, 0.0531, 0.0428],
+       device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0220, 0.0221, 0.0232, 0.0205, 0.0233, 0.0230, 0.0211],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 20:52:17,249 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.08 vs. limit=2.0
+2023-03-28 20:53:19,745 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48273.0, num_to_drop=1, layers_to_drop={3}
+2023-03-28 20:53:39,037 INFO [train.py:892] (3/4) Epoch 27, batch 50, loss[loss=0.179, simple_loss=0.2544, pruned_loss=0.05175, over 19804.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2502, pruned_loss=0.05122, over 890826.79 frames. ], batch size: 211, lr: 5.86e-03, grad_scale: 16.0
+2023-03-28 20:54:11,105 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7653, 5.0760, 5.1134, 5.0221, 4.7334, 5.1060, 4.5753, 4.5987],
+       device='cuda:3'), covar=tensor([0.0456, 0.0489, 0.0470, 0.0417, 0.0582, 0.0517, 0.0734, 0.0976],
+       device='cuda:3'), in_proj_covar=tensor([0.0257, 0.0273, 0.0288, 0.0250, 0.0255, 0.0240, 0.0259, 0.0305],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 20:55:21,749 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.295e+02 3.761e+02 4.503e+02 5.543e+02 9.672e+02, threshold=9.006e+02, percent-clipped=1.0
+2023-03-28 20:55:27,780 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48324.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:55:44,790 INFO [train.py:892] (3/4) Epoch 27, batch 100, loss[loss=0.2196, simple_loss=0.3004, pruned_loss=0.0694, over 19600.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2467, pruned_loss=0.04802, over 1570771.79 frames. ], batch size: 376, lr: 5.86e-03, grad_scale: 16.0
+2023-03-28 20:55:56,415 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5834, 3.6988, 2.2640, 4.4833, 3.9070, 4.4380, 4.4730, 3.3910],
+       device='cuda:3'), covar=tensor([0.0592, 0.0531, 0.1478, 0.0379, 0.0538, 0.0352, 0.0383, 0.0801],
+       device='cuda:3'), in_proj_covar=tensor([0.0139, 0.0139, 0.0140, 0.0146, 0.0130, 0.0129, 0.0142, 0.0142],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 20:57:46,629 INFO [train.py:892] (3/4) Epoch 27, batch 150, loss[loss=0.2049, simple_loss=0.2748, pruned_loss=0.06755, over 19679.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2454, pruned_loss=0.04685, over 2097886.21 frames. ], batch size: 325, lr: 5.86e-03, grad_scale: 16.0
+2023-03-28 20:57:57,096 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:58:23,461 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:58:30,291 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:59:26,432 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.124e+02 3.642e+02 4.290e+02 5.743e+02 1.049e+03, threshold=8.581e+02, percent-clipped=1.0
+2023-03-28 20:59:30,183 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:59:40,797 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 20:59:49,313 INFO [train.py:892] (3/4) Epoch 27, batch 200, loss[loss=0.1635, simple_loss=0.247, pruned_loss=0.03997, over 19889.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.249, pruned_loss=0.04836, over 2507496.58 frames. ], batch size: 52, lr: 5.85e-03, grad_scale: 16.0
+2023-03-28 21:00:15,427 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48442.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:00:20,330 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48444.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:00:53,451 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48458.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:01:08,513 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48465.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 21:01:09,075 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0
+2023-03-28 21:01:49,459 INFO [train.py:892] (3/4) Epoch 27, batch 250, loss[loss=0.1788, simple_loss=0.26, pruned_loss=0.0488, over 19641.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2471, pruned_loss=0.04738, over 2827401.21 frames. ], batch size: 330, lr: 5.85e-03, grad_scale: 16.0
+2023-03-28 21:01:58,509 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:02:17,786 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48492.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:03:11,478 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48513.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 21:03:17,598 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0813, 4.0690, 2.4456, 4.2762, 4.5018, 1.9801, 3.8393, 3.3992],
+       device='cuda:3'), covar=tensor([0.0684, 0.0904, 0.2910, 0.0887, 0.0589, 0.2872, 0.1014, 0.0854],
+       device='cuda:3'), in_proj_covar=tensor([0.0227, 0.0250, 0.0224, 0.0265, 0.0245, 0.0199, 0.0235, 0.0191],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 21:03:32,745 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.677e+02 3.668e+02 4.339e+02 5.007e+02 7.630e+02, threshold=8.678e+02, percent-clipped=0.0
+2023-03-28 21:03:55,771 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6413, 2.6294, 2.8373, 2.5180, 2.9763, 3.0007, 3.4528, 3.8140],
+       device='cuda:3'), covar=tensor([0.0646, 0.1665, 0.1599, 0.2241, 0.1611, 0.1440, 0.0685, 0.0637],
+       device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0235, 0.0259, 0.0249, 0.0289, 0.0251, 0.0224, 0.0246],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 21:03:56,513 INFO [train.py:892] (3/4) Epoch 27, batch 300, loss[loss=0.1759, simple_loss=0.2552, pruned_loss=0.04827, over 19777.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2478, pruned_loss=0.04758, over 3075830.34 frames. ], batch size: 226, lr: 5.85e-03, grad_scale: 16.0
+2023-03-28 21:04:47,299 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2176, 3.2707, 3.5192, 2.7016, 3.4877, 2.9108, 3.2314, 3.4274],
+       device='cuda:3'), covar=tensor([0.0728, 0.0411, 0.0432, 0.0752, 0.0410, 0.0414, 0.0527, 0.0355],
+       device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0081, 0.0078, 0.0106, 0.0076, 0.0078, 0.0075, 0.0068],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 21:05:21,969 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48568.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 21:05:49,821 INFO [train.py:892] (3/4) Epoch 27, batch 350, loss[loss=0.1772, simple_loss=0.2557, pruned_loss=0.04934, over 19764.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.247, pruned_loss=0.04641, over 3268043.64 frames. ], batch size: 244, lr: 5.84e-03, grad_scale: 16.0
+2023-03-28 21:07:35,026 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.631e+02 3.784e+02 4.446e+02 5.402e+02 1.015e+03, threshold=8.893e+02, percent-clipped=1.0
+2023-03-28 21:07:58,156 INFO [train.py:892] (3/4) Epoch 27, batch 400, loss[loss=0.159, simple_loss=0.2472, pruned_loss=0.03534, over 19720.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2467, pruned_loss=0.04678, over 3419415.95 frames. ], batch size: 54, lr: 5.84e-03, grad_scale: 16.0
+2023-03-28 21:10:03,100 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:10:04,278 INFO [train.py:892] (3/4) Epoch 27, batch 450, loss[loss=0.1921, simple_loss=0.2689, pruned_loss=0.05764, over 19777.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2456, pruned_loss=0.04654, over 3538184.15 frames. ], batch size: 280, lr: 5.84e-03, grad_scale: 16.0
+2023-03-28 21:10:32,332 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.3147, 2.6727, 2.2996, 1.7778, 2.3178, 2.5127, 2.4828, 2.6003],
+       device='cuda:3'), covar=tensor([0.0350, 0.0257, 0.0320, 0.0588, 0.0402, 0.0261, 0.0259, 0.0232],
+       device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0091, 0.0095, 0.0099, 0.0102, 0.0081, 0.0081, 0.0081],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 21:12:00,793 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7069, 4.3492, 4.4649, 4.2618, 4.6629, 3.3126, 3.9579, 2.4134],
+       device='cuda:3'), covar=tensor([0.0163, 0.0216, 0.0125, 0.0169, 0.0129, 0.0820, 0.0690, 0.1316],
+       device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0142, 0.0111, 0.0132, 0.0118, 0.0133, 0.0140, 0.0126],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 21:12:01,750 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.486e+02 3.534e+02 4.421e+02 5.220e+02 9.459e+02, threshold=8.842e+02, percent-clipped=1.0
+2023-03-28 21:12:21,882 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48728.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:12:28,183 INFO [train.py:892] (3/4) Epoch 27, batch 500, loss[loss=0.1653, simple_loss=0.2367, pruned_loss=0.047, over 19806.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.246, pruned_loss=0.04707, over 3629816.50 frames. ], batch size: 224, lr: 5.83e-03, grad_scale: 16.0
+2023-03-28 21:13:22,941 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:14:22,700 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:14:30,883 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:14:34,615 INFO [train.py:892] (3/4) Epoch 27, batch 550, loss[loss=0.1491, simple_loss=0.22, pruned_loss=0.03908, over 19833.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2463, pruned_loss=0.04742, over 3700303.45 frames. ], batch size: 90, lr: 5.83e-03, grad_scale: 16.0
+2023-03-28 21:14:35,920 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1962, 4.2538, 2.4606, 4.5263, 4.6746, 2.0233, 3.8489, 3.3995],
+       device='cuda:3'), covar=tensor([0.0639, 0.0725, 0.2767, 0.0651, 0.0525, 0.2708, 0.0973, 0.0850],
+       device='cuda:3'), in_proj_covar=tensor([0.0229, 0.0251, 0.0226, 0.0265, 0.0245, 0.0199, 0.0236, 0.0192],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 21:15:15,094 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0
+2023-03-28 21:16:22,031 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:16:22,712 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.382e+02 4.009e+02 4.912e+02 5.862e+02 1.286e+03, threshold=9.824e+02, percent-clipped=2.0
+2023-03-28 21:16:44,189 INFO [train.py:892] (3/4) Epoch 27, batch 600, loss[loss=0.1506, simple_loss=0.2291, pruned_loss=0.03607, over 19656.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2486, pruned_loss=0.04876, over 3753296.59 frames. ], batch size: 43, lr: 5.83e-03, grad_scale: 16.0
+2023-03-28 21:16:50,976 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0
+2023-03-28 21:17:02,420 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48838.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:18:12,860 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48868.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 21:18:21,685 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.24 vs. limit=5.0
+2023-03-28 21:18:42,151 INFO [train.py:892] (3/4) Epoch 27, batch 650, loss[loss=0.1765, simple_loss=0.2536, pruned_loss=0.04971, over 19802.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2489, pruned_loss=0.04905, over 3796780.83 frames. ], batch size: 51, lr: 5.83e-03, grad_scale: 16.0
+2023-03-28 21:18:46,950 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48882.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:19:27,548 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48899.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:20:07,152 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48916.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 21:20:19,921 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.871e+02 3.831e+02 4.540e+02 6.098e+02 1.137e+03, threshold=9.081e+02, percent-clipped=2.0
+2023-03-28 21:20:42,896 INFO [train.py:892] (3/4) Epoch 27, batch 700, loss[loss=0.2063, simple_loss=0.2798, pruned_loss=0.06642, over 19633.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2493, pruned_loss=0.04883, over 3829720.84 frames. ], batch size: 351, lr: 5.82e-03, grad_scale: 16.0
+2023-03-28 21:22:18,247 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8910, 3.5370, 3.7192, 3.8981, 3.6756, 3.8608, 3.9657, 4.1672],
+       device='cuda:3'), covar=tensor([0.0694, 0.0527, 0.0510, 0.0407, 0.0765, 0.0604, 0.0503, 0.0361],
+       device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0172, 0.0195, 0.0169, 0.0168, 0.0153, 0.0147, 0.0192],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 21:22:29,418 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.6290, 5.8982, 5.9204, 5.8250, 5.5814, 5.9224, 5.3082, 5.2783],
+       device='cuda:3'), covar=tensor([0.0350, 0.0405, 0.0438, 0.0401, 0.0463, 0.0470, 0.0612, 0.1014],
+       device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0271, 0.0284, 0.0248, 0.0253, 0.0237, 0.0256, 0.0302],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 21:22:48,534 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:22:49,429 INFO [train.py:892] (3/4) Epoch 27, batch 750, loss[loss=0.1621, simple_loss=0.2356, pruned_loss=0.04432, over 19779.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2485, pruned_loss=0.04854, over 3856792.04 frames. ], batch size: 233, lr: 5.82e-03, grad_scale: 16.0
+2023-03-28 21:24:16,071 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.0382, 1.7535, 2.6389, 1.7861, 2.6709, 2.8841, 2.4729, 2.7745],
+       device='cuda:3'), covar=tensor([0.0825, 0.1052, 0.0143, 0.0307, 0.0138, 0.0214, 0.0210, 0.0187],
+       device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0102, 0.0086, 0.0153, 0.0082, 0.0097, 0.0089, 0.0084],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0002, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 21:24:23,144 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5143, 3.8628, 3.9966, 4.7211, 3.1552, 3.5273, 2.8745, 2.8053],
+       device='cuda:3'), covar=tensor([0.0490, 0.2027, 0.0877, 0.0315, 0.2052, 0.0920, 0.1302, 0.1697],
+       device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0333, 0.0245, 0.0196, 0.0246, 0.0204, 0.0214, 0.0215],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 21:24:30,278 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0
+2023-03-28 21:24:30,789 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.511e+02 3.742e+02 4.626e+02 5.475e+02 1.269e+03, threshold=9.252e+02, percent-clipped=2.0
+2023-03-28 21:24:49,079 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49028.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:24:56,453 INFO [train.py:892] (3/4) Epoch 27, batch 800, loss[loss=0.1395, simple_loss=0.218, pruned_loss=0.03046, over 19857.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2482, pruned_loss=0.04811, over 3875564.65 frames. ], batch size: 106, lr: 5.82e-03, grad_scale: 16.0
+2023-03-28 21:25:42,277 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3351, 4.2253, 4.7134, 4.2729, 3.9518, 4.5111, 4.3492, 4.7763],
+       device='cuda:3'), covar=tensor([0.0866, 0.0387, 0.0335, 0.0385, 0.0928, 0.0479, 0.0438, 0.0309],
+       device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0220, 0.0219, 0.0231, 0.0205, 0.0234, 0.0229, 0.0211],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 21:25:52,195 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:26:11,917 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.0613, 2.1744, 2.2989, 2.1432, 2.2261, 2.2692, 2.1605, 2.3267],
+       device='cuda:3'), covar=tensor([0.0369, 0.0307, 0.0280, 0.0300, 0.0412, 0.0301, 0.0428, 0.0322],
+       device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0075, 0.0077, 0.0072, 0.0085, 0.0079, 0.0096, 0.0068],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 21:26:58,629 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49079.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:27:02,254 INFO [train.py:892] (3/4) Epoch 27, batch 850, loss[loss=0.1568, simple_loss=0.2278, pruned_loss=0.04292, over 19810.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2488, pruned_loss=0.04805, over 3891657.72 frames. ], batch size: 149, lr: 5.81e-03, grad_scale: 16.0
+2023-03-28 21:27:51,032 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:27:51,146 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7959, 3.5028, 3.6066, 3.7842, 3.5762, 3.7355, 3.9064, 4.0826],
+       device='cuda:3'), covar=tensor([0.0665, 0.0493, 0.0584, 0.0412, 0.0805, 0.0657, 0.0454, 0.0328],
+       device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0172, 0.0196, 0.0170, 0.0169, 0.0154, 0.0147, 0.0193],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 21:28:42,785 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.493e+02 3.766e+02 4.379e+02 5.390e+02 1.134e+03, threshold=8.757e+02, percent-clipped=1.0
+2023-03-28 21:28:56,331 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49127.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:29:04,517 INFO [train.py:892] (3/4) Epoch 27, batch 900, loss[loss=0.1611, simple_loss=0.2314, pruned_loss=0.04537, over 19781.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2491, pruned_loss=0.04772, over 3903056.79 frames. ], batch size: 131, lr: 5.81e-03, grad_scale: 16.0
+2023-03-28 21:30:36,856 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.99 vs. limit=5.0
+2023-03-28 21:31:02,180 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:31:10,428 INFO [train.py:892] (3/4) Epoch 27, batch 950, loss[loss=0.1677, simple_loss=0.2296, pruned_loss=0.05294, over 19828.00 frames. ], tot_loss[loss=0.172, simple_loss=0.249, pruned_loss=0.04751, over 3913027.54 frames. ], batch size: 128, lr: 5.81e-03, grad_scale: 16.0
+2023-03-28 21:31:41,754 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49194.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:32:49,475 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.549e+02 3.706e+02 4.158e+02 5.021e+02 8.010e+02, threshold=8.316e+02, percent-clipped=0.0
+2023-03-28 21:33:13,386 INFO [train.py:892] (3/4) Epoch 27, batch 1000, loss[loss=0.1595, simple_loss=0.2329, pruned_loss=0.04309, over 19764.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2473, pruned_loss=0.04686, over 3922076.95 frames. ], batch size: 226, lr: 5.81e-03, grad_scale: 16.0
+2023-03-28 21:33:25,904 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:34:24,244 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.0335, 1.6577, 2.6273, 1.6752, 2.6485, 2.7524, 2.4448, 2.6782],
+       device='cuda:3'), covar=tensor([0.0845, 0.1163, 0.0159, 0.0369, 0.0171, 0.0257, 0.0246, 0.0205],
+       device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0101, 0.0085, 0.0152, 0.0081, 0.0096, 0.0088, 0.0084],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0002, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 21:34:34,398 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2939, 3.2294, 3.5200, 2.7324, 3.5103, 3.0223, 3.2217, 3.4803],
+       device='cuda:3'), covar=tensor([0.0575, 0.0397, 0.0509, 0.0768, 0.0373, 0.0430, 0.0529, 0.0368],
+       device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0082, 0.0079, 0.0107, 0.0076, 0.0078, 0.0075, 0.0068],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 21:34:34,483 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7123, 3.1492, 2.5732, 2.2590, 2.7651, 3.0049, 2.9068, 3.0322],
+       device='cuda:3'), covar=tensor([0.0346, 0.0258, 0.0295, 0.0556, 0.0343, 0.0299, 0.0266, 0.0206],
+       device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0091, 0.0095, 0.0099, 0.0102, 0.0081, 0.0080, 0.0082],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 21:35:10,576 INFO [train.py:892] (3/4) Epoch 27, batch 1050, loss[loss=0.1805, simple_loss=0.2605, pruned_loss=0.05022, over 19849.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2493, pruned_loss=0.04716, over 3925791.10 frames. ], batch size: 58, lr: 5.80e-03, grad_scale: 16.0
+2023-03-28 21:35:54,784 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:36:54,139 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.170e+02 3.715e+02 4.308e+02 4.982e+02 8.833e+02, threshold=8.615e+02, percent-clipped=1.0
+2023-03-28 21:37:20,673 INFO [train.py:892] (3/4) Epoch 27, batch 1100, loss[loss=0.1469, simple_loss=0.2216, pruned_loss=0.03611, over 19850.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2477, pruned_loss=0.04712, over 3932138.69 frames. ], batch size: 112, lr: 5.80e-03, grad_scale: 16.0
+2023-03-28 21:38:01,676 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0
+2023-03-28 21:39:23,373 INFO [train.py:892] (3/4) Epoch 27, batch 1150, loss[loss=0.1784, simple_loss=0.2609, pruned_loss=0.04794, over 19794.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2473, pruned_loss=0.04715, over 3936764.17 frames. ], batch size: 105, lr: 5.80e-03, grad_scale: 16.0
+2023-03-28 21:41:04,930 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.904e+02 3.848e+02 4.568e+02 5.710e+02 9.674e+02, threshold=9.135e+02, percent-clipped=2.0
+2023-03-28 21:41:06,057 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:41:13,638 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0561, 3.0070, 3.2714, 2.5486, 3.3013, 2.7906, 3.1051, 3.2485],
+       device='cuda:3'), covar=tensor([0.0637, 0.0437, 0.0464, 0.0769, 0.0395, 0.0449, 0.0399, 0.0326],
+       device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0082, 0.0079, 0.0107, 0.0076, 0.0078, 0.0075, 0.0068],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 21:41:28,171 INFO [train.py:892] (3/4) Epoch 27, batch 1200, loss[loss=0.1723, simple_loss=0.2547, pruned_loss=0.04494, over 19748.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.247, pruned_loss=0.04703, over 3940612.66 frames. ], batch size: 276, lr: 5.79e-03, grad_scale: 16.0
+2023-03-28 21:41:32,121 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49432.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 21:43:28,810 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:43:35,928 INFO [train.py:892] (3/4) Epoch 27, batch 1250, loss[loss=0.1675, simple_loss=0.2401, pruned_loss=0.04742, over 19834.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2471, pruned_loss=0.0474, over 3943424.56 frames. ], batch size: 161, lr: 5.79e-03, grad_scale: 16.0
+2023-03-28 21:43:41,786 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:43:49,008 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6767, 2.8296, 4.8850, 4.1092, 4.6599, 4.7309, 4.7790, 4.4537],
+       device='cuda:3'), covar=tensor([0.0602, 0.0994, 0.0110, 0.1069, 0.0146, 0.0236, 0.0146, 0.0164],
+       device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0102, 0.0086, 0.0153, 0.0082, 0.0097, 0.0089, 0.0084],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0002, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 21:44:06,644 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49493.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 21:44:09,323 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:45:00,947 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3562, 4.6178, 4.6725, 4.5524, 4.3466, 4.6430, 4.1945, 4.2118],
+       device='cuda:3'), covar=tensor([0.0533, 0.0550, 0.0509, 0.0479, 0.0649, 0.0526, 0.0686, 0.0970],
+       device='cuda:3'), in_proj_covar=tensor([0.0256, 0.0271, 0.0285, 0.0249, 0.0254, 0.0237, 0.0256, 0.0302],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 21:45:21,343 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.439e+02 3.880e+02 4.551e+02 5.249e+02 9.846e+02, threshold=9.103e+02, percent-clipped=1.0
+2023-03-28 21:45:30,241 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:45:44,608 INFO [train.py:892] (3/4) Epoch 27, batch 1300, loss[loss=0.1556, simple_loss=0.2301, pruned_loss=0.04054, over 19597.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2487, pruned_loss=0.04817, over 3942603.80 frames. ], batch size: 45, lr: 5.79e-03, grad_scale: 16.0
+2023-03-28 21:46:14,591 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:47:00,224 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7967, 5.0976, 5.1403, 5.0122, 4.7271, 5.1049, 4.6236, 4.6517],
+       device='cuda:3'), covar=tensor([0.0439, 0.0421, 0.0469, 0.0405, 0.0629, 0.0518, 0.0709, 0.0921],
+       device='cuda:3'), in_proj_covar=tensor([0.0256, 0.0270, 0.0285, 0.0250, 0.0256, 0.0238, 0.0256, 0.0302],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 21:47:50,833 INFO [train.py:892] (3/4) Epoch 27, batch 1350, loss[loss=0.1966, simple_loss=0.273, pruned_loss=0.06015, over 19702.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2496, pruned_loss=0.04848, over 3944081.52 frames. ], batch size: 305, lr: 5.78e-03, grad_scale: 16.0
+2023-03-28 21:48:20,140 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:49:30,804 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.556e+02 4.075e+02 4.881e+02 5.911e+02 1.218e+03, threshold=9.761e+02, percent-clipped=5.0
+2023-03-28 21:49:51,520 INFO [train.py:892] (3/4) Epoch 27, batch 1400, loss[loss=0.1724, simple_loss=0.2463, pruned_loss=0.04923, over 19804.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2476, pruned_loss=0.0472, over 3945876.97 frames. ], batch size: 195, lr: 5.78e-03, grad_scale: 16.0
+2023-03-28 21:51:11,791 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-28 21:51:45,987 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-28 21:51:48,488 INFO [train.py:892] (3/4) Epoch 27, batch 1450, loss[loss=0.1484, simple_loss=0.2275, pruned_loss=0.03469, over 19725.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2485, pruned_loss=0.04742, over 3946345.91 frames. ], batch size: 61, lr: 5.78e-03, grad_scale: 16.0
+2023-03-28 21:52:11,604 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.08 vs. limit=2.0
+2023-03-28 21:52:58,783 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9052, 2.8382, 3.0679, 2.7251, 3.2047, 3.1853, 3.7301, 4.1310],
+       device='cuda:3'), covar=tensor([0.0637, 0.1650, 0.1582, 0.2177, 0.1704, 0.1436, 0.0651, 0.0613],
+       device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0240, 0.0265, 0.0253, 0.0293, 0.0255, 0.0229, 0.0251],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 21:53:31,617 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.500e+02 3.941e+02 4.673e+02 5.511e+02 7.393e+02, threshold=9.346e+02, percent-clipped=0.0
+2023-03-28 21:53:54,193 INFO [train.py:892] (3/4) Epoch 27, batch 1500, loss[loss=0.1864, simple_loss=0.2755, pruned_loss=0.04866, over 19699.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.248, pruned_loss=0.0471, over 3946193.50 frames. ], batch size: 56, lr: 5.78e-03, grad_scale: 16.0
+2023-03-28 21:55:22,822 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:55:51,694 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:55:57,529 INFO [train.py:892] (3/4) Epoch 27, batch 1550, loss[loss=0.1593, simple_loss=0.2341, pruned_loss=0.04223, over 19855.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2482, pruned_loss=0.04677, over 3946396.04 frames. ], batch size: 118, lr: 5.77e-03, grad_scale: 16.0
+2023-03-28 21:55:59,158 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.5285, 5.0232, 5.1562, 4.9041, 5.4130, 3.6166, 4.3557, 2.8442],
+       device='cuda:3'), covar=tensor([0.0169, 0.0185, 0.0112, 0.0157, 0.0118, 0.0751, 0.0909, 0.1335],
+       device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0143, 0.0112, 0.0132, 0.0117, 0.0133, 0.0139, 0.0126],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 21:56:16,183 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49788.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 21:57:39,897 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.487e+02 3.954e+02 4.680e+02 6.114e+02 1.039e+03, threshold=9.360e+02, percent-clipped=2.0
+2023-03-28 21:57:53,466 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49827.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 21:58:00,509 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8392, 2.9099, 3.1012, 2.9128, 2.7635, 3.0452, 2.7990, 3.1413],
+       device='cuda:3'), covar=tensor([0.0264, 0.0297, 0.0257, 0.0307, 0.0407, 0.0266, 0.0398, 0.0316],
+       device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0075, 0.0078, 0.0071, 0.0085, 0.0079, 0.0096, 0.0069],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 21:58:01,494 INFO [train.py:892] (3/4) Epoch 27, batch 1600, loss[loss=0.1445, simple_loss=0.2323, pruned_loss=0.02839, over 19656.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2481, pruned_loss=0.0465, over 3946986.62 frames. ], batch size: 58, lr: 5.77e-03, grad_scale: 16.0
+2023-03-28 21:58:23,842 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0
+2023-03-28 22:00:08,577 INFO [train.py:892] (3/4) Epoch 27, batch 1650, loss[loss=0.1584, simple_loss=0.2333, pruned_loss=0.04176, over 19794.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.247, pruned_loss=0.04586, over 3946368.66 frames. ], batch size: 120, lr: 5.77e-03, grad_scale: 16.0
+2023-03-28 22:00:15,167 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8136, 2.8786, 4.2671, 3.3534, 3.5771, 3.3078, 2.5387, 2.5403],
+       device='cuda:3'), covar=tensor([0.1016, 0.3103, 0.0550, 0.0968, 0.1588, 0.1399, 0.2286, 0.2674],
+       device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0382, 0.0342, 0.0279, 0.0369, 0.0365, 0.0364, 0.0334],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 22:00:36,704 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:01:35,666 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6241, 4.5641, 4.9825, 4.7485, 4.8799, 4.4461, 4.7165, 4.4919],
+       device='cuda:3'), covar=tensor([0.1491, 0.1581, 0.0975, 0.1303, 0.0895, 0.1005, 0.1911, 0.2034],
+       device='cuda:3'), in_proj_covar=tensor([0.0293, 0.0324, 0.0366, 0.0293, 0.0273, 0.0271, 0.0354, 0.0389],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 22:01:55,287 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.537e+02 3.614e+02 4.321e+02 5.513e+02 1.291e+03, threshold=8.641e+02, percent-clipped=1.0
+2023-03-28 22:02:20,816 INFO [train.py:892] (3/4) Epoch 27, batch 1700, loss[loss=0.1429, simple_loss=0.2164, pruned_loss=0.03465, over 19810.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2458, pruned_loss=0.04545, over 3947232.45 frames. ], batch size: 98, lr: 5.76e-03, grad_scale: 16.0
+2023-03-28 22:02:44,476 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:03:56,412 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6396, 2.7549, 4.0088, 3.1011, 3.4602, 3.1650, 2.3639, 2.4997],
+       device='cuda:3'), covar=tensor([0.1138, 0.3328, 0.0627, 0.1117, 0.1617, 0.1564, 0.2450, 0.2733],
+       device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0380, 0.0341, 0.0278, 0.0367, 0.0364, 0.0363, 0.0333],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 22:04:12,201 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6468, 4.9171, 4.9807, 4.8541, 4.6400, 4.9007, 4.4478, 4.4599],
+       device='cuda:3'), covar=tensor([0.0471, 0.0428, 0.0414, 0.0405, 0.0552, 0.0509, 0.0621, 0.0929],
+       device='cuda:3'), in_proj_covar=tensor([0.0256, 0.0271, 0.0286, 0.0249, 0.0254, 0.0238, 0.0256, 0.0302],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 22:04:25,023 INFO [train.py:892] (3/4) Epoch 27, batch 1750, loss[loss=0.1891, simple_loss=0.2642, pruned_loss=0.05699, over 19792.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2452, pruned_loss=0.04508, over 3949097.80 frames. ], batch size: 280, lr: 5.76e-03, grad_scale: 16.0
+2023-03-28 22:05:32,216 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1940, 3.2678, 2.0456, 3.8673, 3.4473, 3.7558, 3.8709, 3.0466],
+       device='cuda:3'), covar=tensor([0.0603, 0.0571, 0.1490, 0.0516, 0.0574, 0.0429, 0.0580, 0.0774],
+       device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0141, 0.0141, 0.0148, 0.0131, 0.0131, 0.0144, 0.0142],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 22:05:54,959 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.637e+02 3.767e+02 4.699e+02 5.313e+02 7.903e+02, threshold=9.399e+02, percent-clipped=0.0
+2023-03-28 22:06:12,870 INFO [train.py:892] (3/4) Epoch 27, batch 1800, loss[loss=0.1638, simple_loss=0.2415, pruned_loss=0.04299, over 19828.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2448, pruned_loss=0.04516, over 3949936.76 frames. ], batch size: 93, lr: 5.76e-03, grad_scale: 16.0
+2023-03-28 22:07:45,093 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0839, 3.4005, 2.8201, 2.4304, 2.8681, 3.2985, 3.1885, 3.2455],
+       device='cuda:3'), covar=tensor([0.0242, 0.0275, 0.0283, 0.0512, 0.0358, 0.0199, 0.0213, 0.0202],
+       device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0091, 0.0094, 0.0098, 0.0100, 0.0081, 0.0080, 0.0081],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 22:07:55,289 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:08:00,315 INFO [train.py:892] (3/4) Epoch 27, batch 1850, loss[loss=0.1643, simple_loss=0.2486, pruned_loss=0.03999, over 19671.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2479, pruned_loss=0.04596, over 3949267.96 frames. ], batch size: 55, lr: 5.76e-03, grad_scale: 16.0
+2023-03-28 22:09:11,413 INFO [train.py:892] (3/4) Epoch 28, batch 0, loss[loss=0.1687, simple_loss=0.2374, pruned_loss=0.05001, over 19844.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.2374, pruned_loss=0.05001, over 19844.00 frames. ], batch size: 145, lr: 5.65e-03, grad_scale: 16.0
+2023-03-28 22:09:11,414 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-28 22:09:45,364 INFO [train.py:926] (3/4) Epoch 28, validation: loss=0.1765, simple_loss=0.2481, pruned_loss=0.05251, over 2883724.00 frames. 
+2023-03-28 22:09:45,366 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-28 22:09:52,071 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50088.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 22:10:30,969 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9666, 2.9706, 4.3724, 3.3298, 3.7738, 3.4306, 2.5059, 2.5870],
+       device='cuda:3'), covar=tensor([0.0961, 0.3217, 0.0567, 0.1050, 0.1501, 0.1487, 0.2490, 0.2707],
+       device='cuda:3'), in_proj_covar=tensor([0.0347, 0.0380, 0.0341, 0.0277, 0.0366, 0.0364, 0.0363, 0.0333],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 22:11:19,946 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.385e+02 4.071e+02 4.813e+02 5.962e+02 1.168e+03, threshold=9.627e+02, percent-clipped=3.0
+2023-03-28 22:11:21,122 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:11:32,563 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:11:37,452 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50128.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:11:58,853 INFO [train.py:892] (3/4) Epoch 28, batch 50, loss[loss=0.1378, simple_loss=0.2104, pruned_loss=0.03264, over 19841.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.244, pruned_loss=0.04588, over 890156.19 frames. ], batch size: 109, lr: 5.65e-03, grad_scale: 16.0
+2023-03-28 22:11:59,966 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50136.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 22:14:00,316 INFO [train.py:892] (3/4) Epoch 28, batch 100, loss[loss=0.1445, simple_loss=0.2204, pruned_loss=0.03429, over 19746.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2461, pruned_loss=0.04574, over 1569760.97 frames. ], batch size: 44, lr: 5.64e-03, grad_scale: 32.0
+2023-03-28 22:14:01,539 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.2303, 2.5574, 2.3222, 1.7899, 2.4336, 2.5123, 2.5647, 2.5394],
+       device='cuda:3'), covar=tensor([0.0404, 0.0337, 0.0322, 0.0608, 0.0388, 0.0310, 0.0299, 0.0285],
+       device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0092, 0.0096, 0.0100, 0.0102, 0.0083, 0.0081, 0.0083],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 22:14:08,159 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50189.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:14:18,527 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8030, 4.8934, 5.1730, 4.8947, 5.0848, 4.7554, 4.9154, 4.6806],
+       device='cuda:3'), covar=tensor([0.1472, 0.1516, 0.1044, 0.1323, 0.0751, 0.0874, 0.1853, 0.2036],
+       device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0323, 0.0367, 0.0295, 0.0273, 0.0272, 0.0354, 0.0391],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-28 22:14:24,092 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50197.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 22:14:26,138 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6003, 3.3408, 3.9853, 2.9686, 3.9977, 3.1866, 3.4822, 3.9564],
+       device='cuda:3'), covar=tensor([0.0570, 0.0379, 0.0232, 0.0717, 0.0292, 0.0411, 0.0389, 0.0267],
+       device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0081, 0.0108, 0.0076, 0.0079, 0.0076, 0.0069],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 22:15:23,505 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.658e+02 4.112e+02 4.923e+02 6.330e+02 1.508e+03, threshold=9.845e+02, percent-clipped=2.0
+2023-03-28 22:15:54,976 INFO [train.py:892] (3/4) Epoch 28, batch 150, loss[loss=0.1659, simple_loss=0.2534, pruned_loss=0.03915, over 19852.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2459, pruned_loss=0.04506, over 2098105.73 frames. ], batch size: 56, lr: 5.64e-03, grad_scale: 32.0
+2023-03-28 22:16:50,309 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50258.0, num_to_drop=1, layers_to_drop={3}
+2023-03-28 22:17:57,280 INFO [train.py:892] (3/4) Epoch 28, batch 200, loss[loss=0.1739, simple_loss=0.2536, pruned_loss=0.04715, over 19860.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2458, pruned_loss=0.04533, over 2508632.74 frames. ], batch size: 51, lr: 5.64e-03, grad_scale: 32.0
+2023-03-28 22:18:22,384 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3504, 3.6407, 3.7469, 4.4556, 3.0406, 3.3431, 2.7099, 2.7095],
+       device='cuda:3'), covar=tensor([0.0502, 0.1967, 0.0926, 0.0336, 0.1877, 0.0920, 0.1352, 0.1660],
+       device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0332, 0.0245, 0.0197, 0.0247, 0.0204, 0.0214, 0.0216],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 22:19:26,391 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.556e+02 4.059e+02 4.750e+02 5.729e+02 1.085e+03, threshold=9.500e+02, percent-clipped=1.0
+2023-03-28 22:19:44,118 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4703, 2.5569, 2.7577, 2.4291, 2.9278, 2.8814, 3.3661, 3.6429],
+       device='cuda:3'), covar=tensor([0.0701, 0.1683, 0.1606, 0.2199, 0.1715, 0.1505, 0.0679, 0.0697],
+       device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0240, 0.0265, 0.0255, 0.0295, 0.0255, 0.0231, 0.0253],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 22:20:01,080 INFO [train.py:892] (3/4) Epoch 28, batch 250, loss[loss=0.1516, simple_loss=0.2311, pruned_loss=0.03608, over 19836.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2445, pruned_loss=0.04547, over 2829066.94 frames. ], batch size: 171, lr: 5.64e-03, grad_scale: 16.0
+2023-03-28 22:21:44,521 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5175, 2.1534, 3.5003, 3.0048, 3.5282, 3.5465, 3.4012, 3.4063],
+       device='cuda:3'), covar=tensor([0.0729, 0.1020, 0.0117, 0.0476, 0.0131, 0.0256, 0.0215, 0.0183],
+       device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0102, 0.0085, 0.0152, 0.0082, 0.0096, 0.0088, 0.0084],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0002, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 22:22:07,549 INFO [train.py:892] (3/4) Epoch 28, batch 300, loss[loss=0.1897, simple_loss=0.2623, pruned_loss=0.05856, over 19663.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2458, pruned_loss=0.04599, over 3076853.87 frames. ], batch size: 50, lr: 5.63e-03, grad_scale: 16.0
+2023-03-28 22:23:39,869 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:23:40,969 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.722e+02 4.064e+02 4.778e+02 6.064e+02 1.211e+03, threshold=9.555e+02, percent-clipped=3.0
+2023-03-28 22:24:15,471 INFO [train.py:892] (3/4) Epoch 28, batch 350, loss[loss=0.158, simple_loss=0.2466, pruned_loss=0.03472, over 19618.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.245, pruned_loss=0.04527, over 3271163.94 frames. ], batch size: 52, lr: 5.63e-03, grad_scale: 16.0
+2023-03-28 22:25:39,668 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:26:17,100 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50484.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:26:20,283 INFO [train.py:892] (3/4) Epoch 28, batch 400, loss[loss=0.1611, simple_loss=0.2412, pruned_loss=0.04049, over 19730.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.247, pruned_loss=0.04661, over 3419466.04 frames. ], batch size: 104, lr: 5.63e-03, grad_scale: 16.0
+2023-03-28 22:27:29,137 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6564, 4.0151, 4.2692, 4.8751, 3.1792, 3.5689, 3.0539, 2.9790],
+       device='cuda:3'), covar=tensor([0.0501, 0.1898, 0.0738, 0.0318, 0.2015, 0.1032, 0.1229, 0.1626],
+       device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0335, 0.0247, 0.0198, 0.0248, 0.0207, 0.0216, 0.0218],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 22:27:38,227 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.2828, 4.7472, 4.8790, 4.6005, 5.1372, 3.3455, 4.1978, 2.4341],
+       device='cuda:3'), covar=tensor([0.0155, 0.0203, 0.0126, 0.0185, 0.0124, 0.0896, 0.0812, 0.1543],
+       device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0143, 0.0112, 0.0132, 0.0117, 0.0133, 0.0139, 0.0126],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 22:27:49,853 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.616e+02 4.023e+02 4.684e+02 5.835e+02 1.086e+03, threshold=9.368e+02, percent-clipped=2.0
+2023-03-28 22:28:11,299 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4376, 2.5083, 2.6138, 2.5399, 2.5027, 2.6767, 2.4005, 2.7264],
+       device='cuda:3'), covar=tensor([0.0345, 0.0303, 0.0340, 0.0282, 0.0397, 0.0282, 0.0466, 0.0262],
+       device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0075, 0.0078, 0.0071, 0.0085, 0.0078, 0.0096, 0.0068],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 22:28:16,138 INFO [train.py:892] (3/4) Epoch 28, batch 450, loss[loss=0.2481, simple_loss=0.3328, pruned_loss=0.08167, over 19244.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2489, pruned_loss=0.04743, over 3535742.42 frames. ], batch size: 483, lr: 5.63e-03, grad_scale: 16.0
+2023-03-28 22:28:42,929 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50546.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:28:46,671 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4200, 3.1196, 3.4316, 3.0062, 3.6545, 3.6523, 4.2401, 4.7209],
+       device='cuda:3'), covar=tensor([0.0597, 0.1717, 0.1489, 0.2203, 0.1775, 0.1353, 0.0609, 0.0587],
+       device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0242, 0.0267, 0.0256, 0.0297, 0.0257, 0.0232, 0.0255],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 22:29:01,680 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50553.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 22:29:57,193 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0
+2023-03-28 22:30:25,771 INFO [train.py:892] (3/4) Epoch 28, batch 500, loss[loss=0.1629, simple_loss=0.2466, pruned_loss=0.03966, over 19644.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2475, pruned_loss=0.04688, over 3627863.95 frames. ], batch size: 69, lr: 5.62e-03, grad_scale: 16.0
+2023-03-28 22:31:19,327 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:31:40,367 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.55 vs. limit=5.0
+2023-03-28 22:31:59,941 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.484e+02 3.842e+02 4.586e+02 5.688e+02 1.000e+03, threshold=9.173e+02, percent-clipped=1.0
+2023-03-28 22:32:20,786 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5218, 2.7742, 4.5594, 3.9235, 4.3779, 4.4649, 4.3640, 4.1620],
+       device='cuda:3'), covar=tensor([0.0501, 0.0904, 0.0100, 0.0817, 0.0132, 0.0203, 0.0153, 0.0161],
+       device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0102, 0.0086, 0.0152, 0.0083, 0.0096, 0.0088, 0.0084],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0002, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 22:32:31,003 INFO [train.py:892] (3/4) Epoch 28, batch 550, loss[loss=0.1567, simple_loss=0.2295, pruned_loss=0.04194, over 19767.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2458, pruned_loss=0.04592, over 3699944.15 frames. ], batch size: 152, lr: 5.62e-03, grad_scale: 16.0
+2023-03-28 22:34:36,429 INFO [train.py:892] (3/4) Epoch 28, batch 600, loss[loss=0.2016, simple_loss=0.2688, pruned_loss=0.06716, over 19738.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.247, pruned_loss=0.04636, over 3755309.95 frames. ], batch size: 134, lr: 5.62e-03, grad_scale: 16.0
+2023-03-28 22:36:08,583 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.507e+02 3.618e+02 4.238e+02 5.077e+02 1.172e+03, threshold=8.476e+02, percent-clipped=2.0
+2023-03-28 22:36:42,064 INFO [train.py:892] (3/4) Epoch 28, batch 650, loss[loss=0.177, simple_loss=0.2577, pruned_loss=0.04818, over 19854.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2464, pruned_loss=0.04656, over 3797688.99 frames. ], batch size: 60, lr: 5.61e-03, grad_scale: 16.0
+2023-03-28 22:38:38,804 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:38:43,603 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:38:46,766 INFO [train.py:892] (3/4) Epoch 28, batch 700, loss[loss=0.1633, simple_loss=0.2332, pruned_loss=0.04671, over 19774.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2469, pruned_loss=0.04666, over 3830987.56 frames. ], batch size: 152, lr: 5.61e-03, grad_scale: 16.0
+2023-03-28 22:39:45,299 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50808.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 22:40:20,257 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.436e+02 3.963e+02 4.473e+02 5.641e+02 1.140e+03, threshold=8.946e+02, percent-clipped=3.0
+2023-03-28 22:40:44,036 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50832.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:40:51,130 INFO [train.py:892] (3/4) Epoch 28, batch 750, loss[loss=0.1527, simple_loss=0.2315, pruned_loss=0.03695, over 19886.00 frames. ], tot_loss[loss=0.1694, simple_loss=0.2462, pruned_loss=0.04629, over 3857679.55 frames. ], batch size: 47, lr: 5.61e-03, grad_scale: 16.0
+2023-03-28 22:41:08,961 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50843.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:41:33,243 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50853.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 22:42:13,674 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50869.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 22:42:55,254 INFO [train.py:892] (3/4) Epoch 28, batch 800, loss[loss=0.1693, simple_loss=0.2407, pruned_loss=0.04888, over 19835.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2462, pruned_loss=0.04617, over 3878291.07 frames. ], batch size: 208, lr: 5.61e-03, grad_scale: 16.0
+2023-03-28 22:43:31,405 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50901.0, num_to_drop=1, layers_to_drop={1}
+2023-03-28 22:43:33,506 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:44:24,105 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.705e+02 4.018e+02 4.822e+02 5.652e+02 1.134e+03, threshold=9.644e+02, percent-clipped=5.0
+2023-03-28 22:44:41,009 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-28 22:44:54,706 INFO [train.py:892] (3/4) Epoch 28, batch 850, loss[loss=0.1909, simple_loss=0.2689, pruned_loss=0.05644, over 19710.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2461, pruned_loss=0.04614, over 3894744.36 frames. ], batch size: 325, lr: 5.60e-03, grad_scale: 16.0
+2023-03-28 22:45:21,209 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:46:57,617 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6913, 3.1631, 3.5284, 3.0541, 3.8961, 3.8590, 4.5024, 5.0210],
+       device='cuda:3'), covar=tensor([0.0480, 0.1599, 0.1380, 0.2211, 0.1554, 0.1244, 0.0520, 0.0437],
+       device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0239, 0.0263, 0.0254, 0.0295, 0.0254, 0.0229, 0.0251],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 22:46:58,587 INFO [train.py:892] (3/4) Epoch 28, batch 900, loss[loss=0.1651, simple_loss=0.2463, pruned_loss=0.04196, over 19851.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.246, pruned_loss=0.04557, over 3905126.47 frames. ], batch size: 56, lr: 5.60e-03, grad_scale: 16.0
+2023-03-28 22:47:51,568 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51007.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:48:31,186 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.571e+02 3.452e+02 4.322e+02 5.014e+02 9.389e+02, threshold=8.643e+02, percent-clipped=0.0
+2023-03-28 22:49:04,829 INFO [train.py:892] (3/4) Epoch 28, batch 950, loss[loss=0.1496, simple_loss=0.2329, pruned_loss=0.03315, over 19669.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2461, pruned_loss=0.04533, over 3913510.99 frames. ], batch size: 58, lr: 5.60e-03, grad_scale: 16.0
+2023-03-28 22:49:50,098 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.9087, 1.8886, 1.9812, 1.8971, 1.9054, 1.9497, 1.8610, 1.9241],
+       device='cuda:3'), covar=tensor([0.0332, 0.0301, 0.0300, 0.0299, 0.0405, 0.0325, 0.0435, 0.0307],
+       device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0074, 0.0077, 0.0071, 0.0085, 0.0078, 0.0095, 0.0068],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 22:50:10,981 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51062.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:51:05,056 INFO [train.py:892] (3/4) Epoch 28, batch 1000, loss[loss=0.1606, simple_loss=0.2311, pruned_loss=0.04507, over 19797.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2452, pruned_loss=0.04526, over 3921266.54 frames. ], batch size: 162, lr: 5.60e-03, grad_scale: 16.0
+2023-03-28 22:52:34,540 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.807e+02 3.916e+02 4.760e+02 5.853e+02 1.174e+03, threshold=9.520e+02, percent-clipped=7.0
+2023-03-28 22:52:36,026 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51123.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:52:54,287 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1594, 2.9080, 3.1765, 2.8400, 3.4458, 3.3920, 4.0494, 4.4484],
+       device='cuda:3'), covar=tensor([0.0612, 0.1769, 0.1667, 0.2179, 0.1626, 0.1549, 0.0595, 0.0519],
+       device='cuda:3'), in_proj_covar=tensor([0.0249, 0.0239, 0.0263, 0.0252, 0.0294, 0.0252, 0.0228, 0.0249],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 22:53:06,050 INFO [train.py:892] (3/4) Epoch 28, batch 1050, loss[loss=0.1579, simple_loss=0.2354, pruned_loss=0.0402, over 19801.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2451, pruned_loss=0.04549, over 3928257.22 frames. ], batch size: 224, lr: 5.59e-03, grad_scale: 16.0
+2023-03-28 22:53:14,164 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51138.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:54:05,582 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.19 vs. limit=5.0
+2023-03-28 22:54:08,849 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51164.0, num_to_drop=1, layers_to_drop={2}
+2023-03-28 22:54:13,051 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0
+2023-03-28 22:54:25,389 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51172.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 22:55:00,669 INFO [train.py:892] (3/4) Epoch 28, batch 1100, loss[loss=0.1661, simple_loss=0.2381, pruned_loss=0.04706, over 19802.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2451, pruned_loss=0.04559, over 3933834.68 frames. ], batch size: 120, lr: 5.59e-03, grad_scale: 16.0
+2023-03-28 22:55:40,416 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:55:47,931 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6131, 3.2186, 3.7035, 2.8362, 3.9216, 3.1462, 3.4479, 3.6681],
+       device='cuda:3'), covar=tensor([0.0630, 0.0450, 0.0537, 0.0734, 0.0336, 0.0378, 0.0385, 0.0335],
+       device='cuda:3'), in_proj_covar=tensor([0.0074, 0.0082, 0.0081, 0.0107, 0.0076, 0.0079, 0.0076, 0.0069],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 22:55:56,285 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.5686, 5.9813, 6.0937, 5.9053, 5.7468, 5.7342, 5.8115, 5.6242],
+       device='cuda:3'), covar=tensor([0.1274, 0.1157, 0.0819, 0.1116, 0.0662, 0.0774, 0.1595, 0.1943],
+       device='cuda:3'), in_proj_covar=tensor([0.0294, 0.0325, 0.0366, 0.0293, 0.0272, 0.0274, 0.0353, 0.0389],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-28 22:56:33,422 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1245, 2.9020, 3.1115, 2.5323, 3.3708, 2.6750, 3.1296, 3.1467],
+       device='cuda:3'), covar=tensor([0.0477, 0.0540, 0.0640, 0.0797, 0.0322, 0.0480, 0.0457, 0.0362],
+       device='cuda:3'), in_proj_covar=tensor([0.0075, 0.0082, 0.0081, 0.0107, 0.0076, 0.0079, 0.0076, 0.0069],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 22:56:34,347 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.745e+02 3.704e+02 4.632e+02 5.921e+02 1.286e+03, threshold=9.265e+02, percent-clipped=1.0
+2023-03-28 22:56:57,588 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51233.0, num_to_drop=1, layers_to_drop={3}
+2023-03-28 22:57:02,696 INFO [train.py:892] (3/4) Epoch 28, batch 1150, loss[loss=0.1545, simple_loss=0.2246, pruned_loss=0.04219, over 19853.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.244, pruned_loss=0.04562, over 3938609.00 frames. ], batch size: 106, lr: 5.59e-03, grad_scale: 16.0
+2023-03-28 22:57:43,013 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 22:59:16,859 INFO [train.py:892] (3/4) Epoch 28, batch 1200, loss[loss=0.234, simple_loss=0.3063, pruned_loss=0.08085, over 19608.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2454, pruned_loss=0.04617, over 3941003.16 frames. ], batch size: 387, lr: 5.58e-03, grad_scale: 16.0
+2023-03-28 22:59:56,102 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:00:00,827 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.2803, 2.3487, 2.5125, 2.3024, 2.3422, 2.4275, 2.3855, 2.4581],
+       device='cuda:3'), covar=tensor([0.0320, 0.0282, 0.0267, 0.0314, 0.0424, 0.0304, 0.0409, 0.0319],
+       device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0075, 0.0078, 0.0072, 0.0086, 0.0078, 0.0095, 0.0068],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 23:00:45,070 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.551e+02 3.659e+02 4.255e+02 5.414e+02 8.940e+02, threshold=8.510e+02, percent-clipped=0.0
+2023-03-28 23:01:16,027 INFO [train.py:892] (3/4) Epoch 28, batch 1250, loss[loss=0.1572, simple_loss=0.2261, pruned_loss=0.04415, over 19802.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2442, pruned_loss=0.04546, over 3943899.14 frames. ], batch size: 151, lr: 5.58e-03, grad_scale: 16.0
+2023-03-28 23:02:51,232 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7461, 2.1739, 2.5752, 2.9124, 3.3849, 3.5656, 3.5012, 3.5051],
+       device='cuda:3'), covar=tensor([0.1007, 0.1752, 0.1327, 0.0748, 0.0443, 0.0298, 0.0426, 0.0411],
+       device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0170, 0.0177, 0.0150, 0.0134, 0.0129, 0.0121, 0.0114],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 23:03:06,466 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51382.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:03:13,792 INFO [train.py:892] (3/4) Epoch 28, batch 1300, loss[loss=0.165, simple_loss=0.2341, pruned_loss=0.04794, over 19875.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2463, pruned_loss=0.04639, over 3942373.47 frames. ], batch size: 159, lr: 5.58e-03, grad_scale: 16.0
+2023-03-28 23:03:31,669 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:04:35,798 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:04:45,690 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.672e+02 4.378e+02 5.169e+02 6.239e+02 1.127e+03, threshold=1.034e+03, percent-clipped=3.0
+2023-03-28 23:05:20,072 INFO [train.py:892] (3/4) Epoch 28, batch 1350, loss[loss=0.1742, simple_loss=0.2459, pruned_loss=0.0512, over 19800.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.248, pruned_loss=0.04745, over 3943785.38 frames. ], batch size: 150, lr: 5.58e-03, grad_scale: 16.0
+2023-03-28 23:05:25,336 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51438.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:05:36,867 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51443.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:05:59,688 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:06:25,743 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51464.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 23:07:16,854 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0
+2023-03-28 23:07:22,889 INFO [train.py:892] (3/4) Epoch 28, batch 1400, loss[loss=0.1819, simple_loss=0.2515, pruned_loss=0.05619, over 19778.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2485, pruned_loss=0.04754, over 3945243.03 frames. ], batch size: 193, lr: 5.57e-03, grad_scale: 16.0
+2023-03-28 23:07:23,996 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51486.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:07:44,184 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:08:23,616 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51512.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 23:08:48,563 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.195e+02 3.889e+02 4.620e+02 5.536e+02 9.901e+02, threshold=9.240e+02, percent-clipped=0.0
+2023-03-28 23:09:01,734 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51528.0, num_to_drop=1, layers_to_drop={3}
+2023-03-28 23:09:20,654 INFO [train.py:892] (3/4) Epoch 28, batch 1450, loss[loss=0.156, simple_loss=0.2304, pruned_loss=0.04079, over 19821.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2473, pruned_loss=0.04666, over 3947148.05 frames. ], batch size: 103, lr: 5.57e-03, grad_scale: 16.0
+2023-03-28 23:09:30,808 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0569, 4.8719, 5.4597, 4.8976, 4.3620, 5.2475, 5.0737, 5.6438],
+       device='cuda:3'), covar=tensor([0.0864, 0.0353, 0.0368, 0.0383, 0.0768, 0.0450, 0.0429, 0.0308],
+       device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0220, 0.0220, 0.0233, 0.0205, 0.0238, 0.0231, 0.0213],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 23:10:03,562 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0
+2023-03-28 23:10:14,063 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:10:53,453 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0
+2023-03-28 23:11:27,606 INFO [train.py:892] (3/4) Epoch 28, batch 1500, loss[loss=0.1405, simple_loss=0.22, pruned_loss=0.03053, over 19746.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2462, pruned_loss=0.04595, over 3948628.90 frames. ], batch size: 89, lr: 5.57e-03, grad_scale: 16.0
+2023-03-28 23:12:07,070 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:13:00,848 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.657e+02 3.781e+02 4.719e+02 5.439e+02 9.463e+02, threshold=9.439e+02, percent-clipped=1.0
+2023-03-28 23:13:30,305 INFO [train.py:892] (3/4) Epoch 28, batch 1550, loss[loss=0.1857, simple_loss=0.2508, pruned_loss=0.06029, over 19799.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2453, pruned_loss=0.04576, over 3948901.46 frames. ], batch size: 172, lr: 5.57e-03, grad_scale: 16.0
+2023-03-28 23:14:07,809 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:14:39,380 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5089, 3.3098, 3.3418, 3.5464, 3.3808, 3.4808, 3.6165, 3.7849],
+       device='cuda:3'), covar=tensor([0.0711, 0.0516, 0.0650, 0.0468, 0.0734, 0.0624, 0.0494, 0.0363],
+       device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0172, 0.0198, 0.0171, 0.0168, 0.0153, 0.0146, 0.0194],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 23:15:34,965 INFO [train.py:892] (3/4) Epoch 28, batch 1600, loss[loss=0.1695, simple_loss=0.2513, pruned_loss=0.04383, over 19795.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2451, pruned_loss=0.04528, over 3948993.56 frames. ], batch size: 185, lr: 5.56e-03, grad_scale: 16.0
+2023-03-28 23:16:54,083 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:17:03,723 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.853e+02 3.747e+02 4.417e+02 5.191e+02 1.106e+03, threshold=8.834e+02, percent-clipped=1.0
+2023-03-28 23:17:34,746 INFO [train.py:892] (3/4) Epoch 28, batch 1650, loss[loss=0.1463, simple_loss=0.2229, pruned_loss=0.03488, over 19773.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2438, pruned_loss=0.04479, over 3950503.11 frames. ], batch size: 91, lr: 5.56e-03, grad_scale: 16.0
+2023-03-28 23:17:40,885 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:18:01,036 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:18:03,540 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:18:51,242 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:19:37,714 INFO [train.py:892] (3/4) Epoch 28, batch 1700, loss[loss=0.162, simple_loss=0.2301, pruned_loss=0.04697, over 19774.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.243, pruned_loss=0.04419, over 3951571.01 frames. ], batch size: 130, lr: 5.56e-03, grad_scale: 16.0
+2023-03-28 23:20:14,467 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4832, 4.2656, 4.2827, 4.5438, 4.2600, 4.6077, 4.6445, 4.7862],
+       device='cuda:3'), covar=tensor([0.0572, 0.0359, 0.0444, 0.0303, 0.0644, 0.0415, 0.0387, 0.0290],
+       device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0173, 0.0199, 0.0172, 0.0170, 0.0155, 0.0147, 0.0195],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-28 23:20:27,742 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0
+2023-03-28 23:20:36,137 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51808.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:20:55,456 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.3942, 4.9012, 4.9847, 4.7173, 5.2617, 3.1633, 4.1339, 2.4938],
+       device='cuda:3'), covar=tensor([0.0162, 0.0175, 0.0133, 0.0187, 0.0127, 0.1005, 0.0964, 0.1647],
+       device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0144, 0.0113, 0.0134, 0.0119, 0.0135, 0.0142, 0.0128],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 23:21:06,663 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.935e+02 3.915e+02 4.350e+02 5.709e+02 1.199e+03, threshold=8.700e+02, percent-clipped=1.0
+2023-03-28 23:21:17,322 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51828.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 23:21:29,939 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9589, 2.7909, 3.1073, 2.7869, 3.3183, 3.2158, 3.8136, 4.1717],
+       device='cuda:3'), covar=tensor([0.0571, 0.1660, 0.1434, 0.2019, 0.1596, 0.1471, 0.0600, 0.0531],
+       device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0237, 0.0264, 0.0251, 0.0293, 0.0252, 0.0227, 0.0249],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 23:21:33,161 INFO [train.py:892] (3/4) Epoch 28, batch 1750, loss[loss=0.1435, simple_loss=0.2242, pruned_loss=0.03143, over 19640.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2429, pruned_loss=0.04443, over 3951841.29 frames. ], batch size: 72, lr: 5.55e-03, grad_scale: 16.0
+2023-03-28 23:22:06,313 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:22:22,914 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1121, 3.0473, 4.4975, 3.3696, 3.6855, 3.5317, 2.4440, 2.6758],
+       device='cuda:3'), covar=tensor([0.0934, 0.3093, 0.0508, 0.1134, 0.1684, 0.1438, 0.2575, 0.2675],
+       device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0385, 0.0344, 0.0283, 0.0371, 0.0370, 0.0369, 0.0338],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 23:22:53,572 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51876.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 23:23:13,531 INFO [train.py:892] (3/4) Epoch 28, batch 1800, loss[loss=0.1842, simple_loss=0.256, pruned_loss=0.05618, over 19642.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2434, pruned_loss=0.0447, over 3951325.69 frames. ], batch size: 299, lr: 5.55e-03, grad_scale: 16.0
+2023-03-28 23:23:19,232 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.80 vs. limit=5.0
+2023-03-28 23:23:41,874 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0108, 3.2296, 3.2981, 3.1885, 3.0619, 3.2721, 3.0354, 3.2696],
+       device='cuda:3'), covar=tensor([0.0226, 0.0312, 0.0272, 0.0253, 0.0363, 0.0222, 0.0371, 0.0322],
+       device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0076, 0.0079, 0.0073, 0.0087, 0.0079, 0.0096, 0.0069],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 23:23:50,514 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3786, 3.6883, 3.8613, 4.4527, 3.0178, 3.3391, 2.7913, 2.7124],
+       device='cuda:3'), covar=tensor([0.0528, 0.1913, 0.0841, 0.0380, 0.1938, 0.0949, 0.1279, 0.1668],
+       device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0332, 0.0246, 0.0199, 0.0246, 0.0206, 0.0215, 0.0215],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 23:24:26,531 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.458e+02 3.828e+02 4.368e+02 5.479e+02 1.522e+03, threshold=8.737e+02, percent-clipped=1.0
+2023-03-28 23:24:33,228 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.2607, 4.8122, 4.8842, 4.6571, 5.1875, 3.2736, 4.1619, 2.6492],
+       device='cuda:3'), covar=tensor([0.0213, 0.0204, 0.0165, 0.0196, 0.0153, 0.0950, 0.1016, 0.1510],
+       device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0144, 0.0113, 0.0134, 0.0119, 0.0134, 0.0142, 0.0128],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 23:24:52,087 INFO [train.py:892] (3/4) Epoch 28, batch 1850, loss[loss=0.1757, simple_loss=0.263, pruned_loss=0.04419, over 19680.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2449, pruned_loss=0.04435, over 3949956.44 frames. ], batch size: 55, lr: 5.55e-03, grad_scale: 16.0
+2023-03-28 23:25:59,286 INFO [train.py:892] (3/4) Epoch 29, batch 0, loss[loss=0.1552, simple_loss=0.2248, pruned_loss=0.04276, over 19761.00 frames. ], tot_loss[loss=0.1552, simple_loss=0.2248, pruned_loss=0.04276, over 19761.00 frames. ], batch size: 152, lr: 5.45e-03, grad_scale: 16.0
+2023-03-28 23:25:59,287 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-28 23:26:37,599 INFO [train.py:926] (3/4) Epoch 29, validation: loss=0.1782, simple_loss=0.2489, pruned_loss=0.05378, over 2883724.00 frames. 
+2023-03-28 23:26:37,600 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-28 23:28:41,200 INFO [train.py:892] (3/4) Epoch 29, batch 50, loss[loss=0.2784, simple_loss=0.3418, pruned_loss=0.1075, over 19204.00 frames. ], tot_loss[loss=0.163, simple_loss=0.2395, pruned_loss=0.04326, over 891214.89 frames. ], batch size: 452, lr: 5.45e-03, grad_scale: 16.0
+2023-03-28 23:29:49,636 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.75 vs. limit=5.0
+2023-03-28 23:30:05,630 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.398e+02 3.680e+02 4.499e+02 5.217e+02 8.656e+02, threshold=8.998e+02, percent-clipped=0.0
+2023-03-28 23:30:16,638 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5173, 2.0594, 2.4392, 2.7523, 3.2658, 3.3007, 3.3105, 3.2779],
+       device='cuda:3'), covar=tensor([0.1080, 0.1869, 0.1420, 0.0815, 0.0490, 0.0390, 0.0457, 0.0511],
+       device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0171, 0.0178, 0.0151, 0.0135, 0.0130, 0.0122, 0.0115],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 23:30:44,871 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52038.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:30:49,313 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:30:50,314 INFO [train.py:892] (3/4) Epoch 29, batch 100, loss[loss=0.1316, simple_loss=0.2117, pruned_loss=0.02574, over 19824.00 frames. ], tot_loss[loss=0.1619, simple_loss=0.2391, pruned_loss=0.04237, over 1570148.01 frames. ], batch size: 57, lr: 5.45e-03, grad_scale: 16.0
+2023-03-28 23:31:10,330 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:32:05,867 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2000, 3.0388, 4.9719, 4.2603, 4.7456, 4.9234, 4.8689, 4.4637],
+       device='cuda:3'), covar=tensor([0.0394, 0.0847, 0.0088, 0.0852, 0.0122, 0.0168, 0.0131, 0.0155],
+       device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0102, 0.0086, 0.0153, 0.0084, 0.0097, 0.0090, 0.0086],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 23:32:44,176 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52086.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:32:54,795 INFO [train.py:892] (3/4) Epoch 29, batch 150, loss[loss=0.1452, simple_loss=0.2128, pruned_loss=0.03876, over 19834.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2413, pruned_loss=0.04372, over 2096901.37 frames. ], batch size: 127, lr: 5.44e-03, grad_scale: 16.0
+2023-03-28 23:33:10,167 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:33:23,283 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:33:27,872 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:34:15,962 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.333e+02 4.011e+02 4.815e+02 5.907e+02 9.716e+02, threshold=9.631e+02, percent-clipped=1.0
+2023-03-28 23:34:59,547 INFO [train.py:892] (3/4) Epoch 29, batch 200, loss[loss=0.1762, simple_loss=0.2564, pruned_loss=0.04796, over 19749.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2407, pruned_loss=0.04336, over 2508573.53 frames. ], batch size: 110, lr: 5.44e-03, grad_scale: 16.0
+2023-03-28 23:35:24,857 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:35:25,565 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.58 vs. limit=5.0
+2023-03-28 23:36:59,115 INFO [train.py:892] (3/4) Epoch 29, batch 250, loss[loss=0.1735, simple_loss=0.2599, pruned_loss=0.04356, over 19729.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2427, pruned_loss=0.04395, over 2827728.90 frames. ], batch size: 50, lr: 5.44e-03, grad_scale: 16.0
+2023-03-28 23:37:05,240 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52193.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:37:15,918 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0
+2023-03-28 23:37:17,888 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:37:39,094 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5426, 2.7356, 2.6922, 2.6906, 2.5435, 2.7213, 2.4958, 2.7041],
+       device='cuda:3'), covar=tensor([0.0350, 0.0321, 0.0330, 0.0276, 0.0412, 0.0284, 0.0409, 0.0401],
+       device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0076, 0.0080, 0.0073, 0.0088, 0.0080, 0.0097, 0.0069],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 23:38:17,956 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.564e+02 3.825e+02 4.477e+02 5.151e+02 8.477e+02, threshold=8.954e+02, percent-clipped=0.0
+2023-03-28 23:38:54,112 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2574, 2.6766, 3.4050, 3.3660, 3.8866, 4.3764, 4.1425, 4.3051],
+       device='cuda:3'), covar=tensor([0.0765, 0.1628, 0.1045, 0.0616, 0.0344, 0.0248, 0.0374, 0.0390],
+       device='cuda:3'), in_proj_covar=tensor([0.0157, 0.0169, 0.0176, 0.0149, 0.0132, 0.0128, 0.0120, 0.0113],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-28 23:39:01,957 INFO [train.py:892] (3/4) Epoch 29, batch 300, loss[loss=0.1763, simple_loss=0.2566, pruned_loss=0.04798, over 19698.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2424, pruned_loss=0.04363, over 3077176.02 frames. ], batch size: 265, lr: 5.44e-03, grad_scale: 16.0
+2023-03-28 23:39:33,543 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7702, 4.5142, 4.5731, 4.3359, 4.7583, 3.0872, 3.9151, 2.4759],
+       device='cuda:3'), covar=tensor([0.0191, 0.0214, 0.0137, 0.0181, 0.0142, 0.0979, 0.0728, 0.1328],
+       device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0145, 0.0114, 0.0134, 0.0119, 0.0135, 0.0143, 0.0128],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 23:39:35,652 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52254.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:41:06,907 INFO [train.py:892] (3/4) Epoch 29, batch 350, loss[loss=0.261, simple_loss=0.3294, pruned_loss=0.09628, over 19182.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2432, pruned_loss=0.0438, over 3269716.10 frames. ], batch size: 452, lr: 5.43e-03, grad_scale: 8.0
+2023-03-28 23:42:02,668 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52312.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:42:07,890 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-28 23:42:20,573 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.54 vs. limit=5.0
+2023-03-28 23:42:30,783 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.445e+02 3.965e+02 4.571e+02 5.589e+02 1.011e+03, threshold=9.142e+02, percent-clipped=1.0
+2023-03-28 23:43:02,912 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.7043, 6.0328, 6.0888, 5.9492, 5.7196, 6.0430, 5.3951, 5.4857],
+       device='cuda:3'), covar=tensor([0.0426, 0.0417, 0.0424, 0.0452, 0.0513, 0.0512, 0.0742, 0.0923],
+       device='cuda:3'), in_proj_covar=tensor([0.0263, 0.0282, 0.0294, 0.0256, 0.0260, 0.0245, 0.0264, 0.0309],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 23:43:04,923 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4970, 2.5025, 1.6273, 2.7562, 2.5537, 2.6602, 2.7433, 2.2415],
+       device='cuda:3'), covar=tensor([0.0691, 0.0759, 0.1436, 0.0634, 0.0697, 0.0567, 0.0623, 0.0974],
+       device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0143, 0.0143, 0.0150, 0.0133, 0.0133, 0.0147, 0.0144],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 23:43:07,784 INFO [train.py:892] (3/4) Epoch 29, batch 400, loss[loss=0.1515, simple_loss=0.2334, pruned_loss=0.0348, over 19889.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2429, pruned_loss=0.04394, over 3421105.51 frames. ], batch size: 63, lr: 5.43e-03, grad_scale: 8.0
+2023-03-28 23:44:27,747 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52373.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:45:13,547 INFO [train.py:892] (3/4) Epoch 29, batch 450, loss[loss=0.1642, simple_loss=0.2401, pruned_loss=0.04413, over 19811.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2425, pruned_loss=0.04417, over 3538792.91 frames. ], batch size: 72, lr: 5.43e-03, grad_scale: 8.0
+2023-03-28 23:45:28,510 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:45:46,535 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52403.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:46:29,424 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9732, 4.5946, 4.6753, 4.4417, 4.8826, 3.1568, 4.0444, 2.6800],
+       device='cuda:3'), covar=tensor([0.0140, 0.0193, 0.0129, 0.0189, 0.0140, 0.0976, 0.0797, 0.1310],
+       device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0144, 0.0113, 0.0133, 0.0119, 0.0134, 0.0142, 0.0127],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 23:46:38,513 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.695e+02 3.819e+02 4.668e+02 5.658e+02 1.148e+03, threshold=9.335e+02, percent-clipped=4.0
+2023-03-28 23:46:42,594 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.52 vs. limit=2.0
+2023-03-28 23:47:02,967 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3756, 2.6140, 4.4216, 3.8591, 4.2009, 4.3399, 4.1709, 4.0450],
+       device='cuda:3'), covar=tensor([0.0517, 0.0979, 0.0104, 0.0719, 0.0142, 0.0213, 0.0174, 0.0173],
+       device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0102, 0.0087, 0.0154, 0.0084, 0.0098, 0.0090, 0.0086],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 23:47:22,589 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8484, 3.8580, 2.3651, 4.1020, 4.2181, 1.9327, 3.4962, 3.3491],
+       device='cuda:3'), covar=tensor([0.0683, 0.0821, 0.2779, 0.0735, 0.0585, 0.2857, 0.1087, 0.0786],
+       device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0256, 0.0229, 0.0272, 0.0252, 0.0203, 0.0238, 0.0195],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 23:47:23,561 INFO [train.py:892] (3/4) Epoch 29, batch 500, loss[loss=0.1594, simple_loss=0.2261, pruned_loss=0.04634, over 19781.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2429, pruned_loss=0.04442, over 3628070.74 frames. ], batch size: 131, lr: 5.43e-03, grad_scale: 8.0
+2023-03-28 23:47:50,896 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52451.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:49:23,626 INFO [train.py:892] (3/4) Epoch 29, batch 550, loss[loss=0.1694, simple_loss=0.2402, pruned_loss=0.04932, over 19758.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2425, pruned_loss=0.0441, over 3700727.29 frames. ], batch size: 209, lr: 5.42e-03, grad_scale: 8.0
+2023-03-28 23:50:29,904 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2432, 4.3297, 2.5420, 4.5838, 4.7711, 2.0357, 3.8732, 3.4553],
+       device='cuda:3'), covar=tensor([0.0694, 0.0845, 0.2772, 0.0805, 0.0525, 0.2971, 0.1125, 0.0929],
+       device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0255, 0.0230, 0.0273, 0.0252, 0.0204, 0.0238, 0.0196],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-28 23:50:40,696 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4695, 4.2956, 4.8000, 4.3602, 3.9921, 4.5762, 4.4425, 4.8892],
+       device='cuda:3'), covar=tensor([0.0805, 0.0387, 0.0367, 0.0407, 0.1019, 0.0529, 0.0560, 0.0342],
+       device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0221, 0.0220, 0.0233, 0.0206, 0.0238, 0.0230, 0.0212],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-28 23:50:45,599 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.600e+02 4.376e+02 5.079e+02 6.009e+02 1.236e+03, threshold=1.016e+03, percent-clipped=4.0
+2023-03-28 23:51:25,986 INFO [train.py:892] (3/4) Epoch 29, batch 600, loss[loss=0.1383, simple_loss=0.2114, pruned_loss=0.03258, over 19904.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2434, pruned_loss=0.04467, over 3756704.69 frames. ], batch size: 116, lr: 5.42e-03, grad_scale: 8.0
+2023-03-28 23:51:36,736 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0
+2023-03-28 23:51:48,683 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:52:05,069 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52555.0, num_to_drop=1, layers_to_drop={0}
+2023-03-28 23:53:25,812 INFO [train.py:892] (3/4) Epoch 29, batch 650, loss[loss=0.1515, simple_loss=0.2318, pruned_loss=0.03559, over 19779.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2429, pruned_loss=0.04433, over 3798696.91 frames. ], batch size: 91, lr: 5.42e-03, grad_scale: 8.0
+2023-03-28 23:54:27,212 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52616.0, num_to_drop=1, layers_to_drop={3}
+2023-03-28 23:54:45,473 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.294e+02 3.664e+02 4.679e+02 5.983e+02 1.020e+03, threshold=9.358e+02, percent-clipped=1.0
+2023-03-28 23:55:31,369 INFO [train.py:892] (3/4) Epoch 29, batch 700, loss[loss=0.1965, simple_loss=0.2781, pruned_loss=0.05742, over 19687.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2427, pruned_loss=0.04422, over 3833764.04 frames. ], batch size: 337, lr: 5.41e-03, grad_scale: 8.0
+2023-03-28 23:55:56,571 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.63 vs. limit=5.0
+2023-03-28 23:56:38,833 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0
+2023-03-28 23:56:40,297 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:57:32,026 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.81 vs. limit=5.0
+2023-03-28 23:57:34,771 INFO [train.py:892] (3/4) Epoch 29, batch 750, loss[loss=0.1533, simple_loss=0.222, pruned_loss=0.0423, over 19866.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2428, pruned_loss=0.04403, over 3859538.83 frames. ], batch size: 154, lr: 5.41e-03, grad_scale: 8.0
+2023-03-28 23:57:48,686 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-28 23:58:51,488 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.283e+02 3.891e+02 4.308e+02 5.320e+02 9.738e+02, threshold=8.617e+02, percent-clipped=1.0
+2023-03-28 23:59:16,901 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-28 23:59:30,951 INFO [train.py:892] (3/4) Epoch 29, batch 800, loss[loss=0.1541, simple_loss=0.2336, pruned_loss=0.03728, over 19796.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2432, pruned_loss=0.04404, over 3879586.22 frames. ], batch size: 40, lr: 5.41e-03, grad_scale: 8.0
+2023-03-28 23:59:40,268 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:00:01,579 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:01:37,824 INFO [train.py:892] (3/4) Epoch 29, batch 850, loss[loss=0.1631, simple_loss=0.2384, pruned_loss=0.04387, over 19764.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2422, pruned_loss=0.04329, over 3895939.28 frames. ], batch size: 244, lr: 5.41e-03, grad_scale: 8.0
+2023-03-29 00:01:46,662 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8497, 2.2895, 2.7046, 3.1095, 3.6110, 3.7466, 3.6912, 3.7545],
+       device='cuda:3'), covar=tensor([0.0887, 0.1637, 0.1275, 0.0649, 0.0357, 0.0264, 0.0367, 0.0379],
+       device='cuda:3'), in_proj_covar=tensor([0.0156, 0.0168, 0.0175, 0.0148, 0.0131, 0.0128, 0.0120, 0.0112],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 00:02:06,240 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.12 vs. limit=2.0
+2023-03-29 00:02:27,888 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-29 00:02:32,690 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8889, 3.7936, 4.1621, 3.7648, 3.5847, 4.0211, 3.8570, 4.2231],
+       device='cuda:3'), covar=tensor([0.0799, 0.0346, 0.0343, 0.0407, 0.1147, 0.0543, 0.0484, 0.0364],
+       device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0220, 0.0220, 0.0232, 0.0206, 0.0239, 0.0230, 0.0212],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 00:02:32,828 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:02:58,931 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.946e+02 3.447e+02 4.095e+02 4.667e+02 7.447e+02, threshold=8.190e+02, percent-clipped=0.0
+2023-03-29 00:03:38,965 INFO [train.py:892] (3/4) Epoch 29, batch 900, loss[loss=0.172, simple_loss=0.2617, pruned_loss=0.04117, over 19655.00 frames. ], tot_loss[loss=0.163, simple_loss=0.2406, pruned_loss=0.04266, over 3908922.15 frames. ], batch size: 57, lr: 5.40e-03, grad_scale: 8.0
+2023-03-29 00:03:58,721 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:04:06,989 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.5090, 5.8018, 5.8718, 5.7602, 5.5623, 5.8358, 5.1208, 5.2801],
+       device='cuda:3'), covar=tensor([0.0434, 0.0449, 0.0464, 0.0395, 0.0541, 0.0493, 0.0741, 0.1028],
+       device='cuda:3'), in_proj_covar=tensor([0.0263, 0.0280, 0.0292, 0.0254, 0.0258, 0.0244, 0.0263, 0.0307],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 00:05:35,947 INFO [train.py:892] (3/4) Epoch 29, batch 950, loss[loss=0.1741, simple_loss=0.2482, pruned_loss=0.04997, over 19888.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2419, pruned_loss=0.04301, over 3916438.59 frames. ], batch size: 176, lr: 5.40e-03, grad_scale: 8.0
+2023-03-29 00:05:53,380 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:06:28,128 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52911.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 00:06:58,669 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.525e+02 3.679e+02 4.624e+02 5.272e+02 1.115e+03, threshold=9.248e+02, percent-clipped=1.0
+2023-03-29 00:07:29,069 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.8189, 5.9541, 6.0149, 5.9863, 5.7815, 5.9804, 5.3209, 5.1032],
+       device='cuda:3'), covar=tensor([0.0637, 0.0791, 0.0750, 0.0617, 0.0808, 0.0791, 0.1179, 0.2018],
+       device='cuda:3'), in_proj_covar=tensor([0.0261, 0.0278, 0.0290, 0.0253, 0.0257, 0.0242, 0.0261, 0.0305],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 00:07:41,657 INFO [train.py:892] (3/4) Epoch 29, batch 1000, loss[loss=0.1701, simple_loss=0.243, pruned_loss=0.04859, over 19801.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2435, pruned_loss=0.04394, over 3923570.93 frames. ], batch size: 173, lr: 5.40e-03, grad_scale: 8.0
+2023-03-29 00:07:42,968 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.87 vs. limit=5.0
+2023-03-29 00:08:46,584 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52968.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:09:39,927 INFO [train.py:892] (3/4) Epoch 29, batch 1050, loss[loss=0.1629, simple_loss=0.2379, pruned_loss=0.04398, over 19764.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.2441, pruned_loss=0.04414, over 3927406.26 frames. ], batch size: 244, lr: 5.40e-03, grad_scale: 8.0
+2023-03-29 00:10:17,077 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4036, 2.5066, 2.6671, 2.4120, 2.8426, 2.8267, 3.2409, 3.5416],
+       device='cuda:3'), covar=tensor([0.0797, 0.1783, 0.1903, 0.2267, 0.1691, 0.1545, 0.0790, 0.0760],
+       device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0240, 0.0267, 0.0254, 0.0296, 0.0256, 0.0231, 0.0253],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 00:10:42,610 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53016.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:10:42,775 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0694, 3.9876, 3.9614, 3.7065, 4.1029, 2.9324, 3.3622, 1.8421],
+       device='cuda:3'), covar=tensor([0.0251, 0.0232, 0.0178, 0.0228, 0.0204, 0.1055, 0.0783, 0.1843],
+       device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0145, 0.0114, 0.0134, 0.0119, 0.0135, 0.0142, 0.0128],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 00:10:59,360 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.507e+02 3.902e+02 4.763e+02 5.659e+02 9.296e+02, threshold=9.526e+02, percent-clipped=1.0
+2023-03-29 00:11:39,414 INFO [train.py:892] (3/4) Epoch 29, batch 1100, loss[loss=0.1753, simple_loss=0.2612, pruned_loss=0.04467, over 19780.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2454, pruned_loss=0.04434, over 3930839.12 frames. ], batch size: 53, lr: 5.39e-03, grad_scale: 8.0
+2023-03-29 00:12:17,997 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:12:20,541 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8025, 3.3456, 3.6886, 3.4055, 4.0181, 4.0565, 4.5596, 5.1344],
+       device='cuda:3'), covar=tensor([0.0543, 0.1625, 0.1519, 0.1946, 0.1670, 0.1278, 0.0594, 0.0516],
+       device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0239, 0.0266, 0.0254, 0.0295, 0.0255, 0.0230, 0.0252],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 00:13:40,590 INFO [train.py:892] (3/4) Epoch 29, batch 1150, loss[loss=0.2467, simple_loss=0.3241, pruned_loss=0.08465, over 19476.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2459, pruned_loss=0.04494, over 3934201.69 frames. ], batch size: 396, lr: 5.39e-03, grad_scale: 8.0
+2023-03-29 00:14:25,724 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:14:43,948 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.15 vs. limit=5.0
+2023-03-29 00:14:48,384 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53117.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:15:03,919 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.770e+02 4.130e+02 4.907e+02 5.867e+02 8.841e+02, threshold=9.813e+02, percent-clipped=0.0
+2023-03-29 00:15:41,766 INFO [train.py:892] (3/4) Epoch 29, batch 1200, loss[loss=0.1316, simple_loss=0.2098, pruned_loss=0.02668, over 19846.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2462, pruned_loss=0.04507, over 3935985.17 frames. ], batch size: 112, lr: 5.39e-03, grad_scale: 8.0
+2023-03-29 00:15:56,026 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6358, 4.0353, 4.1033, 4.9047, 3.1662, 3.5802, 3.2174, 2.9476],
+       device='cuda:3'), covar=tensor([0.0483, 0.1697, 0.0808, 0.0272, 0.1945, 0.0977, 0.1158, 0.1614],
+       device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0330, 0.0244, 0.0199, 0.0245, 0.0205, 0.0214, 0.0214],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 00:15:57,891 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6892, 4.3142, 4.4027, 4.6142, 4.4263, 4.7684, 4.7650, 4.9984],
+       device='cuda:3'), covar=tensor([0.0589, 0.0414, 0.0518, 0.0378, 0.0687, 0.0491, 0.0428, 0.0280],
+       device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0174, 0.0201, 0.0174, 0.0171, 0.0157, 0.0149, 0.0196],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 00:17:43,519 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0
+2023-03-29 00:17:48,630 INFO [train.py:892] (3/4) Epoch 29, batch 1250, loss[loss=0.15, simple_loss=0.2273, pruned_loss=0.03637, over 19716.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2453, pruned_loss=0.04493, over 3938746.77 frames. ], batch size: 61, lr: 5.39e-03, grad_scale: 8.0
+2023-03-29 00:18:19,754 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0
+2023-03-29 00:18:39,143 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53211.0, num_to_drop=1, layers_to_drop={2}
+2023-03-29 00:18:44,097 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2813, 2.8812, 3.2189, 2.8769, 3.5283, 3.4666, 4.0971, 4.5319],
+       device='cuda:3'), covar=tensor([0.0570, 0.1772, 0.1496, 0.2243, 0.1536, 0.1428, 0.0599, 0.0551],
+       device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0237, 0.0263, 0.0251, 0.0292, 0.0253, 0.0228, 0.0249],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 00:19:09,956 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.614e+02 3.724e+02 4.127e+02 5.069e+02 9.146e+02, threshold=8.253e+02, percent-clipped=0.0
+2023-03-29 00:19:52,126 INFO [train.py:892] (3/4) Epoch 29, batch 1300, loss[loss=0.169, simple_loss=0.2571, pruned_loss=0.04049, over 19532.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2442, pruned_loss=0.0445, over 3942738.42 frames. ], batch size: 54, lr: 5.38e-03, grad_scale: 8.0
+2023-03-29 00:19:53,191 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:20:39,662 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53259.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 00:21:58,591 INFO [train.py:892] (3/4) Epoch 29, batch 1350, loss[loss=0.1563, simple_loss=0.2273, pruned_loss=0.04268, over 19781.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2442, pruned_loss=0.04445, over 3943238.77 frames. ], batch size: 163, lr: 5.38e-03, grad_scale: 8.0
+2023-03-29 00:22:26,043 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53302.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:22:30,776 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:22:59,219 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53315.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:23:19,523 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.794e+02 3.937e+02 4.567e+02 5.405e+02 8.872e+02, threshold=9.134e+02, percent-clipped=4.0
+2023-03-29 00:24:04,144 INFO [train.py:892] (3/4) Epoch 29, batch 1400, loss[loss=0.1528, simple_loss=0.2352, pruned_loss=0.03513, over 19822.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.2438, pruned_loss=0.04428, over 3943910.91 frames. ], batch size: 147, lr: 5.38e-03, grad_scale: 8.0
+2023-03-29 00:24:37,519 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.1719, 5.4747, 5.5200, 5.4148, 5.1556, 5.4744, 4.9623, 4.9552],
+       device='cuda:3'), covar=tensor([0.0441, 0.0418, 0.0475, 0.0396, 0.0624, 0.0483, 0.0648, 0.0987],
+       device='cuda:3'), in_proj_covar=tensor([0.0265, 0.0283, 0.0292, 0.0257, 0.0260, 0.0246, 0.0264, 0.0309],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 00:25:04,773 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53365.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:25:09,311 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53367.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:25:31,049 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53376.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:26:04,805 INFO [train.py:892] (3/4) Epoch 29, batch 1450, loss[loss=0.1674, simple_loss=0.2376, pruned_loss=0.04859, over 19687.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2434, pruned_loss=0.04383, over 3945194.10 frames. ], batch size: 64, lr: 5.38e-03, grad_scale: 8.0
+2023-03-29 00:26:49,003 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:26:52,469 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2198, 3.8897, 4.0068, 4.2130, 3.9410, 4.2769, 4.2645, 4.4780],
+       device='cuda:3'), covar=tensor([0.0577, 0.0445, 0.0486, 0.0361, 0.0704, 0.0476, 0.0396, 0.0308],
+       device='cuda:3'), in_proj_covar=tensor([0.0149, 0.0173, 0.0199, 0.0173, 0.0170, 0.0155, 0.0149, 0.0195],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 00:26:59,423 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53412.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:27:27,178 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.651e+02 3.715e+02 4.371e+02 5.322e+02 9.816e+02, threshold=8.743e+02, percent-clipped=2.0
+2023-03-29 00:27:38,960 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53428.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 00:28:09,479 INFO [train.py:892] (3/4) Epoch 29, batch 1500, loss[loss=0.1482, simple_loss=0.2218, pruned_loss=0.03726, over 19833.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2436, pruned_loss=0.04402, over 3946189.46 frames. ], batch size: 101, lr: 5.37e-03, grad_scale: 8.0
+2023-03-29 00:28:49,111 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:29:35,216 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.44 vs. limit=5.0
+2023-03-29 00:30:14,161 INFO [train.py:892] (3/4) Epoch 29, batch 1550, loss[loss=0.2104, simple_loss=0.2837, pruned_loss=0.06855, over 19633.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2432, pruned_loss=0.04364, over 3946654.82 frames. ], batch size: 359, lr: 5.37e-03, grad_scale: 8.0
+2023-03-29 00:31:28,849 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.518e+02 3.925e+02 4.450e+02 5.196e+02 1.158e+03, threshold=8.900e+02, percent-clipped=2.0
+2023-03-29 00:32:15,159 INFO [train.py:892] (3/4) Epoch 29, batch 1600, loss[loss=0.2271, simple_loss=0.3045, pruned_loss=0.07481, over 19558.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.244, pruned_loss=0.04369, over 3947404.79 frames. ], batch size: 376, lr: 5.37e-03, grad_scale: 8.0
+2023-03-29 00:33:32,289 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1110, 2.8166, 5.0578, 4.2860, 4.7935, 4.9264, 4.8213, 4.5468],
+       device='cuda:3'), covar=tensor([0.0419, 0.0947, 0.0087, 0.0917, 0.0133, 0.0208, 0.0146, 0.0148],
+       device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0102, 0.0087, 0.0154, 0.0085, 0.0097, 0.0090, 0.0085],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 00:34:19,766 INFO [train.py:892] (3/4) Epoch 29, batch 1650, loss[loss=0.2109, simple_loss=0.3099, pruned_loss=0.05596, over 18737.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2431, pruned_loss=0.04366, over 3947684.60 frames. ], batch size: 564, lr: 5.37e-03, grad_scale: 8.0
+2023-03-29 00:34:36,545 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53597.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:34:38,920 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9698, 4.7910, 5.3908, 4.9454, 4.3556, 5.1733, 5.0521, 5.5738],
+       device='cuda:3'), covar=tensor([0.0821, 0.0326, 0.0342, 0.0341, 0.0781, 0.0440, 0.0393, 0.0283],
+       device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0224, 0.0224, 0.0236, 0.0209, 0.0242, 0.0233, 0.0215],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 00:35:41,050 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.655e+02 3.870e+02 4.535e+02 5.487e+02 1.255e+03, threshold=9.070e+02, percent-clipped=2.0
+2023-03-29 00:36:23,488 INFO [train.py:892] (3/4) Epoch 29, batch 1700, loss[loss=0.1623, simple_loss=0.2454, pruned_loss=0.03964, over 19653.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2434, pruned_loss=0.0438, over 3947491.28 frames. ], batch size: 47, lr: 5.36e-03, grad_scale: 8.0
+2023-03-29 00:37:04,950 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-29 00:37:12,304 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53660.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:37:27,089 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0
+2023-03-29 00:37:40,616 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53671.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:38:27,042 INFO [train.py:892] (3/4) Epoch 29, batch 1750, loss[loss=0.1559, simple_loss=0.234, pruned_loss=0.03894, over 19778.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2424, pruned_loss=0.04371, over 3947647.87 frames. ], batch size: 87, lr: 5.36e-03, grad_scale: 8.0
+2023-03-29 00:39:14,860 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53712.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:39:37,689 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53723.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 00:39:38,985 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.549e+02 3.643e+02 4.593e+02 5.639e+02 1.111e+03, threshold=9.186e+02, percent-clipped=1.0
+2023-03-29 00:39:44,145 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7306, 2.8623, 2.9727, 2.8030, 2.6965, 2.6909, 2.7871, 3.0068],
+       device='cuda:3'), covar=tensor([0.0287, 0.0324, 0.0248, 0.0321, 0.0396, 0.0358, 0.0344, 0.0349],
+       device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0077, 0.0080, 0.0074, 0.0088, 0.0080, 0.0097, 0.0070],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 00:40:13,040 INFO [train.py:892] (3/4) Epoch 29, batch 1800, loss[loss=0.1867, simple_loss=0.2673, pruned_loss=0.05299, over 19772.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2431, pruned_loss=0.04409, over 3947998.15 frames. ], batch size: 233, lr: 5.36e-03, grad_scale: 8.0
+2023-03-29 00:40:54,034 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:41:55,150 INFO [train.py:892] (3/4) Epoch 29, batch 1850, loss[loss=0.1713, simple_loss=0.2456, pruned_loss=0.04854, over 19817.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2442, pruned_loss=0.0439, over 3947196.51 frames. ], batch size: 57, lr: 5.36e-03, grad_scale: 8.0
+2023-03-29 00:42:59,929 INFO [train.py:892] (3/4) Epoch 30, batch 0, loss[loss=0.1355, simple_loss=0.2073, pruned_loss=0.03184, over 19792.00 frames. ], tot_loss[loss=0.1355, simple_loss=0.2073, pruned_loss=0.03184, over 19792.00 frames. ], batch size: 105, lr: 5.27e-03, grad_scale: 8.0
+2023-03-29 00:42:59,929 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-29 00:43:14,001 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0544, 2.9205, 4.5449, 3.3769, 3.7620, 3.3951, 2.4537, 2.5787],
+       device='cuda:3'), covar=tensor([0.1078, 0.3689, 0.0559, 0.1095, 0.1845, 0.1711, 0.2881, 0.2883],
+       device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0388, 0.0348, 0.0285, 0.0373, 0.0374, 0.0371, 0.0340],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 00:43:34,859 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2069, 2.8176, 2.9062, 3.0856, 3.0633, 2.8163, 4.2224, 4.5272],
+       device='cuda:3'), covar=tensor([0.1241, 0.1635, 0.1564, 0.2146, 0.2071, 0.2044, 0.0575, 0.0347],
+       device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0238, 0.0266, 0.0252, 0.0295, 0.0256, 0.0232, 0.0252],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 00:43:35,880 INFO [train.py:926] (3/4) Epoch 30, validation: loss=0.1794, simple_loss=0.2489, pruned_loss=0.05491, over 2883724.00 frames. 
+2023-03-29 00:43:35,881 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-29 00:43:50,263 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9440, 4.0346, 2.4742, 4.2695, 4.4940, 2.0238, 3.7503, 3.4396],
+       device='cuda:3'), covar=tensor([0.0728, 0.0893, 0.2641, 0.0884, 0.0511, 0.2811, 0.1010, 0.0877],
+       device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0255, 0.0231, 0.0274, 0.0253, 0.0204, 0.0240, 0.0197],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 00:44:13,539 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.2078, 4.8836, 4.8839, 5.2295, 4.9902, 5.4857, 5.3138, 5.5575],
+       device='cuda:3'), covar=tensor([0.0664, 0.0329, 0.0457, 0.0313, 0.0569, 0.0292, 0.0406, 0.0262],
+       device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0175, 0.0202, 0.0174, 0.0172, 0.0157, 0.0150, 0.0197],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 00:44:27,058 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0
+2023-03-29 00:44:48,488 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.508e+02 3.762e+02 4.456e+02 5.146e+02 7.887e+02, threshold=8.911e+02, percent-clipped=0.0
+2023-03-29 00:45:19,717 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.1262, 1.5166, 1.6442, 2.3213, 2.5285, 2.6600, 2.5145, 2.6208],
+       device='cuda:3'), covar=tensor([0.1035, 0.1924, 0.1745, 0.0802, 0.0545, 0.0409, 0.0477, 0.0414],
+       device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0170, 0.0178, 0.0151, 0.0134, 0.0131, 0.0122, 0.0114],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 00:45:28,954 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9707, 4.0989, 2.4907, 4.3598, 4.5546, 1.9812, 3.7914, 3.3957],
+       device='cuda:3'), covar=tensor([0.0712, 0.0800, 0.2594, 0.0691, 0.0440, 0.2779, 0.0941, 0.0890],
+       device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0254, 0.0230, 0.0274, 0.0252, 0.0204, 0.0239, 0.0197],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 00:45:46,671 INFO [train.py:892] (3/4) Epoch 30, batch 50, loss[loss=0.184, simple_loss=0.2668, pruned_loss=0.05058, over 19672.00 frames. ], tot_loss[loss=0.1622, simple_loss=0.2408, pruned_loss=0.0418, over 890121.06 frames. ], batch size: 55, lr: 5.26e-03, grad_scale: 8.0
+2023-03-29 00:47:46,055 INFO [train.py:892] (3/4) Epoch 30, batch 100, loss[loss=0.1731, simple_loss=0.2537, pruned_loss=0.04624, over 19563.00 frames. ], tot_loss[loss=0.1622, simple_loss=0.2409, pruned_loss=0.04173, over 1568767.59 frames. ], batch size: 60, lr: 5.26e-03, grad_scale: 8.0
+2023-03-29 00:47:49,924 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53897.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:47:56,134 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.2966, 4.9849, 4.9939, 5.3496, 4.9785, 5.4696, 5.4512, 5.6832],
+       device='cuda:3'), covar=tensor([0.0611, 0.0367, 0.0406, 0.0296, 0.0668, 0.0371, 0.0370, 0.0289],
+       device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0175, 0.0202, 0.0175, 0.0173, 0.0158, 0.0151, 0.0197],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 00:48:56,409 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.692e+02 3.728e+02 4.256e+02 5.465e+02 1.327e+03, threshold=8.513e+02, percent-clipped=4.0
+2023-03-29 00:49:50,717 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:49:51,902 INFO [train.py:892] (3/4) Epoch 30, batch 150, loss[loss=0.167, simple_loss=0.2331, pruned_loss=0.05042, over 19751.00 frames. ], tot_loss[loss=0.1629, simple_loss=0.2406, pruned_loss=0.04257, over 2097715.73 frames. ], batch size: 140, lr: 5.26e-03, grad_scale: 8.0
+2023-03-29 00:50:07,959 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.57 vs. limit=5.0
+2023-03-29 00:50:27,312 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53960.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:50:55,515 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53971.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:51:57,055 INFO [train.py:892] (3/4) Epoch 30, batch 200, loss[loss=0.1364, simple_loss=0.2127, pruned_loss=0.03003, over 19762.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.2438, pruned_loss=0.04429, over 2508374.50 frames. ], batch size: 100, lr: 5.26e-03, grad_scale: 8.0
+2023-03-29 00:52:32,150 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=54008.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:52:59,031 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=54019.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:53:10,611 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=54023.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 00:53:11,546 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.594e+02 3.930e+02 4.416e+02 5.090e+02 1.215e+03, threshold=8.832e+02, percent-clipped=2.0
+2023-03-29 00:53:26,161 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6848, 4.4036, 4.4012, 4.1087, 4.6277, 3.1115, 3.7607, 2.3770],
+       device='cuda:3'), covar=tensor([0.0200, 0.0225, 0.0159, 0.0220, 0.0162, 0.1024, 0.0917, 0.1551],
+       device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0146, 0.0114, 0.0134, 0.0120, 0.0135, 0.0144, 0.0128],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 00:54:01,473 INFO [train.py:892] (3/4) Epoch 30, batch 250, loss[loss=0.1775, simple_loss=0.2493, pruned_loss=0.05286, over 19845.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2428, pruned_loss=0.04395, over 2829329.47 frames. ], batch size: 190, lr: 5.25e-03, grad_scale: 8.0
+2023-03-29 00:54:14,386 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6619, 3.6984, 2.2582, 3.8848, 4.0135, 1.8784, 3.3447, 3.0554],
+       device='cuda:3'), covar=tensor([0.0860, 0.1021, 0.2896, 0.0855, 0.0698, 0.2723, 0.1041, 0.0968],
+       device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0258, 0.0233, 0.0277, 0.0255, 0.0206, 0.0241, 0.0199],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 00:55:00,479 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.1097, 2.2067, 2.2197, 2.2388, 2.2356, 2.3330, 2.1963, 2.3575],
+       device='cuda:3'), covar=tensor([0.0421, 0.0335, 0.0368, 0.0322, 0.0473, 0.0310, 0.0423, 0.0360],
+       device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0078, 0.0082, 0.0075, 0.0089, 0.0081, 0.0098, 0.0071],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 00:55:05,200 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=54071.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 00:56:07,027 INFO [train.py:892] (3/4) Epoch 30, batch 300, loss[loss=0.1593, simple_loss=0.2324, pruned_loss=0.04315, over 19766.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2432, pruned_loss=0.04381, over 3076484.26 frames. ], batch size: 130, lr: 5.25e-03, grad_scale: 8.0
+2023-03-29 00:56:08,330 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5685, 2.8621, 2.5374, 2.0488, 2.5749, 2.7578, 2.8150, 2.8873],
+       device='cuda:3'), covar=tensor([0.0331, 0.0310, 0.0314, 0.0549, 0.0378, 0.0291, 0.0231, 0.0244],
+       device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0096, 0.0098, 0.0101, 0.0104, 0.0085, 0.0085, 0.0086],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 00:56:28,543 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0
+2023-03-29 00:57:14,798 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.600e+02 3.558e+02 4.130e+02 5.342e+02 1.010e+03, threshold=8.261e+02, percent-clipped=1.0
+2023-03-29 00:58:16,929 INFO [train.py:892] (3/4) Epoch 30, batch 350, loss[loss=0.1378, simple_loss=0.2269, pruned_loss=0.02437, over 19679.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2429, pruned_loss=0.04321, over 3271692.58 frames. ], batch size: 49, lr: 5.25e-03, grad_scale: 8.0
+2023-03-29 00:58:53,323 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1460, 3.1774, 3.5320, 2.6301, 3.6491, 2.9500, 3.2192, 3.4510],
+       device='cuda:3'), covar=tensor([0.0772, 0.0425, 0.0389, 0.0771, 0.0293, 0.0450, 0.0450, 0.0338],
+       device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0084, 0.0082, 0.0109, 0.0078, 0.0080, 0.0078, 0.0070],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 00:59:07,683 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7695, 4.6657, 5.1598, 4.7198, 4.1601, 4.8748, 4.7891, 5.2795],
+       device='cuda:3'), covar=tensor([0.0805, 0.0346, 0.0347, 0.0366, 0.0859, 0.0509, 0.0478, 0.0310],
+       device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0221, 0.0223, 0.0234, 0.0208, 0.0241, 0.0230, 0.0214],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 01:00:25,834 INFO [train.py:892] (3/4) Epoch 30, batch 400, loss[loss=0.154, simple_loss=0.2334, pruned_loss=0.03726, over 19803.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2442, pruned_loss=0.04393, over 3421913.28 frames. ], batch size: 47, lr: 5.25e-03, grad_scale: 8.0
+2023-03-29 01:00:35,748 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-29 01:00:41,504 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4224, 3.3275, 3.7044, 3.3924, 3.2008, 3.6451, 3.4848, 3.7671],
+       device='cuda:3'), covar=tensor([0.0915, 0.0428, 0.0423, 0.0459, 0.1560, 0.0626, 0.0527, 0.0408],
+       device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0221, 0.0223, 0.0234, 0.0208, 0.0241, 0.0230, 0.0214],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 01:01:25,507 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0
+2023-03-29 01:01:38,175 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.477e+02 3.958e+02 4.510e+02 5.368e+02 1.015e+03, threshold=9.019e+02, percent-clipped=3.0
+2023-03-29 01:02:36,032 INFO [train.py:892] (3/4) Epoch 30, batch 450, loss[loss=0.1609, simple_loss=0.2398, pruned_loss=0.04101, over 19959.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2451, pruned_loss=0.04437, over 3537784.10 frames. ], batch size: 53, lr: 5.24e-03, grad_scale: 8.0
+2023-03-29 01:04:30,220 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8271, 2.7866, 1.8258, 3.3206, 3.0433, 3.2055, 3.3423, 2.5891],
+       device='cuda:3'), covar=tensor([0.0674, 0.0757, 0.1714, 0.0620, 0.0621, 0.0590, 0.0558, 0.0913],
+       device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0143, 0.0143, 0.0152, 0.0133, 0.0134, 0.0146, 0.0145],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 01:04:31,236 INFO [train.py:892] (3/4) Epoch 30, batch 500, loss[loss=0.1811, simple_loss=0.2558, pruned_loss=0.05319, over 19836.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2436, pruned_loss=0.04361, over 3631092.24 frames. ], batch size: 171, lr: 5.24e-03, grad_scale: 16.0
+2023-03-29 01:05:11,761 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0
+2023-03-29 01:05:41,313 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.522e+02 3.919e+02 4.339e+02 5.578e+02 1.318e+03, threshold=8.679e+02, percent-clipped=2.0
+2023-03-29 01:06:34,697 INFO [train.py:892] (3/4) Epoch 30, batch 550, loss[loss=0.1564, simple_loss=0.2359, pruned_loss=0.03846, over 19647.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2453, pruned_loss=0.04462, over 3700019.44 frames. ], batch size: 68, lr: 5.24e-03, grad_scale: 16.0
+2023-03-29 01:07:41,887 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8752, 4.7329, 5.2705, 4.7802, 4.3257, 5.0375, 4.8719, 5.4310],
+       device='cuda:3'), covar=tensor([0.0812, 0.0360, 0.0349, 0.0352, 0.0781, 0.0472, 0.0449, 0.0296],
+       device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0223, 0.0224, 0.0235, 0.0208, 0.0243, 0.0231, 0.0215],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 01:08:39,896 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.3203, 4.9580, 4.9493, 5.2965, 4.9872, 5.5411, 5.4082, 5.6276],
+       device='cuda:3'), covar=tensor([0.0604, 0.0410, 0.0498, 0.0343, 0.0589, 0.0339, 0.0453, 0.0302],
+       device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0171, 0.0198, 0.0171, 0.0168, 0.0154, 0.0147, 0.0194],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 01:08:41,051 INFO [train.py:892] (3/4) Epoch 30, batch 600, loss[loss=0.1577, simple_loss=0.2302, pruned_loss=0.04261, over 19779.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2441, pruned_loss=0.04436, over 3755996.52 frames. ], batch size: 46, lr: 5.24e-03, grad_scale: 16.0
+2023-03-29 01:09:49,425 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.185e+02 3.804e+02 4.477e+02 5.601e+02 8.237e+02, threshold=8.953e+02, percent-clipped=0.0
+2023-03-29 01:09:53,144 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-29 01:10:46,780 INFO [train.py:892] (3/4) Epoch 30, batch 650, loss[loss=0.1666, simple_loss=0.2497, pruned_loss=0.04174, over 19878.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2443, pruned_loss=0.04461, over 3798330.53 frames. ], batch size: 48, lr: 5.23e-03, grad_scale: 16.0
+2023-03-29 01:11:10,708 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0
+2023-03-29 01:12:02,520 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0876, 3.4581, 3.4660, 4.0192, 2.7434, 3.3746, 2.5118, 2.5277],
+       device='cuda:3'), covar=tensor([0.0477, 0.1642, 0.0919, 0.0376, 0.1898, 0.0789, 0.1350, 0.1624],
+       device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0329, 0.0246, 0.0200, 0.0246, 0.0207, 0.0215, 0.0215],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 01:12:06,987 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9794, 4.0227, 2.3813, 4.2537, 4.4331, 1.9116, 3.6830, 3.2900],
+       device='cuda:3'), covar=tensor([0.0725, 0.0807, 0.2686, 0.0795, 0.0577, 0.2927, 0.1057, 0.0914],
+       device='cuda:3'), in_proj_covar=tensor([0.0230, 0.0252, 0.0228, 0.0271, 0.0250, 0.0201, 0.0236, 0.0195],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 01:12:28,671 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0
+2023-03-29 01:12:36,260 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8826, 2.4896, 3.0774, 3.1861, 3.6692, 4.0763, 3.9219, 3.9352],
+       device='cuda:3'), covar=tensor([0.1036, 0.1644, 0.1189, 0.0694, 0.0397, 0.0247, 0.0400, 0.0460],
+       device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0172, 0.0178, 0.0152, 0.0135, 0.0132, 0.0123, 0.0116],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 01:12:48,855 INFO [train.py:892] (3/4) Epoch 30, batch 700, loss[loss=0.1501, simple_loss=0.2278, pruned_loss=0.03623, over 19839.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2438, pruned_loss=0.04459, over 3833040.01 frames. ], batch size: 90, lr: 5.23e-03, grad_scale: 16.0
+2023-03-29 01:13:56,162 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.8929, 6.1882, 6.1989, 6.0538, 5.8813, 6.1271, 5.5104, 5.5562],
+       device='cuda:3'), covar=tensor([0.0388, 0.0376, 0.0410, 0.0403, 0.0449, 0.0506, 0.0707, 0.0966],
+       device='cuda:3'), in_proj_covar=tensor([0.0266, 0.0283, 0.0293, 0.0257, 0.0261, 0.0247, 0.0266, 0.0310],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 01:13:59,630 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.972e+02 4.007e+02 4.588e+02 5.401e+02 1.198e+03, threshold=9.175e+02, percent-clipped=2.0
+2023-03-29 01:14:34,194 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3532, 3.3765, 5.0195, 3.8550, 4.0120, 3.9113, 2.7212, 3.0049],
+       device='cuda:3'), covar=tensor([0.0916, 0.2826, 0.0426, 0.0960, 0.1646, 0.1315, 0.2488, 0.2357],
+       device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0388, 0.0348, 0.0284, 0.0372, 0.0372, 0.0371, 0.0340],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 01:14:53,556 INFO [train.py:892] (3/4) Epoch 30, batch 750, loss[loss=0.1556, simple_loss=0.2416, pruned_loss=0.03478, over 19692.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2437, pruned_loss=0.04411, over 3859923.68 frames. ], batch size: 74, lr: 5.23e-03, grad_scale: 16.0
+2023-03-29 01:17:01,219 INFO [train.py:892] (3/4) Epoch 30, batch 800, loss[loss=0.1486, simple_loss=0.2324, pruned_loss=0.03239, over 19841.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2435, pruned_loss=0.04343, over 3879581.64 frames. ], batch size: 90, lr: 5.23e-03, grad_scale: 16.0
+2023-03-29 01:18:07,936 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.698e+02 3.768e+02 4.525e+02 5.531e+02 9.208e+02, threshold=9.049e+02, percent-clipped=1.0
+2023-03-29 01:19:02,299 INFO [train.py:892] (3/4) Epoch 30, batch 850, loss[loss=0.1473, simple_loss=0.2356, pruned_loss=0.02952, over 19620.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2433, pruned_loss=0.04312, over 3894103.74 frames. ], batch size: 65, lr: 5.22e-03, grad_scale: 16.0
+2023-03-29 01:19:48,511 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54664.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 01:20:19,762 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7000, 2.2443, 2.5477, 2.9516, 3.3700, 3.5355, 3.4837, 3.4796],
+       device='cuda:3'), covar=tensor([0.1079, 0.1666, 0.1411, 0.0742, 0.0454, 0.0331, 0.0397, 0.0414],
+       device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0169, 0.0176, 0.0150, 0.0134, 0.0130, 0.0121, 0.0114],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 01:21:04,099 INFO [train.py:892] (3/4) Epoch 30, batch 900, loss[loss=0.1422, simple_loss=0.219, pruned_loss=0.03275, over 19810.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2421, pruned_loss=0.04273, over 3907622.66 frames. ], batch size: 96, lr: 5.22e-03, grad_scale: 16.0
+2023-03-29 01:21:23,653 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4793, 2.5148, 2.7033, 2.1342, 2.7752, 2.3267, 2.6544, 2.6767],
+       device='cuda:3'), covar=tensor([0.0580, 0.0583, 0.0474, 0.0879, 0.0391, 0.0526, 0.0487, 0.0329],
+       device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0085, 0.0083, 0.0111, 0.0079, 0.0082, 0.0080, 0.0071],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 01:22:16,859 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.427e+02 3.830e+02 4.433e+02 5.548e+02 1.086e+03, threshold=8.866e+02, percent-clipped=1.0
+2023-03-29 01:22:22,802 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54725.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 01:23:12,080 INFO [train.py:892] (3/4) Epoch 30, batch 950, loss[loss=0.1737, simple_loss=0.2593, pruned_loss=0.04409, over 19582.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2427, pruned_loss=0.0432, over 3917782.68 frames. ], batch size: 49, lr: 5.22e-03, grad_scale: 16.0
+2023-03-29 01:24:22,649 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5117, 3.4506, 3.9177, 2.9016, 4.0613, 3.2806, 3.5224, 3.7963],
+       device='cuda:3'), covar=tensor([0.0684, 0.0463, 0.0509, 0.0726, 0.0341, 0.0407, 0.0490, 0.0319],
+       device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0085, 0.0083, 0.0111, 0.0079, 0.0082, 0.0080, 0.0071],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 01:25:12,229 INFO [train.py:892] (3/4) Epoch 30, batch 1000, loss[loss=0.1936, simple_loss=0.2714, pruned_loss=0.05788, over 19635.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2424, pruned_loss=0.04311, over 3925064.66 frames. ], batch size: 343, lr: 5.22e-03, grad_scale: 16.0
+2023-03-29 01:26:22,813 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.928e+02 3.838e+02 4.641e+02 5.585e+02 1.320e+03, threshold=9.281e+02, percent-clipped=3.0
+2023-03-29 01:27:18,811 INFO [train.py:892] (3/4) Epoch 30, batch 1050, loss[loss=0.1583, simple_loss=0.242, pruned_loss=0.03734, over 19867.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2426, pruned_loss=0.04335, over 3932234.10 frames. ], batch size: 64, lr: 5.21e-03, grad_scale: 16.0
+2023-03-29 01:27:59,703 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54862.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 01:29:25,272 INFO [train.py:892] (3/4) Epoch 30, batch 1100, loss[loss=0.1515, simple_loss=0.2267, pruned_loss=0.03819, over 19794.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2421, pruned_loss=0.04325, over 3936826.88 frames. ], batch size: 40, lr: 5.21e-03, grad_scale: 16.0
+2023-03-29 01:30:37,603 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54923.0, num_to_drop=1, layers_to_drop={3}
+2023-03-29 01:30:38,554 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.418e+02 3.807e+02 4.386e+02 5.197e+02 8.292e+02, threshold=8.772e+02, percent-clipped=0.0
+2023-03-29 01:31:31,547 INFO [train.py:892] (3/4) Epoch 30, batch 1150, loss[loss=0.1612, simple_loss=0.2399, pruned_loss=0.04126, over 19770.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2432, pruned_loss=0.0441, over 3938600.36 frames. ], batch size: 198, lr: 5.21e-03, grad_scale: 16.0
+2023-03-29 01:31:44,706 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5052, 3.3522, 5.0289, 3.7764, 4.0970, 3.8983, 2.7207, 2.9879],
+       device='cuda:3'), covar=tensor([0.0848, 0.3079, 0.0433, 0.0943, 0.1569, 0.1345, 0.2509, 0.2376],
+       device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0387, 0.0348, 0.0285, 0.0372, 0.0374, 0.0372, 0.0340],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 01:32:32,817 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.9420, 2.1676, 1.9855, 1.4391, 2.0539, 2.1473, 2.0114, 2.1035],
+       device='cuda:3'), covar=tensor([0.0398, 0.0334, 0.0324, 0.0571, 0.0399, 0.0306, 0.0312, 0.0283],
+       device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0096, 0.0098, 0.0101, 0.0104, 0.0086, 0.0085, 0.0086],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 01:33:13,528 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9541, 4.7786, 4.6965, 5.0333, 4.9370, 5.4135, 5.0591, 5.1011],
+       device='cuda:3'), covar=tensor([0.0888, 0.0471, 0.0541, 0.0449, 0.0606, 0.0334, 0.0583, 0.0697],
+       device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0175, 0.0202, 0.0174, 0.0172, 0.0157, 0.0149, 0.0198],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 01:33:31,624 INFO [train.py:892] (3/4) Epoch 30, batch 1200, loss[loss=0.1377, simple_loss=0.2194, pruned_loss=0.028, over 19806.00 frames. ], tot_loss[loss=0.165, simple_loss=0.2429, pruned_loss=0.04357, over 3942002.08 frames. ], batch size: 117, lr: 5.21e-03, grad_scale: 16.0
+2023-03-29 01:34:33,649 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55020.0, num_to_drop=1, layers_to_drop={2}
+2023-03-29 01:34:43,884 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.409e+02 3.740e+02 4.299e+02 5.341e+02 1.989e+03, threshold=8.597e+02, percent-clipped=2.0
+2023-03-29 01:35:37,510 INFO [train.py:892] (3/4) Epoch 30, batch 1250, loss[loss=0.2244, simple_loss=0.2987, pruned_loss=0.0751, over 19607.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2417, pruned_loss=0.04307, over 3943956.98 frames. ], batch size: 387, lr: 5.21e-03, grad_scale: 16.0
+2023-03-29 01:35:42,533 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5158, 4.3675, 4.8513, 4.4605, 4.0447, 4.6902, 4.4989, 4.9809],
+       device='cuda:3'), covar=tensor([0.0881, 0.0419, 0.0393, 0.0392, 0.0902, 0.0509, 0.0456, 0.0350],
+       device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0225, 0.0224, 0.0235, 0.0210, 0.0243, 0.0232, 0.0217],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 01:37:48,070 INFO [train.py:892] (3/4) Epoch 30, batch 1300, loss[loss=0.1657, simple_loss=0.239, pruned_loss=0.04619, over 19806.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.243, pruned_loss=0.04362, over 3944901.38 frames. ], batch size: 181, lr: 5.20e-03, grad_scale: 16.0
+2023-03-29 01:38:08,797 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0803, 5.3277, 5.3836, 5.2530, 4.9736, 5.3418, 4.7885, 4.8404],
+       device='cuda:3'), covar=tensor([0.0394, 0.0430, 0.0444, 0.0418, 0.0596, 0.0502, 0.0760, 0.0964],
+       device='cuda:3'), in_proj_covar=tensor([0.0267, 0.0285, 0.0296, 0.0259, 0.0263, 0.0248, 0.0267, 0.0311],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 01:38:12,815 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4945, 3.3844, 3.7435, 2.7813, 3.9236, 3.1497, 3.4350, 3.7410],
+       device='cuda:3'), covar=tensor([0.0727, 0.0432, 0.0549, 0.0789, 0.0362, 0.0439, 0.0462, 0.0325],
+       device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0086, 0.0083, 0.0111, 0.0079, 0.0081, 0.0079, 0.0071],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 01:38:55,050 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.487e+02 3.571e+02 4.407e+02 5.506e+02 1.177e+03, threshold=8.814e+02, percent-clipped=1.0
+2023-03-29 01:39:52,153 INFO [train.py:892] (3/4) Epoch 30, batch 1350, loss[loss=0.1407, simple_loss=0.2235, pruned_loss=0.02892, over 19834.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2434, pruned_loss=0.0435, over 3944771.74 frames. ], batch size: 76, lr: 5.20e-03, grad_scale: 16.0
+2023-03-29 01:39:58,456 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55148.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 01:40:39,657 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.87 vs. limit=5.0
+2023-03-29 01:41:51,308 INFO [train.py:892] (3/4) Epoch 30, batch 1400, loss[loss=0.1698, simple_loss=0.2568, pruned_loss=0.04142, over 19860.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2429, pruned_loss=0.0431, over 3945430.82 frames. ], batch size: 51, lr: 5.20e-03, grad_scale: 16.0
+2023-03-29 01:42:14,628 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7671, 3.8046, 2.3544, 4.0066, 4.1398, 1.9544, 3.3879, 3.2666],
+       device='cuda:3'), covar=tensor([0.0784, 0.0961, 0.2683, 0.0767, 0.0631, 0.2752, 0.1151, 0.0815],
+       device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0254, 0.0229, 0.0273, 0.0253, 0.0203, 0.0238, 0.0196],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 01:42:29,682 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55209.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 01:42:34,922 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 01:42:52,122 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55218.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 01:43:04,803 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.260e+02 3.753e+02 4.356e+02 5.536e+02 1.082e+03, threshold=8.713e+02, percent-clipped=2.0
+2023-03-29 01:43:37,358 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7636, 2.7769, 1.7176, 3.2015, 2.8925, 3.0545, 3.1888, 2.5943],
+       device='cuda:3'), covar=tensor([0.0701, 0.0705, 0.1593, 0.0559, 0.0669, 0.0529, 0.0548, 0.0859],
+       device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0143, 0.0143, 0.0152, 0.0133, 0.0135, 0.0147, 0.0145],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 01:43:53,893 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.11 vs. limit=2.0
+2023-03-29 01:44:02,207 INFO [train.py:892] (3/4) Epoch 30, batch 1450, loss[loss=0.1528, simple_loss=0.243, pruned_loss=0.03128, over 19885.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2429, pruned_loss=0.04291, over 3946398.48 frames. ], batch size: 52, lr: 5.20e-03, grad_scale: 16.0
+2023-03-29 01:44:12,187 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3927, 3.5488, 2.1306, 4.0672, 3.6829, 4.0462, 4.0900, 3.2989],
+       device='cuda:3'), covar=tensor([0.0653, 0.0585, 0.1624, 0.0673, 0.0685, 0.0402, 0.0661, 0.0767],
+       device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0142, 0.0143, 0.0152, 0.0133, 0.0134, 0.0147, 0.0145],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 01:44:57,878 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5289, 3.7699, 4.0704, 4.7353, 3.0817, 3.4067, 3.0084, 2.9221],
+       device='cuda:3'), covar=tensor([0.0467, 0.2212, 0.0819, 0.0327, 0.1905, 0.1002, 0.1202, 0.1550],
+       device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0328, 0.0245, 0.0200, 0.0244, 0.0206, 0.0214, 0.0216],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 01:45:06,855 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 01:46:06,710 INFO [train.py:892] (3/4) Epoch 30, batch 1500, loss[loss=0.1722, simple_loss=0.2455, pruned_loss=0.04941, over 19803.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2423, pruned_loss=0.04282, over 3947923.93 frames. ], batch size: 211, lr: 5.19e-03, grad_scale: 16.0
+2023-03-29 01:47:08,106 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55320.0, num_to_drop=1, layers_to_drop={2}
+2023-03-29 01:47:16,430 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.567e+02 3.622e+02 4.364e+02 5.637e+02 1.011e+03, threshold=8.727e+02, percent-clipped=2.0
+2023-03-29 01:47:56,064 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0
+2023-03-29 01:48:09,094 INFO [train.py:892] (3/4) Epoch 30, batch 1550, loss[loss=0.1497, simple_loss=0.229, pruned_loss=0.03519, over 19774.00 frames. ], tot_loss[loss=0.1633, simple_loss=0.2416, pruned_loss=0.04253, over 3949402.00 frames. ], batch size: 69, lr: 5.19e-03, grad_scale: 16.0
+2023-03-29 01:49:02,378 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55368.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 01:49:26,970 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7558, 2.8597, 4.2179, 3.2237, 3.4831, 3.2307, 2.3774, 2.5419],
+       device='cuda:3'), covar=tensor([0.1101, 0.3273, 0.0567, 0.1059, 0.1745, 0.1565, 0.2696, 0.2822],
+       device='cuda:3'), in_proj_covar=tensor([0.0348, 0.0383, 0.0344, 0.0282, 0.0367, 0.0371, 0.0368, 0.0336],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 01:50:00,814 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3884, 3.4938, 2.1864, 3.5888, 3.6819, 1.7277, 3.0685, 2.9205],
+       device='cuda:3'), covar=tensor([0.0924, 0.0904, 0.2820, 0.0873, 0.0718, 0.2859, 0.1163, 0.0933],
+       device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0255, 0.0230, 0.0274, 0.0254, 0.0203, 0.0240, 0.0196],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 01:50:11,425 INFO [train.py:892] (3/4) Epoch 30, batch 1600, loss[loss=0.1692, simple_loss=0.2486, pruned_loss=0.04489, over 19855.00 frames. ], tot_loss[loss=0.1628, simple_loss=0.2411, pruned_loss=0.04221, over 3950136.65 frames. ], batch size: 60, lr: 5.19e-03, grad_scale: 16.0
+2023-03-29 01:51:20,601 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.706e+02 3.762e+02 4.518e+02 5.538e+02 9.560e+02, threshold=9.036e+02, percent-clipped=2.0
+2023-03-29 01:52:18,782 INFO [train.py:892] (3/4) Epoch 30, batch 1650, loss[loss=0.2045, simple_loss=0.281, pruned_loss=0.06401, over 19748.00 frames. ], tot_loss[loss=0.1627, simple_loss=0.2413, pruned_loss=0.04204, over 3949465.19 frames. ], batch size: 221, lr: 5.19e-03, grad_scale: 16.0
+2023-03-29 01:53:26,684 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2973, 2.5994, 3.6237, 2.8907, 3.0358, 2.9229, 2.1259, 2.2877],
+       device='cuda:3'), covar=tensor([0.1183, 0.2825, 0.0677, 0.1084, 0.1757, 0.1497, 0.2579, 0.2593],
+       device='cuda:3'), in_proj_covar=tensor([0.0349, 0.0385, 0.0345, 0.0283, 0.0369, 0.0373, 0.0370, 0.0337],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 01:54:04,639 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.17 vs. limit=5.0
+2023-03-29 01:54:24,833 INFO [train.py:892] (3/4) Epoch 30, batch 1700, loss[loss=0.167, simple_loss=0.2339, pruned_loss=0.04999, over 19781.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.243, pruned_loss=0.04308, over 3949282.58 frames. ], batch size: 154, lr: 5.18e-03, grad_scale: 16.0
+2023-03-29 01:54:30,485 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7790, 2.6904, 2.8738, 2.3742, 3.0226, 2.4915, 2.9446, 2.9506],
+       device='cuda:3'), covar=tensor([0.0561, 0.0543, 0.0498, 0.0780, 0.0408, 0.0533, 0.0428, 0.0333],
+       device='cuda:3'), in_proj_covar=tensor([0.0077, 0.0085, 0.0082, 0.0110, 0.0078, 0.0081, 0.0079, 0.0071],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 01:54:45,452 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 01:55:19,704 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55518.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 01:55:32,301 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.784e+02 3.668e+02 4.195e+02 5.159e+02 8.554e+02, threshold=8.391e+02, percent-clipped=0.0
+2023-03-29 01:56:20,132 INFO [train.py:892] (3/4) Epoch 30, batch 1750, loss[loss=0.177, simple_loss=0.2409, pruned_loss=0.05661, over 19841.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2436, pruned_loss=0.04355, over 3949097.05 frames. ], batch size: 144, lr: 5.18e-03, grad_scale: 16.0
+2023-03-29 01:57:01,900 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55566.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 01:57:04,593 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 01:58:06,254 INFO [train.py:892] (3/4) Epoch 30, batch 1800, loss[loss=0.1836, simple_loss=0.2599, pruned_loss=0.05365, over 19539.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2433, pruned_loss=0.04385, over 3949390.08 frames. ], batch size: 46, lr: 5.18e-03, grad_scale: 16.0
+2023-03-29 01:59:02,431 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.794e+02 4.015e+02 4.867e+02 6.116e+02 1.314e+03, threshold=9.734e+02, percent-clipped=8.0
+2023-03-29 01:59:44,988 INFO [train.py:892] (3/4) Epoch 30, batch 1850, loss[loss=0.1632, simple_loss=0.2438, pruned_loss=0.04129, over 19828.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2447, pruned_loss=0.04352, over 3949157.23 frames. ], batch size: 57, lr: 5.18e-03, grad_scale: 16.0
+2023-03-29 02:00:52,116 INFO [train.py:892] (3/4) Epoch 31, batch 0, loss[loss=0.1391, simple_loss=0.2094, pruned_loss=0.03437, over 19847.00 frames. ], tot_loss[loss=0.1391, simple_loss=0.2094, pruned_loss=0.03437, over 19847.00 frames. ], batch size: 137, lr: 5.09e-03, grad_scale: 16.0
+2023-03-29 02:00:52,117 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-29 02:01:29,196 INFO [train.py:926] (3/4) Epoch 31, validation: loss=0.1803, simple_loss=0.2493, pruned_loss=0.05567, over 2883724.00 frames. 
+2023-03-29 02:01:29,197 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-29 02:02:51,619 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.3473, 5.0470, 5.1075, 5.4856, 5.0874, 5.6168, 5.5259, 5.7343],
+       device='cuda:3'), covar=tensor([0.0665, 0.0374, 0.0416, 0.0315, 0.0618, 0.0383, 0.0381, 0.0298],
+       device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0176, 0.0201, 0.0174, 0.0172, 0.0157, 0.0148, 0.0197],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 02:03:38,846 INFO [train.py:892] (3/4) Epoch 31, batch 50, loss[loss=0.1576, simple_loss=0.2415, pruned_loss=0.03683, over 19875.00 frames. ], tot_loss[loss=0.1581, simple_loss=0.2353, pruned_loss=0.04046, over 889906.01 frames. ], batch size: 53, lr: 5.09e-03, grad_scale: 16.0
+2023-03-29 02:04:41,988 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.130e+02 3.666e+02 4.158e+02 5.108e+02 9.085e+02, threshold=8.317e+02, percent-clipped=0.0
+2023-03-29 02:05:06,835 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7723, 2.4726, 4.0650, 3.6031, 3.8881, 4.0280, 3.8796, 3.7613],
+       device='cuda:3'), covar=tensor([0.0720, 0.1034, 0.0119, 0.0601, 0.0177, 0.0251, 0.0200, 0.0186],
+       device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0104, 0.0089, 0.0155, 0.0086, 0.0098, 0.0092, 0.0087],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 02:05:38,272 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55748.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:05:43,146 INFO [train.py:892] (3/4) Epoch 31, batch 100, loss[loss=0.152, simple_loss=0.2306, pruned_loss=0.03674, over 19765.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2419, pruned_loss=0.0436, over 1568227.28 frames. ], batch size: 247, lr: 5.09e-03, grad_scale: 16.0
+2023-03-29 02:06:41,132 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5475, 3.5107, 5.0552, 3.8688, 4.0384, 3.9910, 2.7689, 3.0944],
+       device='cuda:3'), covar=tensor([0.0771, 0.2543, 0.0378, 0.0922, 0.1596, 0.1243, 0.2411, 0.2275],
+       device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0388, 0.0347, 0.0285, 0.0372, 0.0375, 0.0373, 0.0340],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 02:06:48,543 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.1605, 4.8536, 4.8916, 5.1537, 4.6895, 5.3788, 5.2712, 5.4761],
+       device='cuda:3'), covar=tensor([0.0620, 0.0390, 0.0398, 0.0335, 0.0626, 0.0360, 0.0373, 0.0308],
+       device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0176, 0.0202, 0.0175, 0.0173, 0.0158, 0.0149, 0.0197],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 02:07:54,538 INFO [train.py:892] (3/4) Epoch 31, batch 150, loss[loss=0.1814, simple_loss=0.2587, pruned_loss=0.05199, over 19720.00 frames. ], tot_loss[loss=0.1617, simple_loss=0.2395, pruned_loss=0.04195, over 2097143.55 frames. ], batch size: 269, lr: 5.08e-03, grad_scale: 16.0
+2023-03-29 02:08:04,183 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:08:17,845 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55809.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:08:43,611 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5892, 3.5062, 3.4802, 3.2191, 3.5647, 2.7178, 2.8968, 1.8085],
+       device='cuda:3'), covar=tensor([0.0229, 0.0240, 0.0174, 0.0213, 0.0164, 0.1220, 0.0684, 0.1740],
+       device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0146, 0.0113, 0.0134, 0.0119, 0.0135, 0.0143, 0.0127],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 02:08:49,076 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0
+2023-03-29 02:08:54,042 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.568e+02 3.733e+02 4.638e+02 5.672e+02 1.192e+03, threshold=9.276e+02, percent-clipped=1.0
+2023-03-29 02:10:02,878 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:10:09,549 INFO [train.py:892] (3/4) Epoch 31, batch 200, loss[loss=0.1511, simple_loss=0.2247, pruned_loss=0.03873, over 19848.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.2417, pruned_loss=0.04259, over 2505426.23 frames. ], batch size: 112, lr: 5.08e-03, grad_scale: 16.0
+2023-03-29 02:10:13,103 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:10:39,940 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2468, 3.1592, 3.4529, 2.5733, 3.4432, 2.8490, 3.1893, 3.3289],
+       device='cuda:3'), covar=tensor([0.0529, 0.0483, 0.0454, 0.0843, 0.0390, 0.0510, 0.0512, 0.0357],
+       device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0085, 0.0083, 0.0110, 0.0079, 0.0081, 0.0079, 0.0071],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 02:10:54,674 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:11:20,330 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0
+2023-03-29 02:12:23,365 INFO [train.py:892] (3/4) Epoch 31, batch 250, loss[loss=0.1542, simple_loss=0.2261, pruned_loss=0.04111, over 19867.00 frames. ], tot_loss[loss=0.163, simple_loss=0.2414, pruned_loss=0.04227, over 2824233.97 frames. ], batch size: 64, lr: 5.08e-03, grad_scale: 16.0
+2023-03-29 02:12:44,592 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:12:59,729 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:13:03,662 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.1615, 2.3608, 2.3516, 1.9116, 2.4902, 2.0308, 2.3719, 2.3276],
+       device='cuda:3'), covar=tensor([0.0575, 0.0497, 0.0498, 0.0950, 0.0413, 0.0547, 0.0477, 0.0415],
+       device='cuda:3'), in_proj_covar=tensor([0.0078, 0.0085, 0.0083, 0.0111, 0.0079, 0.0081, 0.0080, 0.0072],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 02:13:23,222 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.896e+02 3.622e+02 4.377e+02 5.144e+02 8.817e+02, threshold=8.755e+02, percent-clipped=0.0
+2023-03-29 02:14:32,140 INFO [train.py:892] (3/4) Epoch 31, batch 300, loss[loss=0.1537, simple_loss=0.2294, pruned_loss=0.03903, over 19773.00 frames. ], tot_loss[loss=0.1628, simple_loss=0.2413, pruned_loss=0.04213, over 3075217.67 frames. ], batch size: 182, lr: 5.08e-03, grad_scale: 16.0
+2023-03-29 02:16:45,163 INFO [train.py:892] (3/4) Epoch 31, batch 350, loss[loss=0.1665, simple_loss=0.2457, pruned_loss=0.04363, over 19840.00 frames. ], tot_loss[loss=0.1622, simple_loss=0.2405, pruned_loss=0.04194, over 3270128.23 frames. ], batch size: 190, lr: 5.08e-03, grad_scale: 16.0
+2023-03-29 02:17:12,052 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9856, 4.6298, 4.7205, 4.4223, 4.9579, 3.2808, 4.0271, 2.4146],
+       device='cuda:3'), covar=tensor([0.0157, 0.0189, 0.0133, 0.0198, 0.0125, 0.0916, 0.0816, 0.1462],
+       device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0145, 0.0112, 0.0134, 0.0118, 0.0134, 0.0142, 0.0126],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 02:17:36,573 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5331, 3.5901, 2.2609, 4.2383, 3.7549, 4.1530, 4.2210, 3.3221],
+       device='cuda:3'), covar=tensor([0.0587, 0.0595, 0.1479, 0.0512, 0.0714, 0.0490, 0.0654, 0.0765],
+       device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0144, 0.0144, 0.0152, 0.0132, 0.0135, 0.0147, 0.0145],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 02:17:41,473 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.448e+02 3.652e+02 4.141e+02 5.135e+02 1.534e+03, threshold=8.281e+02, percent-clipped=2.0
+2023-03-29 02:17:46,405 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4668, 3.5374, 2.2242, 3.6624, 3.7591, 1.8256, 3.1657, 2.9292],
+       device='cuda:3'), covar=tensor([0.0816, 0.0816, 0.2678, 0.0809, 0.0662, 0.2584, 0.1077, 0.0900],
+       device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0255, 0.0228, 0.0272, 0.0252, 0.0203, 0.0239, 0.0197],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 02:18:47,907 INFO [train.py:892] (3/4) Epoch 31, batch 400, loss[loss=0.1626, simple_loss=0.2495, pruned_loss=0.03786, over 19813.00 frames. ], tot_loss[loss=0.1618, simple_loss=0.2401, pruned_loss=0.04173, over 3421125.44 frames. ], batch size: 50, lr: 5.07e-03, grad_scale: 16.0
+2023-03-29 02:20:53,872 INFO [train.py:892] (3/4) Epoch 31, batch 450, loss[loss=0.1601, simple_loss=0.2309, pruned_loss=0.04465, over 19875.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2421, pruned_loss=0.04293, over 3537995.83 frames. ], batch size: 159, lr: 5.07e-03, grad_scale: 16.0
+2023-03-29 02:21:03,594 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:21:56,022 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.311e+02 3.637e+02 4.331e+02 5.341e+02 9.053e+02, threshold=8.662e+02, percent-clipped=1.0
+2023-03-29 02:22:47,365 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56145.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:22:54,347 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0238, 4.9170, 5.4308, 4.9534, 4.3745, 5.2143, 5.0953, 5.6080],
+       device='cuda:3'), covar=tensor([0.0816, 0.0317, 0.0300, 0.0376, 0.0711, 0.0417, 0.0376, 0.0274],
+       device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0221, 0.0221, 0.0232, 0.0206, 0.0241, 0.0230, 0.0216],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 02:23:00,060 INFO [train.py:892] (3/4) Epoch 31, batch 500, loss[loss=0.1468, simple_loss=0.2269, pruned_loss=0.03332, over 19811.00 frames. ], tot_loss[loss=0.1658, simple_loss=0.2434, pruned_loss=0.04412, over 3628918.27 frames. ], batch size: 47, lr: 5.07e-03, grad_scale: 16.0
+2023-03-29 02:23:38,927 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2133, 3.4731, 3.0534, 2.6291, 3.0859, 3.4715, 3.3048, 3.4311],
+       device='cuda:3'), covar=tensor([0.0279, 0.0257, 0.0289, 0.0495, 0.0323, 0.0273, 0.0262, 0.0230],
+       device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0094, 0.0098, 0.0100, 0.0103, 0.0085, 0.0085, 0.0085],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 02:24:42,844 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.5550, 5.9749, 6.0937, 5.8406, 5.7183, 5.6503, 5.7570, 5.6030],
+       device='cuda:3'), covar=tensor([0.1446, 0.1262, 0.0826, 0.1106, 0.0589, 0.0842, 0.1806, 0.1893],
+       device='cuda:3'), in_proj_covar=tensor([0.0295, 0.0331, 0.0367, 0.0300, 0.0274, 0.0281, 0.0358, 0.0387],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 02:25:07,735 INFO [train.py:892] (3/4) Epoch 31, batch 550, loss[loss=0.1683, simple_loss=0.2375, pruned_loss=0.04952, over 19802.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2442, pruned_loss=0.04442, over 3697794.80 frames. ], batch size: 162, lr: 5.07e-03, grad_scale: 16.0
+2023-03-29 02:25:15,922 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:25:20,617 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:26:05,181 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.698e+02 3.907e+02 4.783e+02 5.520e+02 2.240e+03, threshold=9.567e+02, percent-clipped=4.0
+2023-03-29 02:27:16,864 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1025, 3.0734, 2.0146, 3.6712, 3.3452, 3.5861, 3.7045, 2.9505],
+       device='cuda:3'), covar=tensor([0.0681, 0.0690, 0.1739, 0.0655, 0.0685, 0.0486, 0.0733, 0.0821],
+       device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0143, 0.0143, 0.0151, 0.0132, 0.0135, 0.0146, 0.0144],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 02:27:17,952 INFO [train.py:892] (3/4) Epoch 31, batch 600, loss[loss=0.1462, simple_loss=0.227, pruned_loss=0.03268, over 19717.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2421, pruned_loss=0.04355, over 3754320.38 frames. ], batch size: 104, lr: 5.06e-03, grad_scale: 16.0
+2023-03-29 02:29:20,430 INFO [train.py:892] (3/4) Epoch 31, batch 650, loss[loss=0.1522, simple_loss=0.2249, pruned_loss=0.03977, over 19764.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.242, pruned_loss=0.04332, over 3797424.08 frames. ], batch size: 110, lr: 5.06e-03, grad_scale: 32.0
+2023-03-29 02:30:21,269 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.457e+02 3.714e+02 4.450e+02 5.010e+02 8.599e+02, threshold=8.900e+02, percent-clipped=0.0
+2023-03-29 02:31:30,806 INFO [train.py:892] (3/4) Epoch 31, batch 700, loss[loss=0.1676, simple_loss=0.241, pruned_loss=0.04709, over 19767.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2417, pruned_loss=0.04313, over 3832269.40 frames. ], batch size: 213, lr: 5.06e-03, grad_scale: 32.0
+2023-03-29 02:32:45,420 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2730, 4.5658, 4.9695, 4.4671, 4.2569, 4.8573, 4.7149, 5.1732],
+       device='cuda:3'), covar=tensor([0.1320, 0.0395, 0.0471, 0.0473, 0.0826, 0.0527, 0.0499, 0.0412],
+       device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0225, 0.0226, 0.0237, 0.0209, 0.0246, 0.0234, 0.0220],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 02:33:34,649 INFO [train.py:892] (3/4) Epoch 31, batch 750, loss[loss=0.2265, simple_loss=0.3417, pruned_loss=0.05568, over 17889.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2428, pruned_loss=0.04336, over 3855360.19 frames. ], batch size: 633, lr: 5.06e-03, grad_scale: 32.0
+2023-03-29 02:33:43,737 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56404.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:34:25,645 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.62 vs. limit=5.0
+2023-03-29 02:34:36,555 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 3.075e+02 3.994e+02 4.748e+02 5.714e+02 1.014e+03, threshold=9.496e+02, percent-clipped=1.0
+2023-03-29 02:35:40,597 INFO [train.py:892] (3/4) Epoch 31, batch 800, loss[loss=0.1771, simple_loss=0.2385, pruned_loss=0.05778, over 19741.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2425, pruned_loss=0.0432, over 3878095.53 frames. ], batch size: 140, lr: 5.06e-03, grad_scale: 32.0
+2023-03-29 02:35:44,466 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:37:14,651 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0
+2023-03-29 02:37:49,444 INFO [train.py:892] (3/4) Epoch 31, batch 850, loss[loss=0.1438, simple_loss=0.2238, pruned_loss=0.03193, over 19534.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2432, pruned_loss=0.04329, over 3893591.48 frames. ], batch size: 46, lr: 5.05e-03, grad_scale: 32.0
+2023-03-29 02:37:50,669 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:37:58,499 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:38:41,522 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56523.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 02:38:42,537 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.680e+02 3.782e+02 4.394e+02 5.442e+02 8.928e+02, threshold=8.788e+02, percent-clipped=0.0
+2023-03-29 02:39:38,501 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-29 02:39:52,780 INFO [train.py:892] (3/4) Epoch 31, batch 900, loss[loss=0.1593, simple_loss=0.2357, pruned_loss=0.04141, over 19931.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2434, pruned_loss=0.04352, over 3906110.72 frames. ], batch size: 51, lr: 5.05e-03, grad_scale: 32.0
+2023-03-29 02:39:56,935 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:40:49,926 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8661, 4.8768, 5.2763, 5.0532, 5.1228, 4.7310, 5.0472, 4.8054],
+       device='cuda:3'), covar=tensor([0.1378, 0.1576, 0.0860, 0.1238, 0.0755, 0.0999, 0.1775, 0.1938],
+       device='cuda:3'), in_proj_covar=tensor([0.0294, 0.0330, 0.0368, 0.0298, 0.0274, 0.0281, 0.0358, 0.0391],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 02:41:04,089 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2854, 4.5120, 4.5550, 4.4418, 4.2423, 4.5059, 4.0148, 4.0942],
+       device='cuda:3'), covar=tensor([0.0496, 0.0501, 0.0468, 0.0444, 0.0643, 0.0522, 0.0708, 0.0992],
+       device='cuda:3'), in_proj_covar=tensor([0.0269, 0.0286, 0.0294, 0.0258, 0.0264, 0.0248, 0.0267, 0.0314],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 02:41:19,019 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56584.0, num_to_drop=1, layers_to_drop={2}
+2023-03-29 02:42:00,184 INFO [train.py:892] (3/4) Epoch 31, batch 950, loss[loss=0.2962, simple_loss=0.3653, pruned_loss=0.1135, over 19408.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2432, pruned_loss=0.04328, over 3914224.54 frames. ], batch size: 431, lr: 5.05e-03, grad_scale: 32.0
+2023-03-29 02:42:56,325 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0
+2023-03-29 02:43:01,365 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.678e+02 3.807e+02 4.389e+02 5.044e+02 8.722e+02, threshold=8.778e+02, percent-clipped=0.0
+2023-03-29 02:44:14,756 INFO [train.py:892] (3/4) Epoch 31, batch 1000, loss[loss=0.1638, simple_loss=0.2404, pruned_loss=0.04357, over 19728.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2433, pruned_loss=0.0429, over 3919944.47 frames. ], batch size: 47, lr: 5.05e-03, grad_scale: 32.0
+2023-03-29 02:46:17,945 INFO [train.py:892] (3/4) Epoch 31, batch 1050, loss[loss=0.1593, simple_loss=0.2349, pruned_loss=0.04186, over 19798.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.244, pruned_loss=0.04317, over 3926568.73 frames. ], batch size: 167, lr: 5.04e-03, grad_scale: 32.0
+2023-03-29 02:46:19,408 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9413, 2.8789, 1.8705, 3.3914, 3.2223, 3.3624, 3.4108, 2.7658],
+       device='cuda:3'), covar=tensor([0.0644, 0.0717, 0.1748, 0.0691, 0.0572, 0.0490, 0.0696, 0.0778],
+       device='cuda:3'), in_proj_covar=tensor([0.0141, 0.0142, 0.0141, 0.0149, 0.0130, 0.0133, 0.0144, 0.0142],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 02:47:17,683 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.497e+02 3.728e+02 4.373e+02 5.189e+02 1.039e+03, threshold=8.746e+02, percent-clipped=2.0
+2023-03-29 02:48:25,995 INFO [train.py:892] (3/4) Epoch 31, batch 1100, loss[loss=0.1655, simple_loss=0.2467, pruned_loss=0.04218, over 19855.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2425, pruned_loss=0.04269, over 3933290.84 frames. ], batch size: 78, lr: 5.04e-03, grad_scale: 32.0
+2023-03-29 02:50:26,112 INFO [train.py:892] (3/4) Epoch 31, batch 1150, loss[loss=0.1611, simple_loss=0.2459, pruned_loss=0.03812, over 19805.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2433, pruned_loss=0.04286, over 3936265.62 frames. ], batch size: 120, lr: 5.04e-03, grad_scale: 32.0
+2023-03-29 02:50:27,134 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:50:32,059 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56803.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:51:22,676 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.381e+02 3.820e+02 4.662e+02 5.416e+02 9.191e+02, threshold=9.324e+02, percent-clipped=1.0
+2023-03-29 02:52:25,331 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9985, 3.2876, 2.8545, 2.4077, 2.8970, 3.2021, 3.1832, 3.2033],
+       device='cuda:3'), covar=tensor([0.0297, 0.0278, 0.0294, 0.0488, 0.0337, 0.0334, 0.0210, 0.0225],
+       device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0097, 0.0100, 0.0102, 0.0106, 0.0088, 0.0087, 0.0088],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 02:52:27,701 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56849.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:52:31,172 INFO [train.py:892] (3/4) Epoch 31, batch 1200, loss[loss=0.1442, simple_loss=0.228, pruned_loss=0.03018, over 19709.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2436, pruned_loss=0.04302, over 3939047.69 frames. ], batch size: 78, lr: 5.04e-03, grad_scale: 32.0
+2023-03-29 02:53:06,351 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56864.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:53:45,290 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56879.0, num_to_drop=1, layers_to_drop={2}
+2023-03-29 02:54:12,175 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8820, 4.7718, 5.3159, 4.8760, 4.2203, 5.0584, 4.9805, 5.5076],
+       device='cuda:3'), covar=tensor([0.0935, 0.0384, 0.0391, 0.0433, 0.0870, 0.0563, 0.0457, 0.0327],
+       device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0223, 0.0223, 0.0236, 0.0208, 0.0245, 0.0233, 0.0219],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 02:54:12,330 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4599, 3.6052, 2.2785, 4.3166, 3.8250, 4.2459, 4.2692, 3.2182],
+       device='cuda:3'), covar=tensor([0.0654, 0.0596, 0.1573, 0.0478, 0.0597, 0.0367, 0.0619, 0.0786],
+       device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0143, 0.0143, 0.0151, 0.0132, 0.0134, 0.0146, 0.0144],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 02:54:42,032 INFO [train.py:892] (3/4) Epoch 31, batch 1250, loss[loss=0.1973, simple_loss=0.2734, pruned_loss=0.06057, over 19645.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2436, pruned_loss=0.04303, over 3941893.66 frames. ], batch size: 351, lr: 5.04e-03, grad_scale: 32.0
+2023-03-29 02:54:54,332 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3889, 4.1296, 4.1779, 3.9497, 4.3652, 3.0761, 3.6680, 2.0646],
+       device='cuda:3'), covar=tensor([0.0170, 0.0238, 0.0152, 0.0199, 0.0138, 0.1006, 0.0708, 0.1568],
+       device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0148, 0.0114, 0.0135, 0.0120, 0.0136, 0.0144, 0.0128],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 02:55:03,275 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56908.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:55:17,014 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0
+2023-03-29 02:55:43,170 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.950e+02 3.644e+02 4.260e+02 5.333e+02 1.198e+03, threshold=8.519e+02, percent-clipped=0.0
+2023-03-29 02:56:45,863 INFO [train.py:892] (3/4) Epoch 31, batch 1300, loss[loss=0.2037, simple_loss=0.27, pruned_loss=0.06877, over 19748.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2452, pruned_loss=0.04374, over 3939769.75 frames. ], batch size: 291, lr: 5.03e-03, grad_scale: 32.0
+2023-03-29 02:57:09,471 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9897, 3.1566, 3.1971, 3.1843, 3.0257, 3.1259, 3.0718, 3.2344],
+       device='cuda:3'), covar=tensor([0.0304, 0.0384, 0.0325, 0.0294, 0.0396, 0.0347, 0.0329, 0.0364],
+       device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0081, 0.0084, 0.0078, 0.0091, 0.0084, 0.0101, 0.0073],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 02:57:31,481 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56969.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:58:08,060 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8151, 2.9357, 3.5562, 3.0860, 3.9948, 3.9157, 4.6050, 5.0810],
+       device='cuda:3'), covar=tensor([0.0447, 0.1873, 0.1385, 0.2077, 0.1553, 0.1329, 0.0501, 0.0386],
+       device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0237, 0.0266, 0.0250, 0.0292, 0.0256, 0.0229, 0.0254],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 02:58:27,110 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56993.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:58:33,576 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56995.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 02:58:45,967 INFO [train.py:892] (3/4) Epoch 31, batch 1350, loss[loss=0.1882, simple_loss=0.2669, pruned_loss=0.05474, over 19684.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2438, pruned_loss=0.04315, over 3941096.24 frames. ], batch size: 59, lr: 5.03e-03, grad_scale: 32.0
+2023-03-29 02:59:37,596 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.342e+02 3.868e+02 4.711e+02 5.636e+02 9.457e+02, threshold=9.422e+02, percent-clipped=6.0
+2023-03-29 03:00:14,797 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5820, 3.7179, 2.2309, 3.8955, 4.0017, 1.8631, 3.3068, 3.0693],
+       device='cuda:3'), covar=tensor([0.0809, 0.0898, 0.2959, 0.0807, 0.0684, 0.2859, 0.1081, 0.0965],
+       device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0259, 0.0232, 0.0278, 0.0257, 0.0206, 0.0241, 0.0200],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 03:00:41,600 INFO [train.py:892] (3/4) Epoch 31, batch 1400, loss[loss=0.1367, simple_loss=0.2128, pruned_loss=0.0303, over 19611.00 frames. ], tot_loss[loss=0.1636, simple_loss=0.2423, pruned_loss=0.04249, over 3942639.10 frames. ], batch size: 46, lr: 5.03e-03, grad_scale: 32.0
+2023-03-29 03:00:50,340 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:00:54,565 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57056.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:02:21,170 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0402, 3.0219, 4.9245, 4.1822, 4.5735, 4.8751, 4.6417, 4.4985],
+       device='cuda:3'), covar=tensor([0.0455, 0.0907, 0.0093, 0.0914, 0.0128, 0.0179, 0.0149, 0.0143],
+       device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0104, 0.0089, 0.0154, 0.0085, 0.0098, 0.0091, 0.0086],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 03:02:44,879 INFO [train.py:892] (3/4) Epoch 31, batch 1450, loss[loss=0.1461, simple_loss=0.2281, pruned_loss=0.03209, over 19620.00 frames. ], tot_loss[loss=0.1625, simple_loss=0.2414, pruned_loss=0.04184, over 3946014.10 frames. ], batch size: 65, lr: 5.03e-03, grad_scale: 32.0
+2023-03-29 03:03:41,326 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.443e+02 3.544e+02 4.145e+02 5.196e+02 9.297e+02, threshold=8.291e+02, percent-clipped=0.0
+2023-03-29 03:03:51,178 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6921, 2.4559, 2.9422, 2.5075, 3.0633, 2.9644, 3.5690, 3.8744],
+       device='cuda:3'), covar=tensor([0.0640, 0.2002, 0.1602, 0.2331, 0.1585, 0.1559, 0.0702, 0.0568],
+       device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0240, 0.0268, 0.0254, 0.0295, 0.0258, 0.0232, 0.0256],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 03:04:03,904 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4869, 2.4149, 4.4063, 3.7378, 4.1049, 4.3463, 4.0997, 4.0974],
+       device='cuda:3'), covar=tensor([0.0526, 0.1093, 0.0107, 0.0806, 0.0150, 0.0208, 0.0203, 0.0170],
+       device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0104, 0.0089, 0.0153, 0.0085, 0.0098, 0.0090, 0.0086],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 03:04:44,659 INFO [train.py:892] (3/4) Epoch 31, batch 1500, loss[loss=0.1688, simple_loss=0.2449, pruned_loss=0.0464, over 19739.00 frames. ], tot_loss[loss=0.1628, simple_loss=0.2416, pruned_loss=0.042, over 3946687.28 frames. ], batch size: 92, lr: 5.02e-03, grad_scale: 32.0
+2023-03-29 03:05:08,650 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57159.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:05:51,048 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.58 vs. limit=5.0
+2023-03-29 03:05:55,125 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57179.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 03:06:12,942 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:06:22,199 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0
+2023-03-29 03:06:41,364 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57197.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:06:49,044 INFO [train.py:892] (3/4) Epoch 31, batch 1550, loss[loss=0.1498, simple_loss=0.2307, pruned_loss=0.03442, over 19872.00 frames. ], tot_loss[loss=0.1625, simple_loss=0.2416, pruned_loss=0.04167, over 3946610.88 frames. ], batch size: 108, lr: 5.02e-03, grad_scale: 32.0
+2023-03-29 03:07:43,734 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.504e+02 3.735e+02 4.518e+02 5.755e+02 1.077e+03, threshold=9.035e+02, percent-clipped=5.0
+2023-03-29 03:07:52,668 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57227.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 03:08:30,511 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57244.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:08:38,675 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57247.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:08:46,365 INFO [train.py:892] (3/4) Epoch 31, batch 1600, loss[loss=0.1468, simple_loss=0.2261, pruned_loss=0.03375, over 19893.00 frames. ], tot_loss[loss=0.1615, simple_loss=0.2408, pruned_loss=0.04108, over 3947842.65 frames. ], batch size: 87, lr: 5.02e-03, grad_scale: 32.0
+2023-03-29 03:08:53,877 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1954, 4.1681, 4.5633, 4.3371, 4.5022, 3.9690, 4.3105, 4.1105],
+       device='cuda:3'), covar=tensor([0.1486, 0.1798, 0.1049, 0.1377, 0.0960, 0.1201, 0.1850, 0.2182],
+       device='cuda:3'), in_proj_covar=tensor([0.0297, 0.0336, 0.0373, 0.0301, 0.0278, 0.0286, 0.0362, 0.0392],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 03:09:04,013 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57258.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 03:09:20,652 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57264.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:10:45,146 INFO [train.py:892] (3/4) Epoch 31, batch 1650, loss[loss=0.1658, simple_loss=0.2616, pruned_loss=0.03495, over 19849.00 frames. ], tot_loss[loss=0.1623, simple_loss=0.2413, pruned_loss=0.04168, over 3949360.89 frames. ], batch size: 56, lr: 5.02e-03, grad_scale: 32.0
+2023-03-29 03:10:58,478 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57305.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:11:38,983 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.517e+02 3.692e+02 4.492e+02 5.232e+02 9.843e+02, threshold=8.984e+02, percent-clipped=1.0
+2023-03-29 03:12:37,208 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57349.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:12:40,563 INFO [train.py:892] (3/4) Epoch 31, batch 1700, loss[loss=0.164, simple_loss=0.2417, pruned_loss=0.04314, over 19764.00 frames. ], tot_loss[loss=0.1619, simple_loss=0.2407, pruned_loss=0.04152, over 3950420.80 frames. ], batch size: 233, lr: 5.02e-03, grad_scale: 32.0
+2023-03-29 03:12:41,489 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57351.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:14:26,288 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57398.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:14:31,699 INFO [train.py:892] (3/4) Epoch 31, batch 1750, loss[loss=0.1632, simple_loss=0.2421, pruned_loss=0.04216, over 19826.00 frames. ], tot_loss[loss=0.1613, simple_loss=0.2398, pruned_loss=0.04137, over 3951094.33 frames. ], batch size: 208, lr: 5.01e-03, grad_scale: 32.0
+2023-03-29 03:15:05,705 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4511, 4.6276, 2.7557, 4.8376, 5.0181, 2.1947, 4.3268, 3.6900],
+       device='cuda:3'), covar=tensor([0.0607, 0.0761, 0.2511, 0.0690, 0.0456, 0.2745, 0.0895, 0.0836],
+       device='cuda:3'), in_proj_covar=tensor([0.0233, 0.0257, 0.0230, 0.0276, 0.0254, 0.0203, 0.0239, 0.0198],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 03:15:20,064 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.125e+02 3.770e+02 4.540e+02 5.122e+02 1.411e+03, threshold=9.080e+02, percent-clipped=2.0
+2023-03-29 03:16:11,422 INFO [train.py:892] (3/4) Epoch 31, batch 1800, loss[loss=0.1585, simple_loss=0.2241, pruned_loss=0.04646, over 19840.00 frames. ], tot_loss[loss=0.1623, simple_loss=0.2411, pruned_loss=0.04175, over 3949364.69 frames. ], batch size: 160, lr: 5.01e-03, grad_scale: 16.0
+2023-03-29 03:16:27,886 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:16:28,001 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57459.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:16:33,399 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57462.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:17:02,552 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2510, 3.5735, 3.0245, 2.6243, 3.1264, 3.5270, 3.4206, 3.4387],
+       device='cuda:3'), covar=tensor([0.0258, 0.0231, 0.0288, 0.0529, 0.0335, 0.0206, 0.0230, 0.0246],
+       device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0097, 0.0100, 0.0103, 0.0106, 0.0087, 0.0087, 0.0087],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 03:17:33,904 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:17:41,494 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0
+2023-03-29 03:17:44,836 INFO [train.py:892] (3/4) Epoch 31, batch 1850, loss[loss=0.1481, simple_loss=0.2305, pruned_loss=0.03287, over 19845.00 frames. ], tot_loss[loss=0.1633, simple_loss=0.2426, pruned_loss=0.04199, over 3949355.66 frames. ], batch size: 58, lr: 5.01e-03, grad_scale: 16.0
+2023-03-29 03:18:57,379 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.1534, 4.7241, 4.7894, 4.5292, 5.0771, 3.2920, 4.1790, 2.5625],
+       device='cuda:3'), covar=tensor([0.0157, 0.0197, 0.0138, 0.0178, 0.0118, 0.0936, 0.0781, 0.1412],
+       device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0147, 0.0114, 0.0135, 0.0119, 0.0136, 0.0143, 0.0128],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 03:18:58,442 INFO [train.py:892] (3/4) Epoch 32, batch 0, loss[loss=0.1517, simple_loss=0.2298, pruned_loss=0.0368, over 19637.00 frames. ], tot_loss[loss=0.1517, simple_loss=0.2298, pruned_loss=0.0368, over 19637.00 frames. ], batch size: 72, lr: 4.93e-03, grad_scale: 16.0
+2023-03-29 03:18:58,442 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-29 03:19:34,345 INFO [train.py:926] (3/4) Epoch 32, validation: loss=0.1821, simple_loss=0.2499, pruned_loss=0.05717, over 2883724.00 frames. 
+2023-03-29 03:19:34,347 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-29 03:19:37,652 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57507.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:20:17,652 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57523.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:20:21,036 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.254e+02 3.434e+02 4.255e+02 5.001e+02 8.278e+02, threshold=8.509e+02, percent-clipped=0.0
+2023-03-29 03:21:01,597 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57542.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:21:28,459 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57553.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 03:21:33,947 INFO [train.py:892] (3/4) Epoch 32, batch 50, loss[loss=0.1503, simple_loss=0.2356, pruned_loss=0.03248, over 19885.00 frames. ], tot_loss[loss=0.1545, simple_loss=0.2323, pruned_loss=0.0384, over 892334.78 frames. ], batch size: 47, lr: 4.93e-03, grad_scale: 16.0
+2023-03-29 03:21:35,062 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:21:53,724 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57564.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:21:58,285 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6712, 3.8353, 2.4890, 4.5172, 4.0266, 4.4540, 4.5166, 3.5329],
+       device='cuda:3'), covar=tensor([0.0562, 0.0500, 0.1346, 0.0596, 0.0547, 0.0396, 0.0456, 0.0695],
+       device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0144, 0.0143, 0.0153, 0.0133, 0.0135, 0.0147, 0.0145],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 03:23:18,868 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57600.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:23:31,175 INFO [train.py:892] (3/4) Epoch 32, batch 100, loss[loss=0.1619, simple_loss=0.2473, pruned_loss=0.03825, over 19730.00 frames. ], tot_loss[loss=0.158, simple_loss=0.2364, pruned_loss=0.03981, over 1570933.37 frames. ], batch size: 63, lr: 4.92e-03, grad_scale: 16.0
+2023-03-29 03:23:47,402 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57612.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:24:17,753 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.599e+02 3.672e+02 4.484e+02 5.551e+02 1.135e+03, threshold=8.969e+02, percent-clipped=1.0
+2023-03-29 03:25:12,318 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:25:16,310 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:25:27,028 INFO [train.py:892] (3/4) Epoch 32, batch 150, loss[loss=0.1711, simple_loss=0.2399, pruned_loss=0.05112, over 19866.00 frames. ], tot_loss[loss=0.1567, simple_loss=0.2348, pruned_loss=0.03931, over 2099771.10 frames. ], batch size: 129, lr: 4.92e-03, grad_scale: 16.0
+2023-03-29 03:25:34,805 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0
+2023-03-29 03:25:41,448 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4425, 4.9080, 5.0939, 4.8442, 5.3530, 3.3908, 4.3353, 2.8640],
+       device='cuda:3'), covar=tensor([0.0149, 0.0186, 0.0129, 0.0163, 0.0127, 0.0834, 0.0814, 0.1317],
+       device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0148, 0.0114, 0.0136, 0.0120, 0.0136, 0.0144, 0.0128],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 03:26:32,282 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0
+2023-03-29 03:27:07,498 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57697.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:27:13,582 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:27:27,796 INFO [train.py:892] (3/4) Epoch 32, batch 200, loss[loss=0.1648, simple_loss=0.2368, pruned_loss=0.04642, over 19818.00 frames. ], tot_loss[loss=0.1586, simple_loss=0.237, pruned_loss=0.04005, over 2509986.26 frames. ], batch size: 202, lr: 4.92e-03, grad_scale: 16.0
+2023-03-29 03:28:10,858 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.596e+02 3.546e+02 4.264e+02 5.379e+02 8.670e+02, threshold=8.529e+02, percent-clipped=0.0
+2023-03-29 03:28:22,052 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8367, 4.5392, 4.5904, 4.8706, 4.5970, 4.9913, 4.9353, 5.1675],
+       device='cuda:3'), covar=tensor([0.0604, 0.0412, 0.0452, 0.0335, 0.0631, 0.0399, 0.0409, 0.0246],
+       device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0176, 0.0201, 0.0174, 0.0174, 0.0158, 0.0150, 0.0197],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 03:28:30,905 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57733.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:29:16,452 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57754.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:29:19,501 INFO [train.py:892] (3/4) Epoch 32, batch 250, loss[loss=0.1496, simple_loss=0.2186, pruned_loss=0.04034, over 19748.00 frames. ], tot_loss[loss=0.1594, simple_loss=0.2378, pruned_loss=0.04051, over 2829637.02 frames. ], batch size: 139, lr: 4.92e-03, grad_scale: 16.0
+2023-03-29 03:30:50,024 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57794.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:31:19,649 INFO [train.py:892] (3/4) Epoch 32, batch 300, loss[loss=0.1931, simple_loss=0.2699, pruned_loss=0.05811, over 19915.00 frames. ], tot_loss[loss=0.1602, simple_loss=0.2385, pruned_loss=0.04091, over 3077997.74 frames. ], batch size: 53, lr: 4.92e-03, grad_scale: 16.0
+2023-03-29 03:31:50,703 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:32:05,153 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.255e+02 3.451e+02 4.259e+02 5.251e+02 1.158e+03, threshold=8.517e+02, percent-clipped=3.0
+2023-03-29 03:32:48,062 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57842.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:33:09,067 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57851.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:33:13,467 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57853.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 03:33:19,325 INFO [train.py:892] (3/4) Epoch 32, batch 350, loss[loss=0.1446, simple_loss=0.2252, pruned_loss=0.03202, over 19799.00 frames. ], tot_loss[loss=0.1615, simple_loss=0.2398, pruned_loss=0.04161, over 3271327.37 frames. ], batch size: 193, lr: 4.91e-03, grad_scale: 16.0
+2023-03-29 03:34:27,286 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9314, 4.0525, 2.4848, 4.2401, 4.4611, 1.9907, 3.6220, 3.4288],
+       device='cuda:3'), covar=tensor([0.0743, 0.0798, 0.2664, 0.0824, 0.0548, 0.2669, 0.1050, 0.0833],
+       device='cuda:3'), in_proj_covar=tensor([0.0234, 0.0257, 0.0230, 0.0275, 0.0254, 0.0203, 0.0240, 0.0198],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 03:34:39,374 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57890.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:35:01,119 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:35:03,046 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:35:14,083 INFO [train.py:892] (3/4) Epoch 32, batch 400, loss[loss=0.2201, simple_loss=0.3362, pruned_loss=0.05202, over 17872.00 frames. ], tot_loss[loss=0.1609, simple_loss=0.2396, pruned_loss=0.04115, over 3419841.77 frames. ], batch size: 633, lr: 4.91e-03, grad_scale: 16.0
+2023-03-29 03:35:59,247 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.499e+02 3.927e+02 4.614e+02 5.637e+02 8.834e+02, threshold=9.228e+02, percent-clipped=1.0
+2023-03-29 03:36:53,745 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57948.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:37:11,498 INFO [train.py:892] (3/4) Epoch 32, batch 450, loss[loss=0.1787, simple_loss=0.2616, pruned_loss=0.04796, over 19643.00 frames. ], tot_loss[loss=0.1626, simple_loss=0.2416, pruned_loss=0.04179, over 3537787.85 frames. ], batch size: 299, lr: 4.91e-03, grad_scale: 16.0
+2023-03-29 03:37:42,684 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.1640, 5.4894, 5.5185, 5.4378, 5.1292, 5.4646, 4.9638, 4.9970],
+       device='cuda:3'), covar=tensor([0.0456, 0.0451, 0.0458, 0.0409, 0.0608, 0.0488, 0.0643, 0.0970],
+       device='cuda:3'), in_proj_covar=tensor([0.0271, 0.0288, 0.0300, 0.0263, 0.0265, 0.0250, 0.0270, 0.0314],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 03:39:10,417 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4579, 3.2040, 3.4817, 3.0052, 3.7118, 3.7389, 4.2634, 4.7123],
+       device='cuda:3'), covar=tensor([0.0481, 0.1491, 0.1377, 0.2132, 0.1542, 0.1258, 0.0581, 0.0508],
+       device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0240, 0.0268, 0.0255, 0.0297, 0.0258, 0.0232, 0.0257],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 03:39:11,339 INFO [train.py:892] (3/4) Epoch 32, batch 500, loss[loss=0.1654, simple_loss=0.2459, pruned_loss=0.04249, over 19877.00 frames. ], tot_loss[loss=0.1631, simple_loss=0.2421, pruned_loss=0.04209, over 3627692.07 frames. ], batch size: 84, lr: 4.91e-03, grad_scale: 16.0
+2023-03-29 03:39:55,699 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.669e+02 3.901e+02 4.425e+02 5.266e+02 9.447e+02, threshold=8.850e+02, percent-clipped=1.0
+2023-03-29 03:40:09,424 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5608, 3.4489, 3.4081, 3.1875, 3.5368, 2.6138, 2.8133, 1.6197],
+       device='cuda:3'), covar=tensor([0.0229, 0.0258, 0.0167, 0.0201, 0.0171, 0.1368, 0.0746, 0.1944],
+       device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0147, 0.0113, 0.0134, 0.0119, 0.0135, 0.0142, 0.0127],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 03:41:02,869 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58054.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:41:07,809 INFO [train.py:892] (3/4) Epoch 32, batch 550, loss[loss=0.2031, simple_loss=0.2813, pruned_loss=0.0624, over 19642.00 frames. ], tot_loss[loss=0.1636, simple_loss=0.2428, pruned_loss=0.04223, over 3697685.47 frames. ], batch size: 359, lr: 4.91e-03, grad_scale: 16.0
+2023-03-29 03:42:24,400 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58089.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:42:56,943 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58102.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:43:04,369 INFO [train.py:892] (3/4) Epoch 32, batch 600, loss[loss=0.1658, simple_loss=0.2391, pruned_loss=0.04625, over 19671.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.2427, pruned_loss=0.04214, over 3752668.39 frames. ], batch size: 64, lr: 4.90e-03, grad_scale: 16.0
+2023-03-29 03:43:30,607 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-29 03:43:34,464 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:43:48,670 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.610e+02 3.618e+02 4.223e+02 5.011e+02 6.745e+02, threshold=8.447e+02, percent-clipped=0.0
+2023-03-29 03:44:15,988 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7383, 3.4688, 3.7548, 3.0542, 4.0335, 3.3875, 3.5271, 3.8109],
+       device='cuda:3'), covar=tensor([0.0619, 0.0462, 0.0543, 0.0719, 0.0374, 0.0416, 0.0381, 0.0296],
+       device='cuda:3'), in_proj_covar=tensor([0.0079, 0.0085, 0.0084, 0.0110, 0.0079, 0.0082, 0.0080, 0.0073],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 03:44:27,919 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58141.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:44:33,807 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:44:49,998 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58151.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:44:50,033 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1475, 2.5937, 2.9437, 3.2402, 3.7090, 4.1069, 3.9408, 4.0620],
+       device='cuda:3'), covar=tensor([0.0851, 0.1600, 0.1392, 0.0684, 0.0424, 0.0238, 0.0400, 0.0399],
+       device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0169, 0.0179, 0.0152, 0.0138, 0.0133, 0.0125, 0.0116],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 03:45:00,228 INFO [train.py:892] (3/4) Epoch 32, batch 650, loss[loss=0.1554, simple_loss=0.2306, pruned_loss=0.04006, over 19827.00 frames. ], tot_loss[loss=0.1629, simple_loss=0.2417, pruned_loss=0.04202, over 3796265.32 frames. ], batch size: 121, lr: 4.90e-03, grad_scale: 16.0
+2023-03-29 03:45:19,355 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2406, 4.6151, 2.6555, 4.8685, 4.9957, 2.2640, 4.0506, 3.4839],
+       device='cuda:3'), covar=tensor([0.0858, 0.0659, 0.2783, 0.0592, 0.0527, 0.2975, 0.1146, 0.1001],
+       device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0259, 0.0232, 0.0277, 0.0256, 0.0205, 0.0241, 0.0199],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 03:45:21,014 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:45:21,281 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6262, 2.7314, 2.7380, 2.8129, 2.6250, 2.7852, 2.6604, 2.9429],
+       device='cuda:3'), covar=tensor([0.0357, 0.0357, 0.0338, 0.0287, 0.0443, 0.0315, 0.0409, 0.0330],
+       device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0081, 0.0083, 0.0078, 0.0090, 0.0083, 0.0100, 0.0073],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 03:45:49,673 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.58 vs. limit=5.0
+2023-03-29 03:46:37,033 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:46:43,956 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58202.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:46:52,168 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58205.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:46:52,977 INFO [train.py:892] (3/4) Epoch 32, batch 700, loss[loss=0.1529, simple_loss=0.2378, pruned_loss=0.03405, over 19882.00 frames. ], tot_loss[loss=0.163, simple_loss=0.2416, pruned_loss=0.04214, over 3831440.77 frames. ], batch size: 84, lr: 4.90e-03, grad_scale: 16.0
+2023-03-29 03:47:32,724 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0
+2023-03-29 03:47:39,106 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.613e+02 3.735e+02 4.349e+02 5.105e+02 9.277e+02, threshold=8.697e+02, percent-clipped=1.0
+2023-03-29 03:48:54,083 INFO [train.py:892] (3/4) Epoch 32, batch 750, loss[loss=0.1693, simple_loss=0.2482, pruned_loss=0.0452, over 19746.00 frames. ], tot_loss[loss=0.1621, simple_loss=0.2405, pruned_loss=0.04189, over 3858639.07 frames. ], batch size: 209, lr: 4.90e-03, grad_scale: 16.0
+2023-03-29 03:50:29,148 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3697, 2.6159, 3.7061, 2.9498, 3.0739, 2.9859, 2.1933, 2.3726],
+       device='cuda:3'), covar=tensor([0.1174, 0.3029, 0.0688, 0.1141, 0.1886, 0.1570, 0.2587, 0.2702],
+       device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0386, 0.0349, 0.0286, 0.0371, 0.0375, 0.0372, 0.0343],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 03:50:47,188 INFO [train.py:892] (3/4) Epoch 32, batch 800, loss[loss=0.1548, simple_loss=0.2298, pruned_loss=0.03989, over 19837.00 frames. ], tot_loss[loss=0.161, simple_loss=0.2394, pruned_loss=0.04135, over 3880401.22 frames. ], batch size: 177, lr: 4.90e-03, grad_scale: 16.0
+2023-03-29 03:51:28,607 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1745, 3.1106, 3.2124, 2.3648, 3.3328, 2.6880, 3.1026, 3.2084],
+       device='cuda:3'), covar=tensor([0.0555, 0.0405, 0.0519, 0.0930, 0.0344, 0.0566, 0.0491, 0.0331],
+       device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0086, 0.0084, 0.0111, 0.0080, 0.0083, 0.0081, 0.0074],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 03:51:31,608 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.700e+02 3.818e+02 4.452e+02 5.202e+02 1.002e+03, threshold=8.904e+02, percent-clipped=2.0
+2023-03-29 03:52:43,169 INFO [train.py:892] (3/4) Epoch 32, batch 850, loss[loss=0.1674, simple_loss=0.2435, pruned_loss=0.04565, over 19771.00 frames. ], tot_loss[loss=0.1615, simple_loss=0.2402, pruned_loss=0.04143, over 3895572.41 frames. ], batch size: 273, lr: 4.89e-03, grad_scale: 16.0
+2023-03-29 03:52:47,847 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6108, 3.0973, 3.5189, 3.0556, 3.8112, 3.8977, 4.4396, 4.9299],
+       device='cuda:3'), covar=tensor([0.0487, 0.1765, 0.1522, 0.2233, 0.1647, 0.1218, 0.0641, 0.0487],
+       device='cuda:3'), in_proj_covar=tensor([0.0258, 0.0243, 0.0271, 0.0256, 0.0299, 0.0260, 0.0235, 0.0260],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 03:52:59,832 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3719, 4.1749, 4.1750, 3.9456, 4.3561, 2.9695, 3.6306, 2.2595],
+       device='cuda:3'), covar=tensor([0.0182, 0.0216, 0.0142, 0.0187, 0.0135, 0.1087, 0.0623, 0.1474],
+       device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0146, 0.0113, 0.0134, 0.0119, 0.0136, 0.0142, 0.0127],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 03:53:49,675 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0
+2023-03-29 03:54:00,027 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:54:11,139 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.07 vs. limit=2.0
+2023-03-29 03:54:39,279 INFO [train.py:892] (3/4) Epoch 32, batch 900, loss[loss=0.1857, simple_loss=0.2778, pruned_loss=0.04685, over 19822.00 frames. ], tot_loss[loss=0.1616, simple_loss=0.2406, pruned_loss=0.04127, over 3907654.25 frames. ], batch size: 57, lr: 4.89e-03, grad_scale: 16.0
+2023-03-29 03:55:02,929 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6513, 4.7137, 2.8420, 4.9170, 5.1232, 2.1867, 4.3825, 3.7302],
+       device='cuda:3'), covar=tensor([0.0610, 0.0693, 0.2418, 0.0652, 0.0499, 0.2635, 0.0904, 0.0772],
+       device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0257, 0.0230, 0.0275, 0.0255, 0.0203, 0.0239, 0.0197],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 03:55:10,039 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3268, 4.4535, 2.6735, 4.6284, 4.8297, 2.1663, 4.1220, 3.6528],
+       device='cuda:3'), covar=tensor([0.0660, 0.0813, 0.2539, 0.0839, 0.0580, 0.2644, 0.0917, 0.0793],
+       device='cuda:3'), in_proj_covar=tensor([0.0232, 0.0256, 0.0230, 0.0275, 0.0255, 0.0203, 0.0239, 0.0197],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 03:55:21,926 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.766e+02 3.700e+02 4.222e+02 4.872e+02 9.847e+02, threshold=8.445e+02, percent-clipped=1.0
+2023-03-29 03:55:51,527 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58437.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:55:58,838 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.5952, 1.5361, 1.6629, 1.6685, 1.5461, 1.6259, 1.5169, 1.6892],
+       device='cuda:3'), covar=tensor([0.0435, 0.0388, 0.0376, 0.0324, 0.0522, 0.0362, 0.0530, 0.0327],
+       device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0081, 0.0083, 0.0078, 0.0090, 0.0083, 0.0100, 0.0073],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 03:56:33,913 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7088, 2.7457, 4.7990, 4.0060, 4.5639, 4.6874, 4.5898, 4.4385],
+       device='cuda:3'), covar=tensor([0.0538, 0.1040, 0.0098, 0.0953, 0.0135, 0.0184, 0.0151, 0.0149],
+       device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0104, 0.0089, 0.0153, 0.0086, 0.0098, 0.0090, 0.0086],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 03:56:34,990 INFO [train.py:892] (3/4) Epoch 32, batch 950, loss[loss=0.1566, simple_loss=0.2478, pruned_loss=0.03272, over 19838.00 frames. ], tot_loss[loss=0.1623, simple_loss=0.2417, pruned_loss=0.04141, over 3913648.28 frames. ], batch size: 57, lr: 4.89e-03, grad_scale: 16.0
+2023-03-29 03:58:05,418 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58497.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:58:12,043 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58500.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 03:58:20,974 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0
+2023-03-29 03:58:24,204 INFO [train.py:892] (3/4) Epoch 32, batch 1000, loss[loss=0.1868, simple_loss=0.2708, pruned_loss=0.05136, over 19734.00 frames. ], tot_loss[loss=0.1624, simple_loss=0.2417, pruned_loss=0.04154, over 3920491.33 frames. ], batch size: 295, lr: 4.89e-03, grad_scale: 16.0
+2023-03-29 03:59:09,120 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.374e+02 3.686e+02 4.298e+02 5.253e+02 9.198e+02, threshold=8.596e+02, percent-clipped=1.0
+2023-03-29 03:59:34,662 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.16 vs. limit=5.0
+2023-03-29 04:00:20,520 INFO [train.py:892] (3/4) Epoch 32, batch 1050, loss[loss=0.1595, simple_loss=0.2386, pruned_loss=0.04021, over 19747.00 frames. ], tot_loss[loss=0.1622, simple_loss=0.2417, pruned_loss=0.04138, over 3925317.04 frames. ], batch size: 221, lr: 4.88e-03, grad_scale: 16.0
+2023-03-29 04:02:18,312 INFO [train.py:892] (3/4) Epoch 32, batch 1100, loss[loss=0.1537, simple_loss=0.2358, pruned_loss=0.03579, over 19810.00 frames. ], tot_loss[loss=0.1621, simple_loss=0.2415, pruned_loss=0.04136, over 3930468.31 frames. ], batch size: 72, lr: 4.88e-03, grad_scale: 16.0
+2023-03-29 04:03:01,462 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.871e+02 3.808e+02 4.430e+02 5.351e+02 1.082e+03, threshold=8.860e+02, percent-clipped=2.0
+2023-03-29 04:04:08,087 INFO [train.py:892] (3/4) Epoch 32, batch 1150, loss[loss=0.2533, simple_loss=0.3364, pruned_loss=0.08507, over 19244.00 frames. ], tot_loss[loss=0.1626, simple_loss=0.2416, pruned_loss=0.04183, over 3934578.01 frames. ], batch size: 483, lr: 4.88e-03, grad_scale: 16.0
+2023-03-29 04:05:05,323 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58681.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:06:02,816 INFO [train.py:892] (3/4) Epoch 32, batch 1200, loss[loss=0.1502, simple_loss=0.226, pruned_loss=0.0372, over 19693.00 frames. ], tot_loss[loss=0.1626, simple_loss=0.2413, pruned_loss=0.0419, over 3936413.26 frames. ], batch size: 45, lr: 4.88e-03, grad_scale: 16.0
+2023-03-29 04:06:46,112 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.511e+02 3.575e+02 4.310e+02 5.528e+02 1.202e+03, threshold=8.619e+02, percent-clipped=2.0
+2023-03-29 04:06:47,037 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3136, 4.0093, 4.1508, 4.3459, 4.0257, 4.3402, 4.4076, 4.6069],
+       device='cuda:3'), covar=tensor([0.0685, 0.0441, 0.0494, 0.0391, 0.0697, 0.0531, 0.0448, 0.0284],
+       device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0179, 0.0203, 0.0178, 0.0176, 0.0161, 0.0152, 0.0200],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 04:07:29,749 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58742.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:07:57,925 INFO [train.py:892] (3/4) Epoch 32, batch 1250, loss[loss=0.1399, simple_loss=0.2193, pruned_loss=0.03023, over 19825.00 frames. ], tot_loss[loss=0.162, simple_loss=0.2405, pruned_loss=0.04178, over 3939576.33 frames. ], batch size: 93, lr: 4.88e-03, grad_scale: 16.0
+2023-03-29 04:08:46,422 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58775.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:09:34,863 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:09:41,909 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:09:55,339 INFO [train.py:892] (3/4) Epoch 32, batch 1300, loss[loss=0.1509, simple_loss=0.2239, pruned_loss=0.03895, over 19819.00 frames. ], tot_loss[loss=0.1611, simple_loss=0.2397, pruned_loss=0.04126, over 3943875.68 frames. ], batch size: 166, lr: 4.87e-03, grad_scale: 16.0
+2023-03-29 04:10:37,439 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.259e+02 3.414e+02 4.354e+02 4.908e+02 8.205e+02, threshold=8.708e+02, percent-clipped=0.0
+2023-03-29 04:11:01,258 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58836.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:11:07,409 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3447, 3.2437, 3.5937, 3.2947, 3.1449, 3.5554, 3.4185, 3.6598],
+       device='cuda:3'), covar=tensor([0.0839, 0.0419, 0.0407, 0.0456, 0.1554, 0.0609, 0.0488, 0.0408],
+       device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0225, 0.0225, 0.0237, 0.0210, 0.0249, 0.0235, 0.0221],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 04:11:18,734 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.42 vs. limit=5.0
+2023-03-29 04:11:20,361 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:11:28,504 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:11:34,844 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.63 vs. limit=5.0
+2023-03-29 04:11:46,424 INFO [train.py:892] (3/4) Epoch 32, batch 1350, loss[loss=0.1596, simple_loss=0.2406, pruned_loss=0.03926, over 19787.00 frames. ], tot_loss[loss=0.1617, simple_loss=0.2406, pruned_loss=0.0414, over 3945153.12 frames. ], batch size: 48, lr: 4.87e-03, grad_scale: 16.0
+2023-03-29 04:13:32,140 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3034, 2.4574, 4.2931, 3.7083, 4.1504, 4.2187, 4.0747, 4.0151],
+       device='cuda:3'), covar=tensor([0.0544, 0.1029, 0.0112, 0.0660, 0.0137, 0.0223, 0.0178, 0.0175],
+       device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0104, 0.0090, 0.0155, 0.0087, 0.0099, 0.0091, 0.0086],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 04:13:39,902 INFO [train.py:892] (3/4) Epoch 32, batch 1400, loss[loss=0.2184, simple_loss=0.3013, pruned_loss=0.06772, over 19622.00 frames. ], tot_loss[loss=0.1619, simple_loss=0.2406, pruned_loss=0.04161, over 3946891.08 frames. ], batch size: 387, lr: 4.87e-03, grad_scale: 16.0
+2023-03-29 04:13:41,180 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:14:27,791 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.427e+02 3.533e+02 4.416e+02 5.664e+02 1.235e+03, threshold=8.833e+02, percent-clipped=5.0
+2023-03-29 04:14:59,215 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8310, 2.8339, 2.8869, 2.4049, 3.0369, 2.6200, 2.9749, 3.0252],
+       device='cuda:3'), covar=tensor([0.0650, 0.0488, 0.0685, 0.0810, 0.0451, 0.0512, 0.0442, 0.0333],
+       device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0087, 0.0085, 0.0112, 0.0080, 0.0083, 0.0081, 0.0074],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 04:15:40,821 INFO [train.py:892] (3/4) Epoch 32, batch 1450, loss[loss=0.1248, simple_loss=0.2013, pruned_loss=0.02417, over 19853.00 frames. ], tot_loss[loss=0.1615, simple_loss=0.2404, pruned_loss=0.04132, over 3947381.50 frames. ], batch size: 118, lr: 4.87e-03, grad_scale: 16.0
+2023-03-29 04:16:06,829 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58967.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:16:59,288 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5920, 2.1549, 2.4193, 2.8335, 3.2657, 3.3799, 3.2699, 3.3637],
+       device='cuda:3'), covar=tensor([0.1106, 0.1697, 0.1473, 0.0815, 0.0543, 0.0377, 0.0453, 0.0470],
+       device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0169, 0.0178, 0.0153, 0.0137, 0.0134, 0.0126, 0.0117],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 04:17:08,521 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9269, 2.9650, 4.3527, 3.2897, 3.5476, 3.3723, 2.4243, 2.6145],
+       device='cuda:3'), covar=tensor([0.1063, 0.3278, 0.0575, 0.1227, 0.1969, 0.1542, 0.2909, 0.2815],
+       device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0390, 0.0351, 0.0288, 0.0376, 0.0377, 0.0374, 0.0346],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 04:17:32,726 INFO [train.py:892] (3/4) Epoch 32, batch 1500, loss[loss=0.1603, simple_loss=0.246, pruned_loss=0.03725, over 19832.00 frames. ], tot_loss[loss=0.1612, simple_loss=0.2399, pruned_loss=0.0413, over 3948144.13 frames. ], batch size: 57, lr: 4.87e-03, grad_scale: 16.0
+2023-03-29 04:18:18,936 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.783e+02 3.777e+02 4.436e+02 5.379e+02 1.000e+03, threshold=8.872e+02, percent-clipped=1.0
+2023-03-29 04:18:48,036 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:19:27,080 INFO [train.py:892] (3/4) Epoch 32, batch 1550, loss[loss=0.1436, simple_loss=0.2292, pruned_loss=0.02903, over 19795.00 frames. ], tot_loss[loss=0.1606, simple_loss=0.2396, pruned_loss=0.04079, over 3947936.50 frames. ], batch size: 79, lr: 4.86e-03, grad_scale: 16.0
+2023-03-29 04:20:42,849 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8974, 3.9008, 2.3993, 4.1048, 4.2853, 1.9297, 3.5776, 3.2754],
+       device='cuda:3'), covar=tensor([0.0689, 0.0837, 0.2626, 0.0768, 0.0593, 0.2654, 0.1036, 0.0835],
+       device='cuda:3'), in_proj_covar=tensor([0.0235, 0.0259, 0.0231, 0.0278, 0.0257, 0.0205, 0.0242, 0.0198],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 04:21:26,063 INFO [train.py:892] (3/4) Epoch 32, batch 1600, loss[loss=0.2564, simple_loss=0.3272, pruned_loss=0.09282, over 19472.00 frames. ], tot_loss[loss=0.1619, simple_loss=0.2411, pruned_loss=0.04136, over 3946734.17 frames. ], batch size: 396, lr: 4.86e-03, grad_scale: 16.0
+2023-03-29 04:22:08,643 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.456e+02 3.560e+02 4.342e+02 5.121e+02 7.778e+02, threshold=8.685e+02, percent-clipped=0.0
+2023-03-29 04:22:23,663 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59131.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:23:16,342 INFO [train.py:892] (3/4) Epoch 32, batch 1650, loss[loss=0.1404, simple_loss=0.2202, pruned_loss=0.03036, over 19857.00 frames. ], tot_loss[loss=0.1613, simple_loss=0.2404, pruned_loss=0.04109, over 3947131.14 frames. ], batch size: 104, lr: 4.86e-03, grad_scale: 16.0
+2023-03-29 04:23:33,830 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:25:14,206 INFO [train.py:892] (3/4) Epoch 32, batch 1700, loss[loss=0.16, simple_loss=0.2342, pruned_loss=0.0429, over 19811.00 frames. ], tot_loss[loss=0.1611, simple_loss=0.2402, pruned_loss=0.04097, over 3946582.30 frames. ], batch size: 181, lr: 4.86e-03, grad_scale: 16.0
+2023-03-29 04:26:00,352 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.757e+02 3.824e+02 4.316e+02 5.380e+02 8.214e+02, threshold=8.632e+02, percent-clipped=0.0
+2023-03-29 04:26:01,441 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:27:06,758 INFO [train.py:892] (3/4) Epoch 32, batch 1750, loss[loss=0.1614, simple_loss=0.2365, pruned_loss=0.04314, over 19630.00 frames. ], tot_loss[loss=0.1614, simple_loss=0.2403, pruned_loss=0.04125, over 3945483.63 frames. ], batch size: 68, lr: 4.86e-03, grad_scale: 16.0
+2023-03-29 04:27:18,831 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59262.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:28:46,844 INFO [train.py:892] (3/4) Epoch 32, batch 1800, loss[loss=0.1426, simple_loss=0.219, pruned_loss=0.03308, over 19765.00 frames. ], tot_loss[loss=0.1613, simple_loss=0.2401, pruned_loss=0.04122, over 3945316.61 frames. ], batch size: 102, lr: 4.85e-03, grad_scale: 16.0
+2023-03-29 04:29:23,551 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.122e+02 3.692e+02 4.509e+02 5.287e+02 9.435e+02, threshold=9.017e+02, percent-clipped=2.0
+2023-03-29 04:29:40,846 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.75 vs. limit=5.0
+2023-03-29 04:29:45,489 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:30:19,889 INFO [train.py:892] (3/4) Epoch 32, batch 1850, loss[loss=0.18, simple_loss=0.2611, pruned_loss=0.04941, over 19808.00 frames. ], tot_loss[loss=0.1614, simple_loss=0.241, pruned_loss=0.0409, over 3945435.22 frames. ], batch size: 57, lr: 4.85e-03, grad_scale: 16.0
+2023-03-29 04:31:27,297 INFO [train.py:892] (3/4) Epoch 33, batch 0, loss[loss=0.1348, simple_loss=0.2114, pruned_loss=0.02905, over 19820.00 frames. ], tot_loss[loss=0.1348, simple_loss=0.2114, pruned_loss=0.02905, over 19820.00 frames. ], batch size: 98, lr: 4.78e-03, grad_scale: 16.0
+2023-03-29 04:31:27,298 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-29 04:32:02,114 INFO [train.py:926] (3/4) Epoch 33, validation: loss=0.1828, simple_loss=0.2501, pruned_loss=0.05775, over 2883724.00 frames. 
+2023-03-29 04:32:02,115 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-29 04:32:58,919 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59385.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:34:00,948 INFO [train.py:892] (3/4) Epoch 33, batch 50, loss[loss=0.1589, simple_loss=0.2372, pruned_loss=0.04032, over 19744.00 frames. ], tot_loss[loss=0.1535, simple_loss=0.2315, pruned_loss=0.03775, over 891237.37 frames. ], batch size: 84, lr: 4.77e-03, grad_scale: 32.0
+2023-03-29 04:34:31,980 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.011e+02 3.734e+02 4.113e+02 4.881e+02 1.279e+03, threshold=8.226e+02, percent-clipped=1.0
+2023-03-29 04:34:45,861 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59431.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:35:56,489 INFO [train.py:892] (3/4) Epoch 33, batch 100, loss[loss=0.1727, simple_loss=0.2556, pruned_loss=0.04488, over 19858.00 frames. ], tot_loss[loss=0.1596, simple_loss=0.2389, pruned_loss=0.04015, over 1568176.27 frames. ], batch size: 60, lr: 4.77e-03, grad_scale: 32.0
+2023-03-29 04:36:10,323 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1762, 2.9634, 3.2549, 2.6055, 3.3546, 2.8433, 3.1233, 3.1904],
+       device='cuda:3'), covar=tensor([0.0599, 0.0541, 0.0592, 0.0799, 0.0386, 0.0502, 0.0542, 0.0430],
+       device='cuda:3'), in_proj_covar=tensor([0.0080, 0.0088, 0.0086, 0.0114, 0.0081, 0.0084, 0.0082, 0.0075],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 04:36:37,047 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59479.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:37:08,524 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0287, 3.3629, 3.5053, 3.9328, 2.7624, 3.3208, 2.5476, 2.5441],
+       device='cuda:3'), covar=tensor([0.0504, 0.1868, 0.0889, 0.0446, 0.1920, 0.0862, 0.1362, 0.1654],
+       device='cuda:3'), in_proj_covar=tensor([0.0247, 0.0331, 0.0251, 0.0206, 0.0250, 0.0210, 0.0220, 0.0219],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 04:37:32,234 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59502.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:37:51,804 INFO [train.py:892] (3/4) Epoch 33, batch 150, loss[loss=0.1493, simple_loss=0.2188, pruned_loss=0.03986, over 19836.00 frames. ], tot_loss[loss=0.1595, simple_loss=0.2384, pruned_loss=0.04035, over 2096795.67 frames. ], batch size: 146, lr: 4.77e-03, grad_scale: 32.0
+2023-03-29 04:38:07,257 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0
+2023-03-29 04:38:12,744 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59520.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:38:22,527 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.374e+02 3.870e+02 4.435e+02 5.412e+02 1.132e+03, threshold=8.870e+02, percent-clipped=1.0
+2023-03-29 04:39:44,917 INFO [train.py:892] (3/4) Epoch 33, batch 200, loss[loss=0.134, simple_loss=0.2107, pruned_loss=0.02864, over 19735.00 frames. ], tot_loss[loss=0.1592, simple_loss=0.2384, pruned_loss=0.03999, over 2508672.70 frames. ], batch size: 44, lr: 4.77e-03, grad_scale: 32.0
+2023-03-29 04:39:48,067 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59562.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:39:50,157 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59563.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:41:35,584 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59610.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:41:36,991 INFO [train.py:892] (3/4) Epoch 33, batch 250, loss[loss=0.1472, simple_loss=0.2287, pruned_loss=0.03286, over 19864.00 frames. ], tot_loss[loss=0.1594, simple_loss=0.2388, pruned_loss=0.03997, over 2828192.26 frames. ], batch size: 46, lr: 4.77e-03, grad_scale: 32.0
+2023-03-29 04:41:48,590 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-29 04:42:08,470 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.374e+02 3.603e+02 4.122e+02 4.867e+02 8.945e+02, threshold=8.243e+02, percent-clipped=1.0
+2023-03-29 04:43:31,402 INFO [train.py:892] (3/4) Epoch 33, batch 300, loss[loss=0.1588, simple_loss=0.2333, pruned_loss=0.04217, over 19900.00 frames. ], tot_loss[loss=0.1598, simple_loss=0.2393, pruned_loss=0.04018, over 3076901.54 frames. ], batch size: 94, lr: 4.76e-03, grad_scale: 32.0
+2023-03-29 04:44:18,260 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1914, 2.9496, 3.2783, 2.8417, 3.4260, 3.3913, 3.9935, 4.3941],
+       device='cuda:3'), covar=tensor([0.0560, 0.1680, 0.1519, 0.2162, 0.1661, 0.1593, 0.0619, 0.0578],
+       device='cuda:3'), in_proj_covar=tensor([0.0256, 0.0241, 0.0269, 0.0254, 0.0298, 0.0257, 0.0233, 0.0257],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 04:45:17,711 INFO [train.py:892] (3/4) Epoch 33, batch 350, loss[loss=0.1461, simple_loss=0.2171, pruned_loss=0.03758, over 19858.00 frames. ], tot_loss[loss=0.1597, simple_loss=0.2394, pruned_loss=0.03999, over 3270809.70 frames. ], batch size: 158, lr: 4.76e-03, grad_scale: 32.0
+2023-03-29 04:45:51,033 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.193e+02 3.458e+02 4.083e+02 4.831e+02 8.519e+02, threshold=8.167e+02, percent-clipped=1.0
+2023-03-29 04:47:10,418 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.12 vs. limit=5.0
+2023-03-29 04:47:13,330 INFO [train.py:892] (3/4) Epoch 33, batch 400, loss[loss=0.1496, simple_loss=0.2304, pruned_loss=0.0344, over 19797.00 frames. ], tot_loss[loss=0.1603, simple_loss=0.2401, pruned_loss=0.04024, over 3419615.17 frames. ], batch size: 65, lr: 4.76e-03, grad_scale: 32.0
+2023-03-29 04:47:27,765 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.2084, 1.7790, 3.4326, 2.9604, 3.4682, 3.4645, 3.2441, 3.3901],
+       device='cuda:3'), covar=tensor([0.1131, 0.1654, 0.0177, 0.0655, 0.0181, 0.0326, 0.0314, 0.0257],
+       device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0105, 0.0089, 0.0155, 0.0087, 0.0100, 0.0092, 0.0087],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 04:47:38,011 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7365, 3.9936, 4.2115, 4.8619, 3.1429, 3.6107, 3.0147, 2.8938],
+       device='cuda:3'), covar=tensor([0.0448, 0.1777, 0.0793, 0.0335, 0.2056, 0.0982, 0.1255, 0.1605],
+       device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0328, 0.0250, 0.0204, 0.0248, 0.0209, 0.0219, 0.0216],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 04:48:23,420 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0384, 3.7789, 3.8298, 4.0167, 3.7738, 4.0429, 4.1541, 4.3227],
+       device='cuda:3'), covar=tensor([0.0777, 0.0463, 0.0563, 0.0452, 0.0804, 0.0581, 0.0462, 0.0323],
+       device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0176, 0.0200, 0.0176, 0.0174, 0.0159, 0.0150, 0.0197],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 04:49:08,632 INFO [train.py:892] (3/4) Epoch 33, batch 450, loss[loss=0.1431, simple_loss=0.2284, pruned_loss=0.02887, over 19884.00 frames. ], tot_loss[loss=0.1611, simple_loss=0.2411, pruned_loss=0.04056, over 3535357.41 frames. ], batch size: 95, lr: 4.76e-03, grad_scale: 32.0
+2023-03-29 04:49:28,970 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:49:39,997 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.419e+02 3.507e+02 4.283e+02 5.159e+02 1.029e+03, threshold=8.565e+02, percent-clipped=3.0
+2023-03-29 04:50:19,125 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0
+2023-03-29 04:50:33,117 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:50:50,701 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.81 vs. limit=5.0
+2023-03-29 04:50:54,176 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59858.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:50:59,542 INFO [train.py:892] (3/4) Epoch 33, batch 500, loss[loss=0.1399, simple_loss=0.2194, pruned_loss=0.03021, over 19813.00 frames. ], tot_loss[loss=0.1613, simple_loss=0.2409, pruned_loss=0.04085, over 3626276.07 frames. ], batch size: 96, lr: 4.76e-03, grad_scale: 32.0
+2023-03-29 04:51:20,676 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59868.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:52:06,982 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59889.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:52:52,876 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:52:55,882 INFO [train.py:892] (3/4) Epoch 33, batch 550, loss[loss=0.1433, simple_loss=0.2259, pruned_loss=0.03033, over 19749.00 frames. ], tot_loss[loss=0.162, simple_loss=0.2416, pruned_loss=0.04118, over 3696369.03 frames. ], batch size: 110, lr: 4.75e-03, grad_scale: 32.0
+2023-03-29 04:53:05,388 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6461, 4.4883, 4.4688, 4.2365, 4.6982, 3.1149, 3.8933, 2.3806],
+       device='cuda:3'), covar=tensor([0.0226, 0.0214, 0.0167, 0.0203, 0.0152, 0.0988, 0.0707, 0.1476],
+       device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0147, 0.0114, 0.0135, 0.0120, 0.0135, 0.0142, 0.0127],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 04:53:28,204 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.466e+02 3.637e+02 4.346e+02 5.121e+02 1.028e+03, threshold=8.693e+02, percent-clipped=1.0
+2023-03-29 04:54:26,674 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:54:50,041 INFO [train.py:892] (3/4) Epoch 33, batch 600, loss[loss=0.1408, simple_loss=0.2274, pruned_loss=0.02714, over 19782.00 frames. ], tot_loss[loss=0.1636, simple_loss=0.2429, pruned_loss=0.04212, over 3750648.70 frames. ], batch size: 94, lr: 4.75e-03, grad_scale: 32.0
+2023-03-29 04:56:47,929 INFO [train.py:892] (3/4) Epoch 33, batch 650, loss[loss=0.1467, simple_loss=0.2295, pruned_loss=0.03197, over 19737.00 frames. ], tot_loss[loss=0.1628, simple_loss=0.2422, pruned_loss=0.04166, over 3792741.62 frames. ], batch size: 77, lr: 4.75e-03, grad_scale: 32.0
+2023-03-29 04:56:49,270 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0
+2023-03-29 04:57:13,237 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9343, 4.0044, 2.4536, 4.2232, 4.4019, 1.9773, 3.6282, 3.3462],
+       device='cuda:3'), covar=tensor([0.0724, 0.0898, 0.2803, 0.0880, 0.0603, 0.2907, 0.1124, 0.0932],
+       device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0261, 0.0234, 0.0279, 0.0258, 0.0206, 0.0243, 0.0200],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 04:57:20,325 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.343e+02 3.751e+02 4.539e+02 5.312e+02 8.817e+02, threshold=9.077e+02, percent-clipped=1.0
+2023-03-29 04:57:55,025 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:58:22,742 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:58:38,896 INFO [train.py:892] (3/4) Epoch 33, batch 700, loss[loss=0.1795, simple_loss=0.2642, pruned_loss=0.04741, over 19763.00 frames. ], tot_loss[loss=0.1625, simple_loss=0.2421, pruned_loss=0.04151, over 3826062.01 frames. ], batch size: 226, lr: 4.75e-03, grad_scale: 32.0
+2023-03-29 04:58:48,359 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60065.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:58:58,763 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1472, 3.4259, 3.5099, 4.0900, 2.8730, 3.1782, 2.8535, 2.6552],
+       device='cuda:3'), covar=tensor([0.0547, 0.1945, 0.1091, 0.0460, 0.2014, 0.0992, 0.1265, 0.1667],
+       device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0327, 0.0249, 0.0204, 0.0247, 0.0209, 0.0217, 0.0217],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 04:59:18,788 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 04:59:32,160 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-29 05:00:08,763 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60101.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 05:00:13,276 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3696, 2.5405, 4.5095, 3.8359, 4.1739, 4.4070, 4.2205, 4.0906],
+       device='cuda:3'), covar=tensor([0.0521, 0.1024, 0.0107, 0.0784, 0.0175, 0.0231, 0.0182, 0.0172],
+       device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0105, 0.0089, 0.0153, 0.0087, 0.0099, 0.0091, 0.0087],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 05:00:28,991 INFO [train.py:892] (3/4) Epoch 33, batch 750, loss[loss=0.1514, simple_loss=0.2183, pruned_loss=0.04221, over 19764.00 frames. ], tot_loss[loss=0.1621, simple_loss=0.2414, pruned_loss=0.04143, over 3853099.41 frames. ], batch size: 152, lr: 4.75e-03, grad_scale: 32.0
+2023-03-29 05:00:33,771 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60113.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 05:00:59,947 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.577e+02 3.667e+02 4.352e+02 5.271e+02 8.551e+02, threshold=8.703e+02, percent-clipped=0.0
+2023-03-29 05:01:03,763 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:01:23,416 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4765, 3.8431, 3.9334, 4.5515, 3.0476, 3.4781, 2.8373, 2.7893],
+       device='cuda:3'), covar=tensor([0.0463, 0.1734, 0.0817, 0.0349, 0.1949, 0.0928, 0.1340, 0.1603],
+       device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0329, 0.0249, 0.0204, 0.0248, 0.0209, 0.0219, 0.0218],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 05:01:34,885 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:01:53,533 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0
+2023-03-29 05:02:17,800 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:02:23,116 INFO [train.py:892] (3/4) Epoch 33, batch 800, loss[loss=0.174, simple_loss=0.2524, pruned_loss=0.04784, over 19731.00 frames. ], tot_loss[loss=0.163, simple_loss=0.2423, pruned_loss=0.04182, over 3872962.66 frames. ], batch size: 269, lr: 4.74e-03, grad_scale: 32.0
+2023-03-29 05:02:31,897 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2251, 4.0630, 4.5027, 4.1258, 3.8103, 4.3671, 4.1527, 4.6027],
+       device='cuda:3'), covar=tensor([0.0793, 0.0381, 0.0358, 0.0414, 0.1036, 0.0523, 0.0470, 0.0338],
+       device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0222, 0.0222, 0.0235, 0.0208, 0.0245, 0.0232, 0.0218],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 05:03:47,818 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:04:03,564 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60204.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:04:08,047 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60206.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:04:17,806 INFO [train.py:892] (3/4) Epoch 33, batch 850, loss[loss=0.1418, simple_loss=0.2158, pruned_loss=0.03391, over 19862.00 frames. ], tot_loss[loss=0.162, simple_loss=0.2415, pruned_loss=0.04129, over 3889562.54 frames. ], batch size: 99, lr: 4.74e-03, grad_scale: 32.0
+2023-03-29 05:04:48,774 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.588e+02 3.634e+02 4.278e+02 5.134e+02 8.109e+02, threshold=8.556e+02, percent-clipped=0.0
+2023-03-29 05:05:15,499 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.5026, 5.8105, 5.8754, 5.7687, 5.5346, 5.8643, 5.2504, 5.3025],
+       device='cuda:3'), covar=tensor([0.0424, 0.0482, 0.0462, 0.0406, 0.0570, 0.0450, 0.0667, 0.1022],
+       device='cuda:3'), in_proj_covar=tensor([0.0276, 0.0293, 0.0306, 0.0266, 0.0271, 0.0256, 0.0273, 0.0321],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 05:05:37,089 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:06:04,343 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0632, 5.2624, 5.4717, 5.2797, 5.2775, 5.1066, 5.2269, 4.9937],
+       device='cuda:3'), covar=tensor([0.1385, 0.1437, 0.0945, 0.1214, 0.0779, 0.0835, 0.1822, 0.1949],
+       device='cuda:3'), in_proj_covar=tensor([0.0294, 0.0337, 0.0371, 0.0301, 0.0277, 0.0285, 0.0366, 0.0391],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 05:06:08,124 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60259.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:06:11,088 INFO [train.py:892] (3/4) Epoch 33, batch 900, loss[loss=0.1567, simple_loss=0.233, pruned_loss=0.04022, over 19795.00 frames. ], tot_loss[loss=0.1606, simple_loss=0.2397, pruned_loss=0.04071, over 3904036.86 frames. ], batch size: 185, lr: 4.74e-03, grad_scale: 32.0
+2023-03-29 05:08:06,296 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0
+2023-03-29 05:08:06,815 INFO [train.py:892] (3/4) Epoch 33, batch 950, loss[loss=0.2053, simple_loss=0.2818, pruned_loss=0.0644, over 19713.00 frames. ], tot_loss[loss=0.1606, simple_loss=0.2402, pruned_loss=0.04046, over 3913132.00 frames. ], batch size: 305, lr: 4.74e-03, grad_scale: 32.0
+2023-03-29 05:08:37,043 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.481e+02 3.823e+02 4.629e+02 5.710e+02 9.734e+02, threshold=9.258e+02, percent-clipped=1.0
+2023-03-29 05:09:56,798 INFO [train.py:892] (3/4) Epoch 33, batch 1000, loss[loss=0.2619, simple_loss=0.3338, pruned_loss=0.09498, over 19415.00 frames. ], tot_loss[loss=0.1612, simple_loss=0.2409, pruned_loss=0.04074, over 3920131.49 frames. ], batch size: 431, lr: 4.74e-03, grad_scale: 32.0
+2023-03-29 05:11:13,844 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60396.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 05:11:41,770 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60408.0, num_to_drop=1, layers_to_drop={3}
+2023-03-29 05:11:48,663 INFO [train.py:892] (3/4) Epoch 33, batch 1050, loss[loss=0.1755, simple_loss=0.2578, pruned_loss=0.04657, over 19701.00 frames. ], tot_loss[loss=0.1615, simple_loss=0.2411, pruned_loss=0.04093, over 3926541.39 frames. ], batch size: 305, lr: 4.74e-03, grad_scale: 32.0
+2023-03-29 05:12:11,168 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60421.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:12:21,548 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.510e+02 3.870e+02 4.484e+02 5.211e+02 8.559e+02, threshold=8.968e+02, percent-clipped=0.0
+2023-03-29 05:12:39,939 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0
+2023-03-29 05:12:41,340 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60434.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:13:27,701 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5900, 3.6692, 2.4349, 4.3222, 3.9891, 4.3482, 4.3909, 3.5354],
+       device='cuda:3'), covar=tensor([0.0581, 0.0537, 0.1439, 0.0597, 0.0547, 0.0383, 0.0548, 0.0689],
+       device='cuda:3'), in_proj_covar=tensor([0.0142, 0.0142, 0.0142, 0.0151, 0.0133, 0.0136, 0.0146, 0.0144],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 05:13:40,390 INFO [train.py:892] (3/4) Epoch 33, batch 1100, loss[loss=0.1517, simple_loss=0.2316, pruned_loss=0.03593, over 19875.00 frames. ], tot_loss[loss=0.1611, simple_loss=0.2406, pruned_loss=0.04083, over 3933078.05 frames. ], batch size: 64, lr: 4.73e-03, grad_scale: 32.0
+2023-03-29 05:13:45,524 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0718, 2.9678, 3.1439, 2.8273, 3.3807, 3.2938, 3.9305, 4.3115],
+       device='cuda:3'), covar=tensor([0.0644, 0.1637, 0.1559, 0.2232, 0.1582, 0.1651, 0.0624, 0.0539],
+       device='cuda:3'), in_proj_covar=tensor([0.0256, 0.0241, 0.0269, 0.0254, 0.0298, 0.0258, 0.0234, 0.0258],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 05:14:38,169 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5305, 1.9456, 2.3157, 2.7695, 3.1163, 3.1358, 3.1076, 3.1922],
+       device='cuda:3'), covar=tensor([0.1008, 0.1756, 0.1415, 0.0727, 0.0490, 0.0398, 0.0418, 0.0441],
+       device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0168, 0.0176, 0.0150, 0.0135, 0.0133, 0.0124, 0.0116],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 05:15:21,756 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:15:35,180 INFO [train.py:892] (3/4) Epoch 33, batch 1150, loss[loss=0.161, simple_loss=0.2382, pruned_loss=0.04187, over 19958.00 frames. ], tot_loss[loss=0.1606, simple_loss=0.2398, pruned_loss=0.04073, over 3938048.46 frames. ], batch size: 53, lr: 4.73e-03, grad_scale: 32.0
+2023-03-29 05:16:08,395 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.786e+02 3.806e+02 4.491e+02 5.149e+02 7.976e+02, threshold=8.981e+02, percent-clipped=0.0
+2023-03-29 05:16:54,339 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:17:08,876 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:17:13,042 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60554.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:17:29,245 INFO [train.py:892] (3/4) Epoch 33, batch 1200, loss[loss=0.1549, simple_loss=0.241, pruned_loss=0.03434, over 19781.00 frames. ], tot_loss[loss=0.1593, simple_loss=0.2384, pruned_loss=0.04008, over 3941064.29 frames. ], batch size: 87, lr: 4.73e-03, grad_scale: 32.0
+2023-03-29 05:18:42,753 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60593.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:19:02,557 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8236, 4.9293, 5.1776, 5.0300, 5.0402, 4.6993, 4.9228, 4.7489],
+       device='cuda:3'), covar=tensor([0.1421, 0.1531, 0.0942, 0.1212, 0.0787, 0.0892, 0.1865, 0.1920],
+       device='cuda:3'), in_proj_covar=tensor([0.0293, 0.0334, 0.0367, 0.0299, 0.0275, 0.0283, 0.0361, 0.0388],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 05:19:21,771 INFO [train.py:892] (3/4) Epoch 33, batch 1250, loss[loss=0.1584, simple_loss=0.2365, pruned_loss=0.04015, over 19735.00 frames. ], tot_loss[loss=0.1599, simple_loss=0.239, pruned_loss=0.04039, over 3944015.63 frames. ], batch size: 118, lr: 4.73e-03, grad_scale: 32.0
+2023-03-29 05:19:44,923 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.11 vs. limit=2.0
+2023-03-29 05:19:52,007 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.888e+02 3.769e+02 4.430e+02 5.364e+02 1.022e+03, threshold=8.861e+02, percent-clipped=3.0
+2023-03-29 05:21:14,320 INFO [train.py:892] (3/4) Epoch 33, batch 1300, loss[loss=0.1415, simple_loss=0.2273, pruned_loss=0.02792, over 19786.00 frames. ], tot_loss[loss=0.1599, simple_loss=0.2391, pruned_loss=0.04033, over 3946880.04 frames. ], batch size: 111, lr: 4.73e-03, grad_scale: 32.0
+2023-03-29 05:21:15,903 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0
+2023-03-29 05:22:35,177 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:23:03,638 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:23:08,910 INFO [train.py:892] (3/4) Epoch 33, batch 1350, loss[loss=0.1598, simple_loss=0.231, pruned_loss=0.04428, over 19830.00 frames. ], tot_loss[loss=0.1599, simple_loss=0.2395, pruned_loss=0.04013, over 3945225.28 frames. ], batch size: 166, lr: 4.72e-03, grad_scale: 16.0
+2023-03-29 05:23:27,924 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:23:34,154 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60721.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:23:44,306 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.407e+02 3.398e+02 4.212e+02 5.272e+02 1.001e+03, threshold=8.423e+02, percent-clipped=0.0
+2023-03-29 05:24:02,012 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:24:25,149 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60744.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:24:53,340 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:25:04,413 INFO [train.py:892] (3/4) Epoch 33, batch 1400, loss[loss=0.1622, simple_loss=0.2436, pruned_loss=0.04038, over 19851.00 frames. ], tot_loss[loss=0.1605, simple_loss=0.2398, pruned_loss=0.04057, over 3946236.68 frames. ], batch size: 60, lr: 4.72e-03, grad_scale: 16.0
+2023-03-29 05:25:16,536 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0
+2023-03-29 05:25:21,961 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60769.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:25:40,836 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4411, 3.5705, 2.2296, 3.6686, 3.7746, 1.7542, 3.1154, 2.9059],
+       device='cuda:3'), covar=tensor([0.0788, 0.0861, 0.2665, 0.0735, 0.0609, 0.2593, 0.1121, 0.0943],
+       device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0261, 0.0233, 0.0280, 0.0259, 0.0205, 0.0243, 0.0200],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 05:25:44,817 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60779.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:25:51,140 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60782.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:26:54,145 INFO [train.py:892] (3/4) Epoch 33, batch 1450, loss[loss=0.175, simple_loss=0.2548, pruned_loss=0.04763, over 19642.00 frames. ], tot_loss[loss=0.1611, simple_loss=0.2407, pruned_loss=0.04081, over 3945597.81 frames. ], batch size: 330, lr: 4.72e-03, grad_scale: 16.0
+2023-03-29 05:27:25,507 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.764e+02 3.944e+02 4.679e+02 5.484e+02 1.088e+03, threshold=9.358e+02, percent-clipped=4.0
+2023-03-29 05:28:30,977 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60854.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:28:46,124 INFO [train.py:892] (3/4) Epoch 33, batch 1500, loss[loss=0.1597, simple_loss=0.2278, pruned_loss=0.04584, over 19798.00 frames. ], tot_loss[loss=0.1611, simple_loss=0.2404, pruned_loss=0.04086, over 3945493.90 frames. ], batch size: 114, lr: 4.72e-03, grad_scale: 16.0
+2023-03-29 05:30:20,982 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:30:40,170 INFO [train.py:892] (3/4) Epoch 33, batch 1550, loss[loss=0.1603, simple_loss=0.2287, pruned_loss=0.04594, over 19875.00 frames. ], tot_loss[loss=0.1612, simple_loss=0.2403, pruned_loss=0.04098, over 3946965.70 frames. ], batch size: 125, lr: 4.72e-03, grad_scale: 16.0
+2023-03-29 05:31:04,546 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0
+2023-03-29 05:31:12,868 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.819e+02 3.799e+02 4.581e+02 5.362e+02 9.999e+02, threshold=9.162e+02, percent-clipped=1.0
+2023-03-29 05:32:34,458 INFO [train.py:892] (3/4) Epoch 33, batch 1600, loss[loss=0.134, simple_loss=0.2043, pruned_loss=0.03188, over 19828.00 frames. ], tot_loss[loss=0.1609, simple_loss=0.24, pruned_loss=0.04088, over 3947909.08 frames. ], batch size: 147, lr: 4.71e-03, grad_scale: 16.0
+2023-03-29 05:33:50,754 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1124, 3.2193, 2.0138, 3.2516, 3.3301, 1.6783, 2.8349, 2.5624],
+       device='cuda:3'), covar=tensor([0.0864, 0.0855, 0.2618, 0.0844, 0.0664, 0.2403, 0.1045, 0.0994],
+       device='cuda:3'), in_proj_covar=tensor([0.0236, 0.0261, 0.0233, 0.0280, 0.0259, 0.0206, 0.0242, 0.0200],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 05:34:25,220 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6629, 4.5286, 5.0248, 4.5789, 4.1584, 4.7965, 4.6404, 5.1587],
+       device='cuda:3'), covar=tensor([0.0792, 0.0364, 0.0335, 0.0401, 0.0832, 0.0494, 0.0435, 0.0287],
+       device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0223, 0.0223, 0.0238, 0.0209, 0.0247, 0.0236, 0.0220],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 05:34:30,689 INFO [train.py:892] (3/4) Epoch 33, batch 1650, loss[loss=0.1579, simple_loss=0.2353, pruned_loss=0.04029, over 19757.00 frames. ], tot_loss[loss=0.1603, simple_loss=0.2394, pruned_loss=0.04063, over 3948496.73 frames. ], batch size: 88, lr: 4.71e-03, grad_scale: 16.0
+2023-03-29 05:35:04,280 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.576e+02 3.785e+02 4.446e+02 5.538e+02 8.034e+02, threshold=8.891e+02, percent-clipped=0.0
+2023-03-29 05:35:18,733 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0
+2023-03-29 05:35:27,900 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0796, 3.3266, 2.8704, 2.5757, 3.0318, 3.3739, 3.1900, 3.2341],
+       device='cuda:3'), covar=tensor([0.0289, 0.0326, 0.0298, 0.0452, 0.0342, 0.0218, 0.0265, 0.0203],
+       device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0098, 0.0101, 0.0103, 0.0106, 0.0089, 0.0090, 0.0089],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 05:35:45,771 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0
+2023-03-29 05:36:00,542 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0461, 4.6937, 4.7354, 5.0223, 4.6936, 5.1985, 5.1540, 5.3220],
+       device='cuda:3'), covar=tensor([0.0596, 0.0414, 0.0462, 0.0345, 0.0666, 0.0412, 0.0416, 0.0292],
+       device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0180, 0.0205, 0.0179, 0.0177, 0.0162, 0.0154, 0.0200],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 05:36:19,921 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2906, 3.0017, 3.2990, 2.9372, 3.5297, 3.4445, 4.0900, 4.5278],
+       device='cuda:3'), covar=tensor([0.0574, 0.1701, 0.1537, 0.2240, 0.1686, 0.1513, 0.0677, 0.0576],
+       device='cuda:3'), in_proj_covar=tensor([0.0260, 0.0245, 0.0274, 0.0258, 0.0303, 0.0262, 0.0238, 0.0263],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 05:36:28,730 INFO [train.py:892] (3/4) Epoch 33, batch 1700, loss[loss=0.1563, simple_loss=0.2269, pruned_loss=0.04279, over 19896.00 frames. ], tot_loss[loss=0.1596, simple_loss=0.2389, pruned_loss=0.04014, over 3948932.95 frames. ], batch size: 62, lr: 4.71e-03, grad_scale: 16.0
+2023-03-29 05:36:29,967 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61061.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:37:08,455 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61074.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:38:23,926 INFO [train.py:892] (3/4) Epoch 33, batch 1750, loss[loss=0.1571, simple_loss=0.245, pruned_loss=0.03463, over 19917.00 frames. ], tot_loss[loss=0.1592, simple_loss=0.2384, pruned_loss=0.03996, over 3949443.48 frames. ], batch size: 45, lr: 4.71e-03, grad_scale: 16.0
+2023-03-29 05:38:45,396 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:38:52,716 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.343e+02 3.701e+02 4.217e+02 4.986e+02 8.389e+02, threshold=8.433e+02, percent-clipped=0.0
+2023-03-29 05:39:59,798 INFO [train.py:892] (3/4) Epoch 33, batch 1800, loss[loss=0.1783, simple_loss=0.242, pruned_loss=0.05727, over 19841.00 frames. ], tot_loss[loss=0.1593, simple_loss=0.2384, pruned_loss=0.04013, over 3950740.59 frames. ], batch size: 124, lr: 4.71e-03, grad_scale: 16.0
+2023-03-29 05:40:07,430 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.2626, 2.3275, 2.3641, 2.3816, 2.4128, 2.3718, 2.4224, 2.3773],
+       device='cuda:3'), covar=tensor([0.0418, 0.0352, 0.0352, 0.0343, 0.0434, 0.0410, 0.0435, 0.0359],
+       device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0084, 0.0087, 0.0080, 0.0093, 0.0086, 0.0103, 0.0075],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 05:41:31,942 INFO [train.py:892] (3/4) Epoch 33, batch 1850, loss[loss=0.1479, simple_loss=0.245, pruned_loss=0.02536, over 19844.00 frames. ], tot_loss[loss=0.1606, simple_loss=0.241, pruned_loss=0.04014, over 3947159.64 frames. ], batch size: 58, lr: 4.70e-03, grad_scale: 16.0
+2023-03-29 05:42:34,544 INFO [train.py:892] (3/4) Epoch 34, batch 0, loss[loss=0.1495, simple_loss=0.2297, pruned_loss=0.03469, over 19661.00 frames. ], tot_loss[loss=0.1495, simple_loss=0.2297, pruned_loss=0.03469, over 19661.00 frames. ], batch size: 51, lr: 4.63e-03, grad_scale: 16.0
+2023-03-29 05:42:34,545 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-29 05:42:56,669 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5798, 4.0458, 3.8831, 3.8800, 4.0894, 3.9295, 3.8618, 3.6635],
+       device='cuda:3'), covar=tensor([0.2137, 0.1354, 0.1625, 0.1437, 0.0852, 0.0975, 0.1983, 0.2252],
+       device='cuda:3'), in_proj_covar=tensor([0.0299, 0.0340, 0.0372, 0.0303, 0.0279, 0.0289, 0.0368, 0.0397],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 05:43:03,821 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7871, 3.0407, 3.3518, 3.6092, 2.7965, 3.1211, 2.5221, 2.6204],
+       device='cuda:3'), covar=tensor([0.0527, 0.1610, 0.0884, 0.0503, 0.1847, 0.0762, 0.1353, 0.1506],
+       device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0329, 0.0251, 0.0205, 0.0249, 0.0211, 0.0220, 0.0218],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 05:43:07,404 INFO [train.py:926] (3/4) Epoch 34, validation: loss=0.1816, simple_loss=0.2491, pruned_loss=0.05706, over 2883724.00 frames. 
+2023-03-29 05:43:07,405 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-29 05:43:30,437 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.548e+02 3.493e+02 4.214e+02 5.020e+02 1.069e+03, threshold=8.428e+02, percent-clipped=3.0
+2023-03-29 05:44:23,243 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8996, 4.6123, 4.7094, 4.9657, 4.6799, 5.1744, 5.0866, 5.2972],
+       device='cuda:3'), covar=tensor([0.0757, 0.0395, 0.0473, 0.0377, 0.0678, 0.0392, 0.0470, 0.0350],
+       device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0180, 0.0204, 0.0178, 0.0177, 0.0161, 0.0153, 0.0199],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 05:45:06,389 INFO [train.py:892] (3/4) Epoch 34, batch 50, loss[loss=0.1703, simple_loss=0.2557, pruned_loss=0.04248, over 19847.00 frames. ], tot_loss[loss=0.1602, simple_loss=0.2401, pruned_loss=0.04012, over 888390.52 frames. ], batch size: 49, lr: 4.63e-03, grad_scale: 16.0
+2023-03-29 05:47:01,443 INFO [train.py:892] (3/4) Epoch 34, batch 100, loss[loss=0.1396, simple_loss=0.223, pruned_loss=0.02806, over 19774.00 frames. ], tot_loss[loss=0.1597, simple_loss=0.2383, pruned_loss=0.04052, over 1567321.07 frames. ], batch size: 155, lr: 4.63e-03, grad_scale: 16.0
+2023-03-29 05:47:24,787 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.578e+02 3.716e+02 4.447e+02 5.513e+02 1.175e+03, threshold=8.893e+02, percent-clipped=3.0
+2023-03-29 05:48:17,664 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5909, 3.8345, 2.2543, 4.4150, 4.0123, 4.3519, 4.3831, 3.4449],
+       device='cuda:3'), covar=tensor([0.0591, 0.0572, 0.1609, 0.0574, 0.0628, 0.0416, 0.0632, 0.0788],
+       device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0146, 0.0145, 0.0155, 0.0136, 0.0138, 0.0151, 0.0148],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 05:48:57,037 INFO [train.py:892] (3/4) Epoch 34, batch 150, loss[loss=0.1946, simple_loss=0.2669, pruned_loss=0.06115, over 19856.00 frames. ], tot_loss[loss=0.1591, simple_loss=0.238, pruned_loss=0.04012, over 2094679.35 frames. ], batch size: 81, lr: 4.63e-03, grad_scale: 16.0
+2023-03-29 05:49:11,894 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.0374, 2.3505, 2.1222, 1.5827, 2.1639, 2.3631, 2.2112, 2.2485],
+       device='cuda:3'), covar=tensor([0.0423, 0.0309, 0.0353, 0.0594, 0.0414, 0.0292, 0.0302, 0.0298],
+       device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0100, 0.0102, 0.0104, 0.0108, 0.0090, 0.0091, 0.0090],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 05:49:18,285 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:49:49,242 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4820, 4.3500, 4.7923, 4.3833, 4.0628, 4.6302, 4.4149, 4.8770],
+       device='cuda:3'), covar=tensor([0.0737, 0.0369, 0.0348, 0.0380, 0.0907, 0.0495, 0.0487, 0.0322],
+       device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0224, 0.0224, 0.0237, 0.0210, 0.0248, 0.0237, 0.0221],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 05:50:56,788 INFO [train.py:892] (3/4) Epoch 34, batch 200, loss[loss=0.1411, simple_loss=0.2161, pruned_loss=0.03306, over 19831.00 frames. ], tot_loss[loss=0.1594, simple_loss=0.2384, pruned_loss=0.04016, over 2506196.73 frames. ], batch size: 177, lr: 4.63e-03, grad_scale: 16.0
+2023-03-29 05:50:59,760 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61417.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:51:10,352 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=61422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 05:51:18,480 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.390e+02 3.871e+02 4.515e+02 5.267e+02 1.071e+03, threshold=9.030e+02, percent-clipped=3.0
+2023-03-29 05:52:22,307 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0
+2023-03-29 05:52:53,175 INFO [train.py:892] (3/4) Epoch 34, batch 250, loss[loss=0.1558, simple_loss=0.2322, pruned_loss=0.03971, over 19805.00 frames. ], tot_loss[loss=0.1602, simple_loss=0.2386, pruned_loss=0.04093, over 2826782.50 frames. ], batch size: 195, lr: 4.62e-03, grad_scale: 16.0
+2023-03-29 05:54:47,105 INFO [train.py:892] (3/4) Epoch 34, batch 300, loss[loss=0.1539, simple_loss=0.2326, pruned_loss=0.03762, over 19807.00 frames. ], tot_loss[loss=0.1595, simple_loss=0.2384, pruned_loss=0.04027, over 3074001.65 frames. ], batch size: 167, lr: 4.62e-03, grad_scale: 16.0
+2023-03-29 05:55:09,939 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.400e+02 3.531e+02 4.342e+02 5.324e+02 1.066e+03, threshold=8.684e+02, percent-clipped=3.0
+2023-03-29 05:56:44,443 INFO [train.py:892] (3/4) Epoch 34, batch 350, loss[loss=0.1523, simple_loss=0.2252, pruned_loss=0.03971, over 19846.00 frames. ], tot_loss[loss=0.1591, simple_loss=0.2379, pruned_loss=0.04011, over 3268496.63 frames. ], batch size: 137, lr: 4.62e-03, grad_scale: 16.0
+2023-03-29 05:58:37,660 INFO [train.py:892] (3/4) Epoch 34, batch 400, loss[loss=0.1619, simple_loss=0.2461, pruned_loss=0.03884, over 19765.00 frames. ], tot_loss[loss=0.1613, simple_loss=0.2405, pruned_loss=0.0411, over 3419202.91 frames. ], batch size: 70, lr: 4.62e-03, grad_scale: 16.0
+2023-03-29 05:59:04,290 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.292e+02 3.497e+02 4.355e+02 5.264e+02 1.050e+03, threshold=8.709e+02, percent-clipped=2.0
+2023-03-29 05:59:27,100 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0
+2023-03-29 06:00:00,272 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9834, 2.8527, 4.9659, 4.0893, 4.6932, 4.9014, 4.7377, 4.5926],
+       device='cuda:3'), covar=tensor([0.0460, 0.0967, 0.0100, 0.0965, 0.0137, 0.0165, 0.0151, 0.0143],
+       device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0103, 0.0088, 0.0152, 0.0086, 0.0098, 0.0090, 0.0086],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 06:00:25,470 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.73 vs. limit=5.0
+2023-03-29 06:00:32,550 INFO [train.py:892] (3/4) Epoch 34, batch 450, loss[loss=0.1533, simple_loss=0.2317, pruned_loss=0.03744, over 19729.00 frames. ], tot_loss[loss=0.1606, simple_loss=0.2398, pruned_loss=0.04072, over 3537821.91 frames. ], batch size: 47, lr: 4.62e-03, grad_scale: 16.0
+2023-03-29 06:01:42,254 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9747, 2.9983, 1.8859, 3.5175, 3.2887, 3.4159, 3.5922, 2.8306],
+       device='cuda:3'), covar=tensor([0.0640, 0.0722, 0.1610, 0.0593, 0.0602, 0.0565, 0.0515, 0.0826],
+       device='cuda:3'), in_proj_covar=tensor([0.0143, 0.0145, 0.0144, 0.0154, 0.0135, 0.0137, 0.0149, 0.0147],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 06:02:28,800 INFO [train.py:892] (3/4) Epoch 34, batch 500, loss[loss=0.17, simple_loss=0.2401, pruned_loss=0.04992, over 19844.00 frames. ], tot_loss[loss=0.161, simple_loss=0.2401, pruned_loss=0.0409, over 3629878.82 frames. ], batch size: 137, lr: 4.61e-03, grad_scale: 16.0
+2023-03-29 06:02:32,333 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61717.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:02:51,067 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.296e+02 3.666e+02 4.435e+02 5.198e+02 9.761e+02, threshold=8.870e+02, percent-clipped=3.0
+2023-03-29 06:03:07,981 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:03:32,998 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61743.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:04:20,985 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=61765.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:04:22,376 INFO [train.py:892] (3/4) Epoch 34, batch 550, loss[loss=0.1358, simple_loss=0.212, pruned_loss=0.02979, over 19663.00 frames. ], tot_loss[loss=0.1595, simple_loss=0.2386, pruned_loss=0.04024, over 3701802.44 frames. ], batch size: 43, lr: 4.61e-03, grad_scale: 16.0
+2023-03-29 06:04:23,458 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8811, 4.5989, 4.6403, 4.8885, 4.6101, 5.0314, 5.0565, 5.2003],
+       device='cuda:3'), covar=tensor([0.0656, 0.0431, 0.0515, 0.0377, 0.0665, 0.0476, 0.0396, 0.0304],
+       device='cuda:3'), in_proj_covar=tensor([0.0154, 0.0180, 0.0203, 0.0178, 0.0177, 0.0161, 0.0154, 0.0200],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 06:05:28,872 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:05:53,682 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.46 vs. limit=5.0
+2023-03-29 06:05:55,578 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61804.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:06:24,086 INFO [train.py:892] (3/4) Epoch 34, batch 600, loss[loss=0.1552, simple_loss=0.2329, pruned_loss=0.03874, over 19604.00 frames. ], tot_loss[loss=0.1594, simple_loss=0.2386, pruned_loss=0.0401, over 3757071.41 frames. ], batch size: 50, lr: 4.61e-03, grad_scale: 16.0
+2023-03-29 06:06:47,557 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.635e+02 3.514e+02 4.079e+02 4.976e+02 9.879e+02, threshold=8.158e+02, percent-clipped=2.0
+2023-03-29 06:07:00,701 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6192, 2.6785, 2.7039, 2.2915, 2.8027, 2.4229, 2.7860, 2.7354],
+       device='cuda:3'), covar=tensor([0.0569, 0.0502, 0.0555, 0.0808, 0.0448, 0.0491, 0.0487, 0.0393],
+       device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0089, 0.0086, 0.0112, 0.0082, 0.0084, 0.0083, 0.0076],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 06:07:44,935 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3889, 3.2266, 3.5006, 2.7373, 3.5469, 3.0438, 3.2965, 3.4018],
+       device='cuda:3'), covar=tensor([0.0552, 0.0465, 0.0450, 0.0773, 0.0435, 0.0476, 0.0524, 0.0388],
+       device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0089, 0.0086, 0.0113, 0.0082, 0.0084, 0.0083, 0.0076],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 06:08:22,993 INFO [train.py:892] (3/4) Epoch 34, batch 650, loss[loss=0.153, simple_loss=0.22, pruned_loss=0.04299, over 19849.00 frames. ], tot_loss[loss=0.1595, simple_loss=0.2382, pruned_loss=0.0404, over 3800179.63 frames. ], batch size: 144, lr: 4.61e-03, grad_scale: 16.0
+2023-03-29 06:10:10,005 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:10:14,515 INFO [train.py:892] (3/4) Epoch 34, batch 700, loss[loss=0.1467, simple_loss=0.2407, pruned_loss=0.02634, over 19782.00 frames. ], tot_loss[loss=0.1608, simple_loss=0.24, pruned_loss=0.04077, over 3833381.50 frames. ], batch size: 48, lr: 4.61e-03, grad_scale: 16.0
+2023-03-29 06:10:38,150 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.620e+02 3.763e+02 4.447e+02 5.196e+02 1.125e+03, threshold=8.894e+02, percent-clipped=5.0
+2023-03-29 06:12:06,000 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.84 vs. limit=5.0
+2023-03-29 06:12:12,068 INFO [train.py:892] (3/4) Epoch 34, batch 750, loss[loss=0.1412, simple_loss=0.2255, pruned_loss=0.02844, over 19799.00 frames. ], tot_loss[loss=0.16, simple_loss=0.2395, pruned_loss=0.04028, over 3859361.12 frames. ], batch size: 83, lr: 4.61e-03, grad_scale: 16.0
+2023-03-29 06:12:34,579 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61975.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:14:10,214 INFO [train.py:892] (3/4) Epoch 34, batch 800, loss[loss=0.1344, simple_loss=0.2081, pruned_loss=0.03039, over 19743.00 frames. ], tot_loss[loss=0.16, simple_loss=0.2395, pruned_loss=0.04028, over 3878547.72 frames. ], batch size: 140, lr: 4.60e-03, grad_scale: 16.0
+2023-03-29 06:14:31,339 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.697e+02 3.764e+02 4.280e+02 5.112e+02 1.282e+03, threshold=8.560e+02, percent-clipped=3.0
+2023-03-29 06:15:17,292 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2020, 3.9540, 4.0406, 4.2172, 3.9731, 4.3894, 4.2136, 4.4567],
+       device='cuda:3'), covar=tensor([0.0775, 0.0583, 0.0641, 0.0477, 0.0828, 0.0602, 0.0734, 0.0508],
+       device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0181, 0.0206, 0.0179, 0.0179, 0.0163, 0.0155, 0.0202],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 06:16:03,780 INFO [train.py:892] (3/4) Epoch 34, batch 850, loss[loss=0.1313, simple_loss=0.2141, pruned_loss=0.02427, over 19743.00 frames. ], tot_loss[loss=0.159, simple_loss=0.2383, pruned_loss=0.03987, over 3896126.59 frames. ], batch size: 92, lr: 4.60e-03, grad_scale: 16.0
+2023-03-29 06:16:16,169 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4180, 3.7430, 3.8295, 4.4522, 3.0102, 3.3382, 2.7904, 2.7464],
+       device='cuda:3'), covar=tensor([0.0456, 0.2118, 0.0937, 0.0420, 0.2108, 0.1050, 0.1340, 0.1693],
+       device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0330, 0.0252, 0.0207, 0.0250, 0.0211, 0.0222, 0.0220],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 06:16:53,338 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:17:17,394 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62099.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:17:54,507 INFO [train.py:892] (3/4) Epoch 34, batch 900, loss[loss=0.189, simple_loss=0.2641, pruned_loss=0.05697, over 19696.00 frames. ], tot_loss[loss=0.1586, simple_loss=0.2375, pruned_loss=0.03988, over 3908685.14 frames. ], batch size: 315, lr: 4.60e-03, grad_scale: 16.0
+2023-03-29 06:18:03,653 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0
+2023-03-29 06:18:09,185 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62122.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:18:16,039 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.740e+02 3.509e+02 4.216e+02 5.041e+02 1.543e+03, threshold=8.432e+02, percent-clipped=2.0
+2023-03-29 06:19:23,804 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.1082, 1.5681, 1.6445, 2.2949, 2.4702, 2.6389, 2.4216, 2.5683],
+       device='cuda:3'), covar=tensor([0.1138, 0.1979, 0.1858, 0.0842, 0.0609, 0.0428, 0.0562, 0.0525],
+       device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0173, 0.0182, 0.0154, 0.0139, 0.0136, 0.0127, 0.0120],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 06:19:48,624 INFO [train.py:892] (3/4) Epoch 34, batch 950, loss[loss=0.1483, simple_loss=0.2351, pruned_loss=0.03078, over 19776.00 frames. ], tot_loss[loss=0.1594, simple_loss=0.2388, pruned_loss=0.04006, over 3916898.38 frames. ], batch size: 70, lr: 4.60e-03, grad_scale: 16.0
+2023-03-29 06:20:28,517 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62183.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:21:32,062 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4309, 2.1744, 3.5041, 3.0077, 3.5187, 3.5770, 3.2586, 3.3861],
+       device='cuda:3'), covar=tensor([0.0839, 0.1102, 0.0144, 0.0476, 0.0163, 0.0245, 0.0242, 0.0217],
+       device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0103, 0.0089, 0.0153, 0.0087, 0.0098, 0.0090, 0.0086],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 06:21:45,079 INFO [train.py:892] (3/4) Epoch 34, batch 1000, loss[loss=0.228, simple_loss=0.3137, pruned_loss=0.07117, over 19257.00 frames. ], tot_loss[loss=0.1614, simple_loss=0.2407, pruned_loss=0.04104, over 3922162.18 frames. ], batch size: 483, lr: 4.60e-03, grad_scale: 16.0
+2023-03-29 06:21:45,992 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.1008, 5.2235, 5.4748, 5.2722, 5.3151, 4.9985, 5.2262, 5.0400],
+       device='cuda:3'), covar=tensor([0.1394, 0.1746, 0.0910, 0.1130, 0.0698, 0.0797, 0.1910, 0.1877],
+       device='cuda:3'), in_proj_covar=tensor([0.0297, 0.0337, 0.0372, 0.0302, 0.0278, 0.0286, 0.0365, 0.0392],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 06:22:08,033 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.587e+02 3.779e+02 4.502e+02 5.626e+02 1.320e+03, threshold=9.004e+02, percent-clipped=5.0
+2023-03-29 06:22:53,905 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6482, 2.2586, 3.7498, 3.2324, 3.7054, 3.7844, 3.4931, 3.5792],
+       device='cuda:3'), covar=tensor([0.0722, 0.1109, 0.0128, 0.0514, 0.0159, 0.0242, 0.0227, 0.0204],
+       device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0104, 0.0089, 0.0154, 0.0087, 0.0099, 0.0091, 0.0087],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 06:23:39,549 INFO [train.py:892] (3/4) Epoch 34, batch 1050, loss[loss=0.1487, simple_loss=0.23, pruned_loss=0.03371, over 19849.00 frames. ], tot_loss[loss=0.16, simple_loss=0.2393, pruned_loss=0.04032, over 3930240.88 frames. ], batch size: 106, lr: 4.59e-03, grad_scale: 16.0
+2023-03-29 06:23:49,547 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:23:55,986 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:25:31,598 INFO [train.py:892] (3/4) Epoch 34, batch 1100, loss[loss=0.1526, simple_loss=0.2347, pruned_loss=0.03523, over 19730.00 frames. ], tot_loss[loss=0.1598, simple_loss=0.2395, pruned_loss=0.04001, over 3934179.42 frames. ], batch size: 80, lr: 4.59e-03, grad_scale: 16.0
+2023-03-29 06:25:55,966 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.864e+02 3.591e+02 4.112e+02 5.051e+02 8.825e+02, threshold=8.223e+02, percent-clipped=0.0
+2023-03-29 06:26:12,099 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9142, 4.0188, 2.4289, 4.1570, 4.3357, 1.9125, 3.5984, 3.2000],
+       device='cuda:3'), covar=tensor([0.0715, 0.0787, 0.2689, 0.0765, 0.0549, 0.2935, 0.1056, 0.0926],
+       device='cuda:3'), in_proj_covar=tensor([0.0237, 0.0260, 0.0233, 0.0281, 0.0260, 0.0207, 0.0244, 0.0201],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 06:26:15,587 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:27:28,384 INFO [train.py:892] (3/4) Epoch 34, batch 1150, loss[loss=0.1792, simple_loss=0.249, pruned_loss=0.05472, over 19799.00 frames. ], tot_loss[loss=0.1593, simple_loss=0.239, pruned_loss=0.03986, over 3938131.45 frames. ], batch size: 224, lr: 4.59e-03, grad_scale: 16.0
+2023-03-29 06:28:19,832 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:28:26,578 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62391.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:28:46,508 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:29:25,701 INFO [train.py:892] (3/4) Epoch 34, batch 1200, loss[loss=0.1536, simple_loss=0.2297, pruned_loss=0.03877, over 19759.00 frames. ], tot_loss[loss=0.1595, simple_loss=0.239, pruned_loss=0.03998, over 3941705.64 frames. ], batch size: 100, lr: 4.59e-03, grad_scale: 16.0
+2023-03-29 06:29:36,775 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0
+2023-03-29 06:29:49,455 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.475e+02 3.662e+02 4.259e+02 5.186e+02 8.409e+02, threshold=8.517e+02, percent-clipped=1.0
+2023-03-29 06:30:12,883 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:30:40,857 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62447.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:30:51,257 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62452.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:31:20,406 INFO [train.py:892] (3/4) Epoch 34, batch 1250, loss[loss=0.1616, simple_loss=0.2329, pruned_loss=0.04516, over 19765.00 frames. ], tot_loss[loss=0.1598, simple_loss=0.239, pruned_loss=0.04024, over 3943206.11 frames. ], batch size: 152, lr: 4.59e-03, grad_scale: 16.0
+2023-03-29 06:31:42,209 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4756, 3.1284, 3.4167, 2.9699, 3.6021, 3.5955, 4.2553, 4.6932],
+       device='cuda:3'), covar=tensor([0.0505, 0.1616, 0.1511, 0.2323, 0.1728, 0.1370, 0.0647, 0.0507],
+       device='cuda:3'), in_proj_covar=tensor([0.0258, 0.0245, 0.0272, 0.0258, 0.0303, 0.0261, 0.0238, 0.0263],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 06:31:46,616 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.2874, 5.5657, 5.7959, 5.5856, 5.5657, 5.4491, 5.5028, 5.3477],
+       device='cuda:3'), covar=tensor([0.1527, 0.1380, 0.0821, 0.1144, 0.0654, 0.0808, 0.1925, 0.1813],
+       device='cuda:3'), in_proj_covar=tensor([0.0300, 0.0341, 0.0373, 0.0306, 0.0280, 0.0289, 0.0369, 0.0396],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 06:31:46,641 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62478.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:33:12,470 INFO [train.py:892] (3/4) Epoch 34, batch 1300, loss[loss=0.1635, simple_loss=0.2428, pruned_loss=0.04208, over 19740.00 frames. ], tot_loss[loss=0.1599, simple_loss=0.2392, pruned_loss=0.04032, over 3943716.70 frames. ], batch size: 221, lr: 4.58e-03, grad_scale: 16.0
+2023-03-29 06:33:37,403 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.393e+02 3.741e+02 4.359e+02 5.306e+02 1.139e+03, threshold=8.717e+02, percent-clipped=4.0
+2023-03-29 06:33:43,481 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0
+2023-03-29 06:33:51,268 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7574, 2.8824, 2.9124, 2.8008, 2.7581, 2.8556, 2.8313, 2.9834],
+       device='cuda:3'), covar=tensor([0.0306, 0.0336, 0.0315, 0.0319, 0.0444, 0.0337, 0.0367, 0.0374],
+       device='cuda:3'), in_proj_covar=tensor([0.0090, 0.0084, 0.0087, 0.0080, 0.0094, 0.0087, 0.0103, 0.0076],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 06:34:21,166 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62545.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:34:53,843 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8356, 2.4090, 2.8461, 3.0721, 3.5090, 3.8525, 3.6471, 3.6989],
+       device='cuda:3'), covar=tensor([0.0979, 0.1577, 0.1237, 0.0676, 0.0460, 0.0252, 0.0414, 0.0421],
+       device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0171, 0.0181, 0.0153, 0.0138, 0.0135, 0.0126, 0.0119],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 06:35:09,081 INFO [train.py:892] (3/4) Epoch 34, batch 1350, loss[loss=0.1389, simple_loss=0.2181, pruned_loss=0.02979, over 19863.00 frames. ], tot_loss[loss=0.1593, simple_loss=0.2392, pruned_loss=0.03968, over 3943361.38 frames. ], batch size: 118, lr: 4.58e-03, grad_scale: 16.0
+2023-03-29 06:35:18,025 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62570.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:35:36,818 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-29 06:35:50,279 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:36:41,115 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62606.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:36:46,287 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0
+2023-03-29 06:36:59,935 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8561, 2.7942, 1.7684, 3.2980, 3.0543, 3.2726, 3.2992, 2.7926],
+       device='cuda:3'), covar=tensor([0.0698, 0.0813, 0.1771, 0.0736, 0.0783, 0.0546, 0.0799, 0.0799],
+       device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0146, 0.0145, 0.0155, 0.0136, 0.0138, 0.0150, 0.0148],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 06:37:03,352 INFO [train.py:892] (3/4) Epoch 34, batch 1400, loss[loss=0.1735, simple_loss=0.2638, pruned_loss=0.04165, over 19580.00 frames. ], tot_loss[loss=0.1579, simple_loss=0.2377, pruned_loss=0.03907, over 3945382.16 frames. ], batch size: 53, lr: 4.58e-03, grad_scale: 16.0
+2023-03-29 06:37:08,686 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62618.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:37:21,721 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0195, 5.3041, 5.3735, 5.2504, 4.9494, 5.3084, 4.8557, 4.8423],
+       device='cuda:3'), covar=tensor([0.0447, 0.0455, 0.0455, 0.0411, 0.0620, 0.0523, 0.0624, 0.0880],
+       device='cuda:3'), in_proj_covar=tensor([0.0279, 0.0295, 0.0308, 0.0271, 0.0275, 0.0260, 0.0275, 0.0323],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 06:37:26,823 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.572e+02 3.612e+02 4.260e+02 5.220e+02 9.254e+02, threshold=8.520e+02, percent-clipped=0.0
+2023-03-29 06:37:35,476 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:38:13,439 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:38:58,270 INFO [train.py:892] (3/4) Epoch 34, batch 1450, loss[loss=0.1753, simple_loss=0.2458, pruned_loss=0.05239, over 19836.00 frames. ], tot_loss[loss=0.1583, simple_loss=0.238, pruned_loss=0.03926, over 3945339.63 frames. ], batch size: 171, lr: 4.58e-03, grad_scale: 16.0
+2023-03-29 06:39:00,681 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62666.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:40:16,657 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0871, 5.3413, 5.5496, 5.3400, 5.3609, 5.1753, 5.2649, 5.0943],
+       device='cuda:3'), covar=tensor([0.1494, 0.1629, 0.0858, 0.1208, 0.0654, 0.0845, 0.1882, 0.1967],
+       device='cuda:3'), in_proj_covar=tensor([0.0301, 0.0340, 0.0374, 0.0306, 0.0280, 0.0290, 0.0369, 0.0396],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 06:40:21,243 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0
+2023-03-29 06:40:56,794 INFO [train.py:892] (3/4) Epoch 34, batch 1500, loss[loss=0.1449, simple_loss=0.2244, pruned_loss=0.03271, over 19485.00 frames. ], tot_loss[loss=0.1582, simple_loss=0.238, pruned_loss=0.03919, over 3946393.02 frames. ], batch size: 43, lr: 4.58e-03, grad_scale: 32.0
+2023-03-29 06:41:19,207 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.451e+02 3.877e+02 4.374e+02 5.206e+02 9.071e+02, threshold=8.749e+02, percent-clipped=3.0
+2023-03-29 06:41:24,829 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:41:34,646 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0
+2023-03-29 06:42:11,437 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:42:53,779 INFO [train.py:892] (3/4) Epoch 34, batch 1550, loss[loss=0.1732, simple_loss=0.2491, pruned_loss=0.04864, over 19833.00 frames. ], tot_loss[loss=0.1584, simple_loss=0.2382, pruned_loss=0.03927, over 3947409.52 frames. ], batch size: 144, lr: 4.58e-03, grad_scale: 32.0
+2023-03-29 06:43:11,374 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:43:22,826 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:43:32,099 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.9954, 6.2531, 6.3282, 6.1597, 6.0008, 6.2480, 5.6194, 5.6168],
+       device='cuda:3'), covar=tensor([0.0376, 0.0426, 0.0463, 0.0384, 0.0586, 0.0470, 0.0603, 0.0991],
+       device='cuda:3'), in_proj_covar=tensor([0.0277, 0.0291, 0.0306, 0.0268, 0.0273, 0.0258, 0.0273, 0.0319],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 06:44:23,647 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.19 vs. limit=5.0
+2023-03-29 06:44:53,489 INFO [train.py:892] (3/4) Epoch 34, batch 1600, loss[loss=0.1336, simple_loss=0.2091, pruned_loss=0.02906, over 19791.00 frames. ], tot_loss[loss=0.1576, simple_loss=0.2378, pruned_loss=0.0387, over 3946365.11 frames. ], batch size: 105, lr: 4.57e-03, grad_scale: 32.0
+2023-03-29 06:45:16,009 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.873e+02 3.915e+02 4.396e+02 5.347e+02 9.230e+02, threshold=8.791e+02, percent-clipped=1.0
+2023-03-29 06:45:17,177 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62826.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:45:34,351 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:45:59,385 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0
+2023-03-29 06:46:47,392 INFO [train.py:892] (3/4) Epoch 34, batch 1650, loss[loss=0.1483, simple_loss=0.2315, pruned_loss=0.03255, over 19728.00 frames. ], tot_loss[loss=0.1578, simple_loss=0.2381, pruned_loss=0.0388, over 3945747.40 frames. ], batch size: 51, lr: 4.57e-03, grad_scale: 32.0
+2023-03-29 06:47:18,713 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.72 vs. limit=5.0
+2023-03-29 06:48:09,570 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62901.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:48:42,538 INFO [train.py:892] (3/4) Epoch 34, batch 1700, loss[loss=0.1317, simple_loss=0.2132, pruned_loss=0.0251, over 19841.00 frames. ], tot_loss[loss=0.1571, simple_loss=0.2371, pruned_loss=0.03858, over 3948139.09 frames. ], batch size: 115, lr: 4.57e-03, grad_scale: 32.0
+2023-03-29 06:49:05,134 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.243e+02 3.824e+02 4.369e+02 5.372e+02 9.431e+02, threshold=8.739e+02, percent-clipped=1.0
+2023-03-29 06:49:14,476 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:49:41,270 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:50:26,210 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.9822, 2.2293, 2.0620, 1.4804, 2.0758, 2.2612, 2.0974, 2.1454],
+       device='cuda:3'), covar=tensor([0.0433, 0.0334, 0.0365, 0.0639, 0.0459, 0.0325, 0.0352, 0.0321],
+       device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0101, 0.0103, 0.0105, 0.0108, 0.0091, 0.0091, 0.0091],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 06:50:34,868 INFO [train.py:892] (3/4) Epoch 34, batch 1750, loss[loss=0.1402, simple_loss=0.2212, pruned_loss=0.02964, over 19619.00 frames. ], tot_loss[loss=0.1583, simple_loss=0.238, pruned_loss=0.03929, over 3948293.45 frames. ], batch size: 52, lr: 4.57e-03, grad_scale: 32.0
+2023-03-29 06:50:56,825 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:51:06,270 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6028, 2.0877, 2.3904, 2.8797, 3.1683, 3.3191, 3.2351, 3.2438],
+       device='cuda:3'), covar=tensor([0.1101, 0.1788, 0.1524, 0.0761, 0.0645, 0.0389, 0.0497, 0.0591],
+       device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0170, 0.0179, 0.0152, 0.0138, 0.0134, 0.0125, 0.0118],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 06:51:13,407 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.59 vs. limit=5.0
+2023-03-29 06:52:08,807 INFO [train.py:892] (3/4) Epoch 34, batch 1800, loss[loss=0.1441, simple_loss=0.2262, pruned_loss=0.03105, over 19805.00 frames. ], tot_loss[loss=0.1593, simple_loss=0.2389, pruned_loss=0.03983, over 3946423.37 frames. ], batch size: 107, lr: 4.57e-03, grad_scale: 32.0
+2023-03-29 06:52:20,261 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63022.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:52:26,927 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.273e+02 3.774e+02 4.619e+02 5.631e+02 1.047e+03, threshold=9.238e+02, percent-clipped=1.0
+2023-03-29 06:53:07,297 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63047.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:53:40,507 INFO [train.py:892] (3/4) Epoch 34, batch 1850, loss[loss=0.1577, simple_loss=0.2496, pruned_loss=0.03283, over 19821.00 frames. ], tot_loss[loss=0.1602, simple_loss=0.2407, pruned_loss=0.03986, over 3946117.56 frames. ], batch size: 57, lr: 4.56e-03, grad_scale: 32.0
+2023-03-29 06:54:44,119 INFO [train.py:892] (3/4) Epoch 35, batch 0, loss[loss=0.1638, simple_loss=0.2317, pruned_loss=0.04792, over 19827.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2317, pruned_loss=0.04792, over 19827.00 frames. ], batch size: 127, lr: 4.50e-03, grad_scale: 32.0
+2023-03-29 06:54:44,120 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-29 06:55:00,252 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0206, 2.7345, 3.1762, 3.2666, 3.6577, 4.0715, 3.8411, 3.9261],
+       device='cuda:3'), covar=tensor([0.0964, 0.1479, 0.1203, 0.0718, 0.0501, 0.0253, 0.0412, 0.0385],
+       device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0169, 0.0179, 0.0152, 0.0138, 0.0134, 0.0125, 0.0118],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 06:55:18,885 INFO [train.py:926] (3/4) Epoch 35, validation: loss=0.1837, simple_loss=0.2499, pruned_loss=0.05876, over 2883724.00 frames. 
+2023-03-29 06:55:18,886 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-29 06:56:18,857 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:57:16,275 INFO [train.py:892] (3/4) Epoch 35, batch 50, loss[loss=0.1327, simple_loss=0.2142, pruned_loss=0.02559, over 19676.00 frames. ], tot_loss[loss=0.1489, simple_loss=0.2269, pruned_loss=0.03545, over 892266.96 frames. ], batch size: 55, lr: 4.50e-03, grad_scale: 32.0
+2023-03-29 06:57:28,122 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.370e+02 3.277e+02 3.899e+02 4.647e+02 1.054e+03, threshold=7.797e+02, percent-clipped=1.0
+2023-03-29 06:57:36,976 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 06:57:56,263 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0
+2023-03-29 06:59:14,222 INFO [train.py:892] (3/4) Epoch 35, batch 100, loss[loss=0.1361, simple_loss=0.2168, pruned_loss=0.02769, over 19793.00 frames. ], tot_loss[loss=0.1522, simple_loss=0.2312, pruned_loss=0.03658, over 1570122.60 frames. ], batch size: 45, lr: 4.49e-03, grad_scale: 32.0
+2023-03-29 07:00:25,523 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:01:09,264 INFO [train.py:892] (3/4) Epoch 35, batch 150, loss[loss=0.1473, simple_loss=0.23, pruned_loss=0.03232, over 19883.00 frames. ], tot_loss[loss=0.1539, simple_loss=0.2333, pruned_loss=0.03724, over 2097484.90 frames. ], batch size: 88, lr: 4.49e-03, grad_scale: 16.0
+2023-03-29 07:01:10,376 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63221.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:01:22,857 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.863e+02 3.708e+02 4.219e+02 5.359e+02 8.315e+02, threshold=8.439e+02, percent-clipped=1.0
+2023-03-29 07:01:57,366 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:02:14,847 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63249.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:03:02,125 INFO [train.py:892] (3/4) Epoch 35, batch 200, loss[loss=0.1327, simple_loss=0.2116, pruned_loss=0.02691, over 19845.00 frames. ], tot_loss[loss=0.1545, simple_loss=0.2336, pruned_loss=0.03763, over 2509314.04 frames. ], batch size: 109, lr: 4.49e-03, grad_scale: 16.0
+2023-03-29 07:03:30,378 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:03:45,912 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:04:09,733 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8495, 2.8194, 2.9476, 2.2866, 3.0652, 2.5644, 2.8770, 2.9310],
+       device='cuda:3'), covar=tensor([0.0773, 0.0483, 0.0571, 0.0914, 0.0404, 0.0549, 0.0527, 0.0379],
+       device='cuda:3'), in_proj_covar=tensor([0.0081, 0.0089, 0.0086, 0.0114, 0.0082, 0.0085, 0.0083, 0.0076],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 07:04:56,634 INFO [train.py:892] (3/4) Epoch 35, batch 250, loss[loss=0.136, simple_loss=0.2163, pruned_loss=0.02785, over 19699.00 frames. ], tot_loss[loss=0.155, simple_loss=0.2346, pruned_loss=0.0377, over 2827411.42 frames. ], batch size: 82, lr: 4.49e-03, grad_scale: 16.0
+2023-03-29 07:05:00,065 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63322.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:05:09,782 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.149e+02 3.669e+02 4.457e+02 5.249e+02 9.820e+02, threshold=8.914e+02, percent-clipped=1.0
+2023-03-29 07:06:48,149 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63370.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:06:49,432 INFO [train.py:892] (3/4) Epoch 35, batch 300, loss[loss=0.1423, simple_loss=0.2223, pruned_loss=0.03114, over 19707.00 frames. ], tot_loss[loss=0.1551, simple_loss=0.2347, pruned_loss=0.03776, over 3075718.06 frames. ], batch size: 48, lr: 4.49e-03, grad_scale: 16.0
+2023-03-29 07:08:02,805 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2814, 3.6660, 3.7710, 4.2771, 3.0037, 3.2603, 2.7491, 2.8443],
+       device='cuda:3'), covar=tensor([0.0536, 0.2211, 0.1083, 0.0457, 0.2166, 0.1112, 0.1457, 0.1715],
+       device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0334, 0.0252, 0.0209, 0.0252, 0.0214, 0.0224, 0.0220],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 07:08:45,622 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:08:46,587 INFO [train.py:892] (3/4) Epoch 35, batch 350, loss[loss=0.1371, simple_loss=0.2263, pruned_loss=0.02399, over 19680.00 frames. ], tot_loss[loss=0.1562, simple_loss=0.2358, pruned_loss=0.03828, over 3270336.05 frames. ], batch size: 52, lr: 4.49e-03, grad_scale: 16.0
+2023-03-29 07:08:49,684 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63422.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:09:00,273 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.282e+02 3.589e+02 4.143e+02 4.777e+02 8.790e+02, threshold=8.287e+02, percent-clipped=0.0
+2023-03-29 07:09:08,534 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:10:37,167 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0809, 3.0229, 1.9423, 3.6281, 3.3466, 3.5893, 3.6321, 2.9689],
+       device='cuda:3'), covar=tensor([0.0701, 0.0768, 0.1827, 0.0686, 0.0697, 0.0498, 0.0771, 0.0805],
+       device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0146, 0.0145, 0.0155, 0.0136, 0.0138, 0.0150, 0.0148],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 07:10:44,407 INFO [train.py:892] (3/4) Epoch 35, batch 400, loss[loss=0.1378, simple_loss=0.2184, pruned_loss=0.02856, over 19777.00 frames. ], tot_loss[loss=0.1561, simple_loss=0.2356, pruned_loss=0.03831, over 3420315.23 frames. ], batch size: 152, lr: 4.48e-03, grad_scale: 16.0
+2023-03-29 07:10:59,239 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:10:59,813 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0
+2023-03-29 07:11:08,107 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63481.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:11:13,409 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:12:42,403 INFO [train.py:892] (3/4) Epoch 35, batch 450, loss[loss=0.1612, simple_loss=0.2443, pruned_loss=0.03909, over 19749.00 frames. ], tot_loss[loss=0.1561, simple_loss=0.2357, pruned_loss=0.03828, over 3537922.66 frames. ], batch size: 110, lr: 4.48e-03, grad_scale: 16.0
+2023-03-29 07:12:56,140 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.456e+02 3.739e+02 4.362e+02 5.360e+02 8.901e+02, threshold=8.724e+02, percent-clipped=1.0
+2023-03-29 07:14:32,297 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:14:35,599 INFO [train.py:892] (3/4) Epoch 35, batch 500, loss[loss=0.1574, simple_loss=0.2357, pruned_loss=0.03954, over 19798.00 frames. ], tot_loss[loss=0.157, simple_loss=0.2361, pruned_loss=0.03893, over 3629953.61 frames. ], batch size: 65, lr: 4.48e-03, grad_scale: 16.0
+2023-03-29 07:14:48,677 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8564, 2.8475, 1.8103, 3.2729, 3.0628, 3.1747, 3.2967, 2.6569],
+       device='cuda:3'), covar=tensor([0.0629, 0.0745, 0.1734, 0.0595, 0.0636, 0.0590, 0.0578, 0.0863],
+       device='cuda:3'), in_proj_covar=tensor([0.0144, 0.0145, 0.0145, 0.0155, 0.0136, 0.0138, 0.0150, 0.0148],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 07:14:52,656 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63577.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:15:10,998 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:16:32,827 INFO [train.py:892] (3/4) Epoch 35, batch 550, loss[loss=0.1601, simple_loss=0.2483, pruned_loss=0.03602, over 19853.00 frames. ], tot_loss[loss=0.1567, simple_loss=0.2357, pruned_loss=0.03882, over 3701263.58 frames. ], batch size: 58, lr: 4.48e-03, grad_scale: 16.0
+2023-03-29 07:16:46,425 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0552, 3.4091, 3.5084, 4.0222, 2.7793, 3.2569, 2.6171, 2.6397],
+       device='cuda:3'), covar=tensor([0.0535, 0.1966, 0.0933, 0.0413, 0.1880, 0.0899, 0.1339, 0.1568],
+       device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0330, 0.0249, 0.0206, 0.0249, 0.0212, 0.0222, 0.0219],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 07:16:47,213 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.269e+02 3.713e+02 4.298e+02 5.040e+02 8.851e+02, threshold=8.596e+02, percent-clipped=1.0
+2023-03-29 07:16:52,894 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0
+2023-03-29 07:16:54,598 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:17:36,658 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63647.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:18:09,162 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=3.12 vs. limit=5.0
+2023-03-29 07:18:28,407 INFO [train.py:892] (3/4) Epoch 35, batch 600, loss[loss=0.1863, simple_loss=0.2709, pruned_loss=0.05086, over 19754.00 frames. ], tot_loss[loss=0.1569, simple_loss=0.2361, pruned_loss=0.03886, over 3754580.33 frames. ], batch size: 253, lr: 4.48e-03, grad_scale: 16.0
+2023-03-29 07:20:27,065 INFO [train.py:892] (3/4) Epoch 35, batch 650, loss[loss=0.1614, simple_loss=0.2374, pruned_loss=0.04268, over 19738.00 frames. ], tot_loss[loss=0.1578, simple_loss=0.2372, pruned_loss=0.03923, over 3797398.04 frames. ], batch size: 89, lr: 4.48e-03, grad_scale: 16.0
+2023-03-29 07:20:40,720 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.573e+02 3.756e+02 4.198e+02 4.970e+02 9.612e+02, threshold=8.396e+02, percent-clipped=1.0
+2023-03-29 07:22:16,891 INFO [train.py:892] (3/4) Epoch 35, batch 700, loss[loss=0.1423, simple_loss=0.2223, pruned_loss=0.0312, over 19872.00 frames. ], tot_loss[loss=0.1582, simple_loss=0.2378, pruned_loss=0.0393, over 3831130.20 frames. ], batch size: 108, lr: 4.47e-03, grad_scale: 16.0
+2023-03-29 07:22:29,942 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63776.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:22:34,144 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63778.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:23:51,377 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63810.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:24:14,521 INFO [train.py:892] (3/4) Epoch 35, batch 750, loss[loss=0.157, simple_loss=0.2393, pruned_loss=0.03733, over 19806.00 frames. ], tot_loss[loss=0.1587, simple_loss=0.2383, pruned_loss=0.03955, over 3857607.87 frames. ], batch size: 202, lr: 4.47e-03, grad_scale: 16.0
+2023-03-29 07:24:28,556 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.314e+02 3.905e+02 4.508e+02 5.292e+02 1.021e+03, threshold=9.015e+02, percent-clipped=3.0
+2023-03-29 07:24:46,951 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-29 07:25:13,235 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63845.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:26:13,517 INFO [train.py:892] (3/4) Epoch 35, batch 800, loss[loss=0.1447, simple_loss=0.2225, pruned_loss=0.03349, over 19819.00 frames. ], tot_loss[loss=0.1588, simple_loss=0.2382, pruned_loss=0.03968, over 3877810.95 frames. ], batch size: 202, lr: 4.47e-03, grad_scale: 16.0
+2023-03-29 07:26:14,590 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63871.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:26:17,393 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1956, 2.6841, 3.3325, 3.4607, 3.9371, 4.4936, 4.3145, 4.3552],
+       device='cuda:3'), covar=tensor([0.0860, 0.1632, 0.1194, 0.0626, 0.0379, 0.0201, 0.0297, 0.0329],
+       device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0169, 0.0179, 0.0152, 0.0138, 0.0134, 0.0126, 0.0117],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 07:26:29,434 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63877.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:27:10,630 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0
+2023-03-29 07:27:22,024 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0
+2023-03-29 07:27:22,109 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-29 07:27:37,306 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63906.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:28:01,360 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:28:11,426 INFO [train.py:892] (3/4) Epoch 35, batch 850, loss[loss=0.1435, simple_loss=0.2141, pruned_loss=0.03643, over 19824.00 frames. ], tot_loss[loss=0.1587, simple_loss=0.2379, pruned_loss=0.03978, over 3893309.64 frames. ], batch size: 128, lr: 4.47e-03, grad_scale: 16.0
+2023-03-29 07:28:17,419 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.2745, 5.0116, 4.9738, 5.3368, 4.8042, 5.5043, 5.4329, 5.6254],
+       device='cuda:3'), covar=tensor([0.0623, 0.0432, 0.0425, 0.0359, 0.0738, 0.0413, 0.0419, 0.0300],
+       device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0184, 0.0208, 0.0184, 0.0183, 0.0166, 0.0158, 0.0207],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 07:28:21,802 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:28:21,830 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:28:25,041 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.864e+02 3.650e+02 4.447e+02 5.355e+02 8.426e+02, threshold=8.894e+02, percent-clipped=0.0
+2023-03-29 07:28:53,882 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0
+2023-03-29 07:29:02,889 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63942.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:30:06,833 INFO [train.py:892] (3/4) Epoch 35, batch 900, loss[loss=0.1613, simple_loss=0.2397, pruned_loss=0.04142, over 19808.00 frames. ], tot_loss[loss=0.1572, simple_loss=0.2366, pruned_loss=0.03885, over 3906741.34 frames. ], batch size: 202, lr: 4.47e-03, grad_scale: 16.0
+2023-03-29 07:30:21,930 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:30:51,476 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6519, 2.7600, 1.6798, 3.0762, 2.8513, 2.9676, 3.0817, 2.5215],
+       device='cuda:3'), covar=tensor([0.0725, 0.0732, 0.1713, 0.0614, 0.0629, 0.0617, 0.0625, 0.0894],
+       device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0146, 0.0146, 0.0154, 0.0137, 0.0139, 0.0151, 0.0149],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 07:32:03,695 INFO [train.py:892] (3/4) Epoch 35, batch 950, loss[loss=0.1498, simple_loss=0.2241, pruned_loss=0.03777, over 19864.00 frames. ], tot_loss[loss=0.1571, simple_loss=0.2369, pruned_loss=0.03865, over 3917079.29 frames. ], batch size: 157, lr: 4.46e-03, grad_scale: 16.0
+2023-03-29 07:32:08,724 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64023.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:32:16,290 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.345e+02 3.650e+02 4.248e+02 4.892e+02 8.257e+02, threshold=8.496e+02, percent-clipped=0.0
+2023-03-29 07:33:58,701 INFO [train.py:892] (3/4) Epoch 35, batch 1000, loss[loss=0.1433, simple_loss=0.2232, pruned_loss=0.03167, over 19795.00 frames. ], tot_loss[loss=0.1567, simple_loss=0.2364, pruned_loss=0.03848, over 3923817.91 frames. ], batch size: 211, lr: 4.46e-03, grad_scale: 16.0
+2023-03-29 07:34:10,136 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64076.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:34:14,377 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64078.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:34:28,227 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:35:51,129 INFO [train.py:892] (3/4) Epoch 35, batch 1050, loss[loss=0.1394, simple_loss=0.223, pruned_loss=0.02792, over 19872.00 frames. ], tot_loss[loss=0.1565, simple_loss=0.2363, pruned_loss=0.03832, over 3930981.44 frames. ], batch size: 108, lr: 4.46e-03, grad_scale: 16.0
+2023-03-29 07:35:59,005 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64124.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:36:04,345 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64126.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:36:05,623 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.257e+02 3.875e+02 4.468e+02 5.208e+02 1.393e+03, threshold=8.936e+02, percent-clipped=2.0
+2023-03-29 07:36:11,447 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7320, 2.2385, 2.6334, 2.9806, 3.4039, 3.5306, 3.4300, 3.4627],
+       device='cuda:3'), covar=tensor([0.1059, 0.1737, 0.1408, 0.0755, 0.0478, 0.0349, 0.0455, 0.0512],
+       device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0171, 0.0181, 0.0153, 0.0139, 0.0135, 0.0128, 0.0119],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 07:37:37,863 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64166.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:37:51,331 INFO [train.py:892] (3/4) Epoch 35, batch 1100, loss[loss=0.1361, simple_loss=0.2144, pruned_loss=0.02883, over 19699.00 frames. ], tot_loss[loss=0.1575, simple_loss=0.2373, pruned_loss=0.03879, over 3935117.94 frames. ], batch size: 81, lr: 4.46e-03, grad_scale: 16.0
+2023-03-29 07:38:28,844 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3931, 2.8593, 4.6005, 3.9436, 4.3878, 4.5388, 4.3165, 4.2389],
+       device='cuda:3'), covar=tensor([0.0576, 0.0916, 0.0115, 0.0770, 0.0154, 0.0212, 0.0178, 0.0176],
+       device='cuda:3'), in_proj_covar=tensor([0.0101, 0.0105, 0.0090, 0.0153, 0.0087, 0.0100, 0.0091, 0.0087],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 07:38:58,336 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0901, 4.2420, 2.5241, 4.4827, 4.6531, 1.9264, 3.8833, 3.4137],
+       device='cuda:3'), covar=tensor([0.0707, 0.0788, 0.2809, 0.0718, 0.0606, 0.3060, 0.1024, 0.0901],
+       device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0263, 0.0235, 0.0284, 0.0262, 0.0208, 0.0243, 0.0202],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 07:39:00,736 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64201.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:39:48,467 INFO [train.py:892] (3/4) Epoch 35, batch 1150, loss[loss=0.1557, simple_loss=0.2316, pruned_loss=0.03989, over 19827.00 frames. ], tot_loss[loss=0.1575, simple_loss=0.2372, pruned_loss=0.03887, over 3938545.35 frames. ], batch size: 208, lr: 4.46e-03, grad_scale: 16.0
+2023-03-29 07:39:57,610 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:40:00,931 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.058e+02 3.821e+02 4.310e+02 5.075e+02 9.198e+02, threshold=8.620e+02, percent-clipped=1.0
+2023-03-29 07:40:13,271 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:40:38,275 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64242.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:40:43,855 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8719, 4.1932, 2.3706, 4.4799, 4.6629, 2.0869, 3.6107, 3.3745],
+       device='cuda:3'), covar=tensor([0.0799, 0.0745, 0.2713, 0.0709, 0.0495, 0.3003, 0.1185, 0.0947],
+       device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0264, 0.0235, 0.0285, 0.0263, 0.0209, 0.0243, 0.0203],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 07:40:59,591 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9140, 2.8687, 3.0840, 2.6837, 3.2258, 3.1919, 3.7632, 4.1048],
+       device='cuda:3'), covar=tensor([0.0644, 0.1661, 0.1650, 0.2244, 0.1664, 0.1536, 0.0708, 0.0790],
+       device='cuda:3'), in_proj_covar=tensor([0.0256, 0.0243, 0.0269, 0.0256, 0.0301, 0.0260, 0.0235, 0.0261],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 07:41:30,911 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9547, 3.0064, 1.8781, 3.5032, 3.2558, 3.4568, 3.5311, 2.8435],
+       device='cuda:3'), covar=tensor([0.0709, 0.0695, 0.1780, 0.0674, 0.0580, 0.0461, 0.0577, 0.0781],
+       device='cuda:3'), in_proj_covar=tensor([0.0145, 0.0147, 0.0146, 0.0155, 0.0137, 0.0139, 0.0151, 0.0149],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 07:41:41,989 INFO [train.py:892] (3/4) Epoch 35, batch 1200, loss[loss=0.1333, simple_loss=0.215, pruned_loss=0.02577, over 19774.00 frames. ], tot_loss[loss=0.1591, simple_loss=0.2391, pruned_loss=0.0395, over 3938030.83 frames. ], batch size: 46, lr: 4.46e-03, grad_scale: 16.0
+2023-03-29 07:41:46,068 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:41:48,214 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:42:26,903 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64290.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:42:33,839 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:43:37,224 INFO [train.py:892] (3/4) Epoch 35, batch 1250, loss[loss=0.1621, simple_loss=0.2504, pruned_loss=0.03691, over 19878.00 frames. ], tot_loss[loss=0.1596, simple_loss=0.2397, pruned_loss=0.03975, over 3941040.62 frames. ], batch size: 52, lr: 4.45e-03, grad_scale: 16.0
+2023-03-29 07:43:50,723 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.911e+02 3.890e+02 4.539e+02 5.516e+02 1.564e+03, threshold=9.078e+02, percent-clipped=1.0
+2023-03-29 07:44:36,359 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:45:36,353 INFO [train.py:892] (3/4) Epoch 35, batch 1300, loss[loss=0.147, simple_loss=0.2386, pruned_loss=0.02773, over 19846.00 frames. ], tot_loss[loss=0.1591, simple_loss=0.2392, pruned_loss=0.03954, over 3942611.40 frames. ], batch size: 58, lr: 4.45e-03, grad_scale: 16.0
+2023-03-29 07:45:55,572 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64379.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:46:35,542 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.76 vs. limit=5.0
+2023-03-29 07:46:59,430 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64406.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:47:31,792 INFO [train.py:892] (3/4) Epoch 35, batch 1350, loss[loss=0.1589, simple_loss=0.2399, pruned_loss=0.03896, over 19773.00 frames. ], tot_loss[loss=0.1589, simple_loss=0.2385, pruned_loss=0.03968, over 3945169.40 frames. ], batch size: 69, lr: 4.45e-03, grad_scale: 16.0
+2023-03-29 07:47:43,144 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0021, 3.4118, 3.6626, 3.3568, 3.4565, 3.7934, 3.4963, 3.8307],
+       device='cuda:3'), covar=tensor([0.1929, 0.0699, 0.0885, 0.0764, 0.1490, 0.0857, 0.0896, 0.0718],
+       device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0227, 0.0227, 0.0241, 0.0211, 0.0251, 0.0240, 0.0223],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 07:47:46,400 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.065e+02 3.591e+02 4.239e+02 5.222e+02 9.468e+02, threshold=8.478e+02, percent-clipped=2.0
+2023-03-29 07:48:08,171 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5109, 3.6177, 2.1997, 3.6647, 3.7915, 1.8063, 3.1847, 2.9602],
+       device='cuda:3'), covar=tensor([0.0780, 0.0838, 0.2770, 0.0988, 0.0684, 0.2771, 0.1186, 0.0945],
+       device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0263, 0.0234, 0.0284, 0.0262, 0.0208, 0.0242, 0.0202],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 07:49:14,690 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64465.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:49:16,673 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:49:18,732 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1737, 2.4760, 3.4260, 2.8365, 2.9157, 2.8209, 2.0693, 2.2394],
+       device='cuda:3'), covar=tensor([0.1167, 0.2858, 0.0691, 0.1113, 0.1946, 0.1467, 0.2728, 0.2663],
+       device='cuda:3'), in_proj_covar=tensor([0.0351, 0.0391, 0.0349, 0.0288, 0.0375, 0.0381, 0.0378, 0.0348],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 07:49:25,914 INFO [train.py:892] (3/4) Epoch 35, batch 1400, loss[loss=0.166, simple_loss=0.2414, pruned_loss=0.04525, over 19836.00 frames. ], tot_loss[loss=0.158, simple_loss=0.2371, pruned_loss=0.0394, over 3946708.47 frames. ], batch size: 145, lr: 4.45e-03, grad_scale: 16.0
+2023-03-29 07:50:37,380 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:51:03,651 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0
+2023-03-29 07:51:09,510 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64514.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:51:23,330 INFO [train.py:892] (3/4) Epoch 35, batch 1450, loss[loss=0.1291, simple_loss=0.2105, pruned_loss=0.02386, over 19866.00 frames. ], tot_loss[loss=0.1581, simple_loss=0.2374, pruned_loss=0.03943, over 3948991.08 frames. ], batch size: 46, lr: 4.45e-03, grad_scale: 16.0
+2023-03-29 07:51:35,240 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64526.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:51:36,074 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.625e+02 3.695e+02 4.182e+02 5.095e+02 9.377e+02, threshold=8.363e+02, percent-clipped=1.0
+2023-03-29 07:52:31,842 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64549.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:53:21,044 INFO [train.py:892] (3/4) Epoch 35, batch 1500, loss[loss=0.2893, simple_loss=0.3601, pruned_loss=0.1093, over 19400.00 frames. ], tot_loss[loss=0.1584, simple_loss=0.2378, pruned_loss=0.03953, over 3948895.44 frames. ], batch size: 431, lr: 4.45e-03, grad_scale: 16.0
+2023-03-29 07:53:24,948 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64572.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:54:00,633 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64588.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:54:02,586 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:55:14,011 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:55:15,305 INFO [train.py:892] (3/4) Epoch 35, batch 1550, loss[loss=0.1719, simple_loss=0.2713, pruned_loss=0.03622, over 19537.00 frames. ], tot_loss[loss=0.1593, simple_loss=0.2388, pruned_loss=0.03986, over 3948751.86 frames. ], batch size: 54, lr: 4.44e-03, grad_scale: 16.0
+2023-03-29 07:55:27,759 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.542e+02 3.591e+02 4.055e+02 5.054e+02 8.662e+02, threshold=8.110e+02, percent-clipped=1.0
+2023-03-29 07:56:18,355 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:56:23,014 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:56:55,526 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4921, 3.9768, 4.0602, 4.6705, 3.1662, 3.5813, 2.8444, 2.7968],
+       device='cuda:3'), covar=tensor([0.0526, 0.1895, 0.0859, 0.0376, 0.2092, 0.1048, 0.1443, 0.1712],
+       device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0330, 0.0252, 0.0207, 0.0251, 0.0212, 0.0222, 0.0218],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 07:57:07,380 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-29 07:57:10,753 INFO [train.py:892] (3/4) Epoch 35, batch 1600, loss[loss=0.1446, simple_loss=0.2169, pruned_loss=0.03618, over 19840.00 frames. ], tot_loss[loss=0.1574, simple_loss=0.2373, pruned_loss=0.03876, over 3950011.95 frames. ], batch size: 160, lr: 4.44e-03, grad_scale: 16.0
+2023-03-29 07:57:29,742 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0
+2023-03-29 07:57:31,435 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64679.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:58:22,561 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64701.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:58:27,348 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64703.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:58:31,704 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6082, 3.4681, 3.8433, 3.5023, 3.3191, 3.7782, 3.6126, 3.8966],
+       device='cuda:3'), covar=tensor([0.0824, 0.0420, 0.0390, 0.0445, 0.1462, 0.0597, 0.0505, 0.0384],
+       device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0226, 0.0226, 0.0240, 0.0211, 0.0250, 0.0239, 0.0223],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 07:58:40,525 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64709.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:59:00,108 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0730, 4.1450, 4.4737, 4.2676, 4.4066, 4.0107, 4.2200, 4.0005],
+       device='cuda:3'), covar=tensor([0.1626, 0.1918, 0.0984, 0.1317, 0.1053, 0.0990, 0.2023, 0.2096],
+       device='cuda:3'), in_proj_covar=tensor([0.0300, 0.0340, 0.0376, 0.0304, 0.0278, 0.0288, 0.0366, 0.0394],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 07:59:08,790 INFO [train.py:892] (3/4) Epoch 35, batch 1650, loss[loss=0.1783, simple_loss=0.2568, pruned_loss=0.04987, over 19781.00 frames. ], tot_loss[loss=0.1568, simple_loss=0.2366, pruned_loss=0.03848, over 3950189.32 frames. ], batch size: 247, lr: 4.44e-03, grad_scale: 16.0
+2023-03-29 07:59:21,888 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.745e+02 3.702e+02 4.321e+02 4.955e+02 1.270e+03, threshold=8.641e+02, percent-clipped=4.0
+2023-03-29 07:59:22,958 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64727.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 07:59:52,539 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3264, 4.4132, 4.7223, 4.4909, 4.6166, 4.2622, 4.4754, 4.2232],
+       device='cuda:3'), covar=tensor([0.1618, 0.1869, 0.0957, 0.1409, 0.1013, 0.0970, 0.1878, 0.2083],
+       device='cuda:3'), in_proj_covar=tensor([0.0300, 0.0340, 0.0376, 0.0304, 0.0278, 0.0288, 0.0366, 0.0394],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 08:00:03,441 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64745.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:00:07,591 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1515, 3.3764, 2.9195, 2.5676, 2.9429, 3.2850, 3.1416, 3.2830],
+       device='cuda:3'), covar=tensor([0.0319, 0.0322, 0.0330, 0.0590, 0.0354, 0.0351, 0.0307, 0.0248],
+       device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0101, 0.0104, 0.0104, 0.0108, 0.0091, 0.0092, 0.0091],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 08:00:20,331 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64752.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:00:46,930 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64764.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:01:02,062 INFO [train.py:892] (3/4) Epoch 35, batch 1700, loss[loss=0.1791, simple_loss=0.2601, pruned_loss=0.04907, over 19609.00 frames. ], tot_loss[loss=0.1573, simple_loss=0.2371, pruned_loss=0.0387, over 3951041.43 frames. ], batch size: 48, lr: 4.44e-03, grad_scale: 16.0
+2023-03-29 08:02:26,239 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64806.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:02:39,429 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64813.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:02:55,209 INFO [train.py:892] (3/4) Epoch 35, batch 1750, loss[loss=0.1644, simple_loss=0.2606, pruned_loss=0.0341, over 19543.00 frames. ], tot_loss[loss=0.1572, simple_loss=0.237, pruned_loss=0.03877, over 3950833.10 frames. ], batch size: 54, lr: 4.44e-03, grad_scale: 16.0
+2023-03-29 08:02:56,140 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:03:07,069 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.308e+02 3.852e+02 4.447e+02 5.753e+02 1.014e+03, threshold=8.894e+02, percent-clipped=4.0
+2023-03-29 08:04:35,070 INFO [train.py:892] (3/4) Epoch 35, batch 1800, loss[loss=0.1841, simple_loss=0.2705, pruned_loss=0.04883, over 19660.00 frames. ], tot_loss[loss=0.1569, simple_loss=0.2365, pruned_loss=0.03869, over 3951166.78 frames. ], batch size: 330, lr: 4.44e-03, grad_scale: 16.0
+2023-03-29 08:05:06,248 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:06:09,202 INFO [train.py:892] (3/4) Epoch 35, batch 1850, loss[loss=0.1464, simple_loss=0.2285, pruned_loss=0.03218, over 19826.00 frames. ], tot_loss[loss=0.1576, simple_loss=0.2379, pruned_loss=0.03867, over 3950990.77 frames. ], batch size: 57, lr: 4.43e-03, grad_scale: 16.0
+2023-03-29 08:07:11,825 INFO [train.py:892] (3/4) Epoch 36, batch 0, loss[loss=0.1316, simple_loss=0.2198, pruned_loss=0.02172, over 19769.00 frames. ], tot_loss[loss=0.1316, simple_loss=0.2198, pruned_loss=0.02172, over 19769.00 frames. ], batch size: 87, lr: 4.37e-03, grad_scale: 16.0
+2023-03-29 08:07:11,825 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-29 08:07:46,029 INFO [train.py:926] (3/4) Epoch 36, validation: loss=0.183, simple_loss=0.249, pruned_loss=0.05846, over 2883724.00 frames. 
+2023-03-29 08:07:46,031 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-29 08:07:48,053 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.083e+02 3.485e+02 4.267e+02 5.108e+02 8.561e+02, threshold=8.534e+02, percent-clipped=0.0
+2023-03-29 08:08:10,247 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:08:19,479 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.2057, 1.6503, 1.8681, 2.4439, 2.6888, 2.8035, 2.6762, 2.7626],
+       device='cuda:3'), covar=tensor([0.1198, 0.1891, 0.1722, 0.0828, 0.0594, 0.0463, 0.0518, 0.0492],
+       device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0168, 0.0180, 0.0152, 0.0137, 0.0133, 0.0126, 0.0117],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 08:08:30,835 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9165, 2.4871, 2.9334, 3.0735, 3.6957, 4.0312, 3.9236, 3.9120],
+       device='cuda:3'), covar=tensor([0.1028, 0.1542, 0.1262, 0.0729, 0.0431, 0.0277, 0.0397, 0.0447],
+       device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0168, 0.0179, 0.0152, 0.0137, 0.0133, 0.0126, 0.0117],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 08:08:32,707 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:09:41,974 INFO [train.py:892] (3/4) Epoch 36, batch 50, loss[loss=0.1462, simple_loss=0.23, pruned_loss=0.03126, over 19854.00 frames. ], tot_loss[loss=0.1503, simple_loss=0.2297, pruned_loss=0.03545, over 891958.74 frames. ], batch size: 115, lr: 4.37e-03, grad_scale: 16.0
+2023-03-29 08:09:44,562 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:10:42,243 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65001.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:10:49,916 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65004.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:11:33,524 INFO [train.py:892] (3/4) Epoch 36, batch 100, loss[loss=0.1672, simple_loss=0.2445, pruned_loss=0.04501, over 19780.00 frames. ], tot_loss[loss=0.153, simple_loss=0.2331, pruned_loss=0.03643, over 1570671.60 frames. ], batch size: 215, lr: 4.37e-03, grad_scale: 16.0
+2023-03-29 08:11:35,870 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.481e+02 3.835e+02 4.535e+02 5.519e+02 9.407e+02, threshold=9.071e+02, percent-clipped=1.0
+2023-03-29 08:11:59,528 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65037.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 08:12:28,237 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:12:42,373 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:12:50,284 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65059.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:13:25,810 INFO [train.py:892] (3/4) Epoch 36, batch 150, loss[loss=0.1438, simple_loss=0.2245, pruned_loss=0.03157, over 19837.00 frames. ], tot_loss[loss=0.1557, simple_loss=0.2357, pruned_loss=0.03784, over 2098457.70 frames. ], batch size: 43, lr: 4.37e-03, grad_scale: 16.0
+2023-03-29 08:13:44,450 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5906, 4.5665, 2.7919, 4.8930, 5.0784, 2.3136, 4.2202, 3.7649],
+       device='cuda:3'), covar=tensor([0.0568, 0.0771, 0.2686, 0.0649, 0.0570, 0.2825, 0.0937, 0.0872],
+       device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0261, 0.0233, 0.0281, 0.0261, 0.0206, 0.0242, 0.0201],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 08:14:19,331 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1451, 3.0217, 3.3168, 2.8749, 3.4338, 3.4164, 4.0492, 4.4385],
+       device='cuda:3'), covar=tensor([0.0613, 0.1741, 0.1505, 0.2206, 0.1729, 0.1491, 0.0705, 0.0645],
+       device='cuda:3'), in_proj_covar=tensor([0.0257, 0.0244, 0.0272, 0.0257, 0.0304, 0.0262, 0.0237, 0.0263],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 08:14:25,169 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65101.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:14:39,347 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65108.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:14:58,820 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:15:10,932 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65121.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:15:22,341 INFO [train.py:892] (3/4) Epoch 36, batch 200, loss[loss=0.1257, simple_loss=0.2074, pruned_loss=0.02201, over 19818.00 frames. ], tot_loss[loss=0.1559, simple_loss=0.2357, pruned_loss=0.03807, over 2509020.64 frames. ], batch size: 96, lr: 4.36e-03, grad_scale: 16.0
+2023-03-29 08:15:24,692 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.519e+02 3.588e+02 4.285e+02 5.132e+02 1.208e+03, threshold=8.571e+02, percent-clipped=3.0
+2023-03-29 08:15:30,370 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0
+2023-03-29 08:17:01,851 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:17:15,483 INFO [train.py:892] (3/4) Epoch 36, batch 250, loss[loss=0.1414, simple_loss=0.2164, pruned_loss=0.03317, over 19808.00 frames. ], tot_loss[loss=0.1557, simple_loss=0.2351, pruned_loss=0.03816, over 2828389.73 frames. ], batch size: 117, lr: 4.36e-03, grad_scale: 16.0
+2023-03-29 08:19:08,892 INFO [train.py:892] (3/4) Epoch 36, batch 300, loss[loss=0.1316, simple_loss=0.2123, pruned_loss=0.02546, over 19813.00 frames. ], tot_loss[loss=0.1556, simple_loss=0.235, pruned_loss=0.0381, over 3078405.19 frames. ], batch size: 103, lr: 4.36e-03, grad_scale: 32.0
+2023-03-29 08:19:12,294 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.658e+02 3.728e+02 4.313e+02 5.514e+02 9.328e+02, threshold=8.626e+02, percent-clipped=1.0
+2023-03-29 08:19:54,098 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65245.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:21:04,018 INFO [train.py:892] (3/4) Epoch 36, batch 350, loss[loss=0.1645, simple_loss=0.2374, pruned_loss=0.04584, over 19872.00 frames. ], tot_loss[loss=0.1562, simple_loss=0.2357, pruned_loss=0.03835, over 3271698.03 frames. ], batch size: 138, lr: 4.36e-03, grad_scale: 32.0
+2023-03-29 08:21:23,897 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65285.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:21:42,318 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:22:00,258 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.3185, 2.3559, 2.4575, 2.3561, 2.3605, 2.5339, 2.3461, 2.4467],
+       device='cuda:3'), covar=tensor([0.0405, 0.0345, 0.0350, 0.0337, 0.0496, 0.0311, 0.0462, 0.0402],
+       device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0085, 0.0088, 0.0082, 0.0095, 0.0087, 0.0104, 0.0077],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 08:22:08,916 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65304.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:22:56,436 INFO [train.py:892] (3/4) Epoch 36, batch 400, loss[loss=0.1642, simple_loss=0.2487, pruned_loss=0.03985, over 19633.00 frames. ], tot_loss[loss=0.1561, simple_loss=0.2357, pruned_loss=0.03825, over 3422499.70 frames. ], batch size: 72, lr: 4.36e-03, grad_scale: 32.0
+2023-03-29 08:22:58,380 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.673e+02 3.890e+02 4.309e+02 5.103e+02 7.724e+02, threshold=8.618e+02, percent-clipped=0.0
+2023-03-29 08:23:09,870 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65332.0, num_to_drop=1, layers_to_drop={3}
+2023-03-29 08:23:18,221 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1010, 3.1578, 3.2930, 3.3612, 3.0892, 3.3707, 2.9953, 3.2760],
+       device='cuda:3'), covar=tensor([0.0309, 0.0367, 0.0284, 0.0201, 0.0338, 0.0235, 0.0345, 0.0354],
+       device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0085, 0.0088, 0.0082, 0.0094, 0.0087, 0.0104, 0.0077],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 08:23:29,701 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9101, 5.1512, 5.1933, 5.1073, 4.8545, 5.1891, 4.6759, 4.7298],
+       device='cuda:3'), covar=tensor([0.0479, 0.0491, 0.0465, 0.0430, 0.0602, 0.0490, 0.0676, 0.0985],
+       device='cuda:3'), in_proj_covar=tensor([0.0280, 0.0294, 0.0306, 0.0268, 0.0276, 0.0260, 0.0275, 0.0322],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 08:23:45,328 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:23:58,353 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65352.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:24:14,613 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65359.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:24:54,091 INFO [train.py:892] (3/4) Epoch 36, batch 450, loss[loss=0.1901, simple_loss=0.2731, pruned_loss=0.05354, over 19625.00 frames. ], tot_loss[loss=0.1583, simple_loss=0.238, pruned_loss=0.03933, over 3539966.35 frames. ], batch size: 359, lr: 4.36e-03, grad_scale: 16.0
+2023-03-29 08:25:49,822 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65401.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:26:03,685 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:26:07,321 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:26:15,260 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:26:46,790 INFO [train.py:892] (3/4) Epoch 36, batch 500, loss[loss=0.2162, simple_loss=0.3185, pruned_loss=0.05694, over 18731.00 frames. ], tot_loss[loss=0.1587, simple_loss=0.2387, pruned_loss=0.0394, over 3630519.72 frames. ], batch size: 564, lr: 4.35e-03, grad_scale: 16.0
+2023-03-29 08:26:52,434 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.389e+02 3.662e+02 4.205e+02 4.670e+02 6.884e+02, threshold=8.409e+02, percent-clipped=0.0
+2023-03-29 08:27:38,015 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65449.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:27:55,872 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:28:04,817 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4366, 4.5293, 2.7009, 4.8116, 5.0212, 2.1204, 4.2501, 3.6638],
+       device='cuda:3'), covar=tensor([0.0598, 0.0774, 0.2456, 0.0669, 0.0439, 0.2654, 0.0832, 0.0812],
+       device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0261, 0.0234, 0.0282, 0.0261, 0.0206, 0.0242, 0.0202],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 08:28:34,946 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7302, 4.5201, 5.0640, 4.5553, 4.1655, 4.7989, 4.6386, 5.1833],
+       device='cuda:3'), covar=tensor([0.0863, 0.0409, 0.0370, 0.0430, 0.0884, 0.0590, 0.0549, 0.0353],
+       device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0225, 0.0225, 0.0237, 0.0208, 0.0248, 0.0238, 0.0222],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 08:28:38,060 INFO [train.py:892] (3/4) Epoch 36, batch 550, loss[loss=0.1808, simple_loss=0.2593, pruned_loss=0.05116, over 19795.00 frames. ], tot_loss[loss=0.1576, simple_loss=0.2372, pruned_loss=0.039, over 3702387.20 frames. ], batch size: 247, lr: 4.35e-03, grad_scale: 16.0
+2023-03-29 08:29:04,498 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6321, 3.4664, 3.9019, 3.0030, 3.9889, 3.2471, 3.4449, 3.8962],
+       device='cuda:3'), covar=tensor([0.0729, 0.0409, 0.0556, 0.0765, 0.0387, 0.0406, 0.0464, 0.0295],
+       device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0091, 0.0088, 0.0114, 0.0083, 0.0087, 0.0084, 0.0078],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 08:29:49,252 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6191, 2.6463, 2.8693, 2.5753, 3.0738, 3.0521, 3.5125, 3.8165],
+       device='cuda:3'), covar=tensor([0.0684, 0.1724, 0.1612, 0.2127, 0.1535, 0.1473, 0.0739, 0.0658],
+       device='cuda:3'), in_proj_covar=tensor([0.0257, 0.0244, 0.0272, 0.0258, 0.0304, 0.0262, 0.0237, 0.0263],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 08:29:51,237 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1736, 3.1231, 3.3296, 2.6517, 3.3922, 2.8926, 3.1643, 3.2658],
+       device='cuda:3'), covar=tensor([0.0776, 0.0529, 0.0520, 0.0802, 0.0444, 0.0453, 0.0453, 0.0395],
+       device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0091, 0.0087, 0.0114, 0.0083, 0.0087, 0.0084, 0.0078],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 08:30:30,661 INFO [train.py:892] (3/4) Epoch 36, batch 600, loss[loss=0.1602, simple_loss=0.2397, pruned_loss=0.04033, over 19752.00 frames. ], tot_loss[loss=0.1582, simple_loss=0.238, pruned_loss=0.03924, over 3758614.49 frames. ], batch size: 276, lr: 4.35e-03, grad_scale: 16.0
+2023-03-29 08:30:34,422 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.355e+02 3.593e+02 4.441e+02 5.313e+02 1.329e+03, threshold=8.882e+02, percent-clipped=4.0
+2023-03-29 08:30:42,674 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65532.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 08:32:02,975 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0311, 3.8865, 4.2708, 3.9204, 3.6564, 4.1203, 3.9348, 4.3227],
+       device='cuda:3'), covar=tensor([0.0719, 0.0375, 0.0371, 0.0398, 0.1096, 0.0582, 0.0528, 0.0361],
+       device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0226, 0.0226, 0.0238, 0.0208, 0.0249, 0.0239, 0.0223],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 08:32:03,031 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8136, 2.2997, 3.8726, 3.3856, 3.8274, 3.9152, 3.6495, 3.6679],
+       device='cuda:3'), covar=tensor([0.0656, 0.1043, 0.0126, 0.0542, 0.0168, 0.0230, 0.0207, 0.0187],
+       device='cuda:3'), in_proj_covar=tensor([0.0102, 0.0105, 0.0090, 0.0153, 0.0087, 0.0100, 0.0091, 0.0088],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 08:32:21,959 INFO [train.py:892] (3/4) Epoch 36, batch 650, loss[loss=0.1905, simple_loss=0.2698, pruned_loss=0.05566, over 19803.00 frames. ], tot_loss[loss=0.1583, simple_loss=0.2379, pruned_loss=0.03933, over 3799152.69 frames. ], batch size: 51, lr: 4.35e-03, grad_scale: 16.0
+2023-03-29 08:33:02,661 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65593.0, num_to_drop=1, layers_to_drop={3}
+2023-03-29 08:34:14,087 INFO [train.py:892] (3/4) Epoch 36, batch 700, loss[loss=0.146, simple_loss=0.2206, pruned_loss=0.03565, over 19815.00 frames. ], tot_loss[loss=0.1569, simple_loss=0.2366, pruned_loss=0.03856, over 3832804.46 frames. ], batch size: 123, lr: 4.35e-03, grad_scale: 16.0
+2023-03-29 08:34:18,143 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.459e+02 3.698e+02 4.189e+02 4.947e+02 1.521e+03, threshold=8.379e+02, percent-clipped=3.0
+2023-03-29 08:34:31,284 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65632.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:34:44,888 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.91 vs. limit=5.0
+2023-03-29 08:34:52,689 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:36:12,644 INFO [train.py:892] (3/4) Epoch 36, batch 750, loss[loss=0.1361, simple_loss=0.2143, pruned_loss=0.02891, over 19813.00 frames. ], tot_loss[loss=0.157, simple_loss=0.2367, pruned_loss=0.03861, over 3858055.74 frames. ], batch size: 114, lr: 4.35e-03, grad_scale: 16.0
+2023-03-29 08:36:22,142 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:36:22,827 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0
+2023-03-29 08:37:34,772 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:38:07,215 INFO [train.py:892] (3/4) Epoch 36, batch 800, loss[loss=0.1453, simple_loss=0.2265, pruned_loss=0.03199, over 19879.00 frames. ], tot_loss[loss=0.1577, simple_loss=0.2377, pruned_loss=0.03881, over 3878619.71 frames. ], batch size: 97, lr: 4.34e-03, grad_scale: 16.0
+2023-03-29 08:38:11,741 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.627e+02 3.709e+02 4.446e+02 5.789e+02 1.138e+03, threshold=8.893e+02, percent-clipped=2.0
+2023-03-29 08:39:25,073 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:39:32,062 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0720, 2.4954, 3.0261, 3.2241, 3.7060, 4.1560, 4.0130, 4.0781],
+       device='cuda:3'), covar=tensor([0.0915, 0.1702, 0.1216, 0.0689, 0.0464, 0.0255, 0.0345, 0.0453],
+       device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0171, 0.0181, 0.0155, 0.0140, 0.0135, 0.0127, 0.0120],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 08:39:36,113 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8758, 2.9312, 2.9499, 2.4611, 3.0047, 2.5732, 2.9727, 2.9207],
+       device='cuda:3'), covar=tensor([0.0568, 0.0450, 0.0573, 0.0812, 0.0406, 0.0527, 0.0474, 0.0468],
+       device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0090, 0.0087, 0.0113, 0.0082, 0.0086, 0.0083, 0.0077],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 08:40:02,225 INFO [train.py:892] (3/4) Epoch 36, batch 850, loss[loss=0.1365, simple_loss=0.2139, pruned_loss=0.02952, over 19753.00 frames. ], tot_loss[loss=0.1576, simple_loss=0.238, pruned_loss=0.0386, over 3892779.77 frames. ], batch size: 102, lr: 4.34e-03, grad_scale: 16.0
+2023-03-29 08:41:36,442 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0
+2023-03-29 08:41:53,882 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9020, 2.9960, 3.1104, 3.0208, 2.8578, 2.9956, 2.8035, 3.1330],
+       device='cuda:3'), covar=tensor([0.0322, 0.0353, 0.0324, 0.0270, 0.0416, 0.0327, 0.0443, 0.0340],
+       device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0086, 0.0088, 0.0083, 0.0095, 0.0088, 0.0105, 0.0077],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 08:41:57,338 INFO [train.py:892] (3/4) Epoch 36, batch 900, loss[loss=0.1488, simple_loss=0.23, pruned_loss=0.03381, over 19766.00 frames. ], tot_loss[loss=0.1576, simple_loss=0.2381, pruned_loss=0.03853, over 3904594.55 frames. ], batch size: 217, lr: 4.34e-03, grad_scale: 16.0
+2023-03-29 08:42:01,188 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.659e+02 3.793e+02 4.465e+02 5.730e+02 1.109e+03, threshold=8.930e+02, percent-clipped=1.0
+2023-03-29 08:42:24,394 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65838.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 08:43:50,277 INFO [train.py:892] (3/4) Epoch 36, batch 950, loss[loss=0.1569, simple_loss=0.2302, pruned_loss=0.04182, over 19866.00 frames. ], tot_loss[loss=0.158, simple_loss=0.2385, pruned_loss=0.03872, over 3913709.39 frames. ], batch size: 136, lr: 4.34e-03, grad_scale: 16.0
+2023-03-29 08:44:19,523 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65888.0, num_to_drop=1, layers_to_drop={2}
+2023-03-29 08:44:45,765 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65899.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 08:45:40,598 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2538, 3.1289, 4.8043, 3.6050, 3.8087, 3.5983, 2.6268, 2.8595],
+       device='cuda:3'), covar=tensor([0.1005, 0.3390, 0.0457, 0.1110, 0.1817, 0.1628, 0.2787, 0.2753],
+       device='cuda:3'), in_proj_covar=tensor([0.0352, 0.0392, 0.0349, 0.0289, 0.0375, 0.0382, 0.0379, 0.0350],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 08:45:43,717 INFO [train.py:892] (3/4) Epoch 36, batch 1000, loss[loss=0.1439, simple_loss=0.2142, pruned_loss=0.03684, over 19862.00 frames. ], tot_loss[loss=0.1574, simple_loss=0.2377, pruned_loss=0.03856, over 3921276.92 frames. ], batch size: 122, lr: 4.34e-03, grad_scale: 16.0
+2023-03-29 08:45:47,946 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.017e+02 3.590e+02 4.101e+02 4.810e+02 1.037e+03, threshold=8.201e+02, percent-clipped=3.0
+2023-03-29 08:46:20,458 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:46:39,466 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0575, 4.2191, 4.2533, 4.1353, 4.0420, 4.1998, 3.7930, 3.8157],
+       device='cuda:3'), covar=tensor([0.0562, 0.0559, 0.0559, 0.0517, 0.0728, 0.0583, 0.0719, 0.1019],
+       device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0299, 0.0309, 0.0271, 0.0279, 0.0262, 0.0278, 0.0325],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 08:47:38,768 INFO [train.py:892] (3/4) Epoch 36, batch 1050, loss[loss=0.2076, simple_loss=0.3157, pruned_loss=0.04977, over 18943.00 frames. ], tot_loss[loss=0.1569, simple_loss=0.2372, pruned_loss=0.03834, over 3927023.85 frames. ], batch size: 514, lr: 4.34e-03, grad_scale: 16.0
+2023-03-29 08:48:10,767 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:48:56,528 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1318, 2.9692, 3.2543, 2.6017, 3.2520, 2.8585, 3.1438, 3.2707],
+       device='cuda:3'), covar=tensor([0.0617, 0.0519, 0.0482, 0.0780, 0.0410, 0.0500, 0.0491, 0.0373],
+       device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0091, 0.0087, 0.0114, 0.0083, 0.0086, 0.0084, 0.0078],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 08:49:06,310 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9185, 4.5795, 4.6375, 4.4095, 4.9045, 3.0426, 4.0532, 2.3542],
+       device='cuda:3'), covar=tensor([0.0200, 0.0201, 0.0161, 0.0195, 0.0147, 0.1084, 0.0809, 0.1592],
+       device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0149, 0.0115, 0.0137, 0.0121, 0.0137, 0.0145, 0.0129],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 08:49:35,979 INFO [train.py:892] (3/4) Epoch 36, batch 1100, loss[loss=0.1406, simple_loss=0.2149, pruned_loss=0.03315, over 19845.00 frames. ], tot_loss[loss=0.1573, simple_loss=0.2372, pruned_loss=0.03866, over 3933072.39 frames. ], batch size: 137, lr: 4.33e-03, grad_scale: 16.0
+2023-03-29 08:49:39,607 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.898e+02 3.611e+02 4.375e+02 5.188e+02 7.346e+02, threshold=8.750e+02, percent-clipped=0.0
+2023-03-29 08:51:23,877 INFO [train.py:892] (3/4) Epoch 36, batch 1150, loss[loss=0.1748, simple_loss=0.2584, pruned_loss=0.0456, over 19640.00 frames. ], tot_loss[loss=0.1575, simple_loss=0.2372, pruned_loss=0.03889, over 3936705.26 frames. ], batch size: 351, lr: 4.33e-03, grad_scale: 16.0
+2023-03-29 08:51:33,120 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.67 vs. limit=5.0
+2023-03-29 08:53:15,618 INFO [train.py:892] (3/4) Epoch 36, batch 1200, loss[loss=0.1436, simple_loss=0.2341, pruned_loss=0.02649, over 19654.00 frames. ], tot_loss[loss=0.1574, simple_loss=0.2374, pruned_loss=0.03872, over 3938746.66 frames. ], batch size: 66, lr: 4.33e-03, grad_scale: 16.0
+2023-03-29 08:53:19,997 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.230e+02 3.760e+02 4.454e+02 5.058e+02 1.051e+03, threshold=8.908e+02, percent-clipped=2.0
+2023-03-29 08:53:58,179 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:54:33,430 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66160.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 08:55:09,321 INFO [train.py:892] (3/4) Epoch 36, batch 1250, loss[loss=0.1551, simple_loss=0.2165, pruned_loss=0.04684, over 19753.00 frames. ], tot_loss[loss=0.1575, simple_loss=0.2371, pruned_loss=0.03901, over 3941253.78 frames. ], batch size: 129, lr: 4.33e-03, grad_scale: 16.0
+2023-03-29 08:55:38,391 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66188.0, num_to_drop=1, layers_to_drop={2}
+2023-03-29 08:55:51,144 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66194.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 08:56:16,400 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66205.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 08:56:53,166 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66221.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 08:57:03,492 INFO [train.py:892] (3/4) Epoch 36, batch 1300, loss[loss=0.213, simple_loss=0.3305, pruned_loss=0.04769, over 17908.00 frames. ], tot_loss[loss=0.1575, simple_loss=0.2373, pruned_loss=0.03881, over 3939249.00 frames. ], batch size: 633, lr: 4.33e-03, grad_scale: 16.0
+2023-03-29 08:57:07,570 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.794e+02 3.837e+02 4.506e+02 5.422e+02 1.103e+03, threshold=9.011e+02, percent-clipped=2.0
+2023-03-29 08:57:25,756 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66236.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 08:57:29,842 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5800, 4.3740, 4.4218, 4.1806, 4.6061, 3.1236, 3.9417, 2.2247],
+       device='cuda:3'), covar=tensor([0.0191, 0.0220, 0.0135, 0.0188, 0.0127, 0.0946, 0.0662, 0.1492],
+       device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0150, 0.0116, 0.0137, 0.0121, 0.0137, 0.0145, 0.0130],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 08:57:42,036 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2023, 3.5090, 3.7103, 4.1890, 2.9305, 3.4143, 2.5877, 2.6555],
+       device='cuda:3'), covar=tensor([0.0497, 0.1756, 0.0919, 0.0418, 0.1990, 0.0925, 0.1443, 0.1678],
+       device='cuda:3'), in_proj_covar=tensor([0.0250, 0.0330, 0.0252, 0.0207, 0.0251, 0.0214, 0.0224, 0.0219],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 08:58:16,700 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66258.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 08:58:30,435 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2711, 4.1210, 4.5275, 4.1409, 3.8694, 4.3647, 4.1896, 4.6068],
+       device='cuda:3'), covar=tensor([0.0686, 0.0337, 0.0345, 0.0356, 0.0957, 0.0513, 0.0463, 0.0308],
+       device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0225, 0.0224, 0.0236, 0.0206, 0.0247, 0.0237, 0.0221],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 08:58:57,030 INFO [train.py:892] (3/4) Epoch 36, batch 1350, loss[loss=0.1682, simple_loss=0.2357, pruned_loss=0.05035, over 19915.00 frames. ], tot_loss[loss=0.1568, simple_loss=0.2369, pruned_loss=0.03836, over 3942370.53 frames. ], batch size: 45, lr: 4.33e-03, grad_scale: 16.0
+2023-03-29 09:00:26,816 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.0199, 1.9990, 2.0245, 2.1222, 2.0299, 2.1395, 1.9923, 2.0741],
+       device='cuda:3'), covar=tensor([0.0442, 0.0355, 0.0412, 0.0320, 0.0510, 0.0329, 0.0495, 0.0383],
+       device='cuda:3'), in_proj_covar=tensor([0.0091, 0.0086, 0.0089, 0.0082, 0.0095, 0.0088, 0.0105, 0.0078],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 09:00:37,095 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 09:00:49,681 INFO [train.py:892] (3/4) Epoch 36, batch 1400, loss[loss=0.1499, simple_loss=0.2281, pruned_loss=0.03584, over 19787.00 frames. ], tot_loss[loss=0.1569, simple_loss=0.2368, pruned_loss=0.03853, over 3944943.18 frames. ], batch size: 211, lr: 4.32e-03, grad_scale: 16.0
+2023-03-29 09:00:54,075 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.446e+02 3.517e+02 4.188e+02 4.908e+02 8.356e+02, threshold=8.377e+02, percent-clipped=0.0
+2023-03-29 09:01:40,147 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6778, 4.4665, 4.4488, 4.7056, 4.4132, 4.8263, 4.7664, 5.0115],
+       device='cuda:3'), covar=tensor([0.0640, 0.0371, 0.0530, 0.0406, 0.0711, 0.0473, 0.0441, 0.0260],
+       device='cuda:3'), in_proj_covar=tensor([0.0155, 0.0180, 0.0203, 0.0179, 0.0177, 0.0161, 0.0154, 0.0199],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 09:01:47,828 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3964, 3.1833, 3.4569, 2.6010, 3.4858, 2.9779, 3.3259, 3.5047],
+       device='cuda:3'), covar=tensor([0.0568, 0.0439, 0.0484, 0.0849, 0.0362, 0.0508, 0.0397, 0.0318],
+       device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0090, 0.0086, 0.0113, 0.0083, 0.0086, 0.0083, 0.0077],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 09:02:27,087 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.95 vs. limit=5.0
+2023-03-29 09:02:43,653 INFO [train.py:892] (3/4) Epoch 36, batch 1450, loss[loss=0.1755, simple_loss=0.2549, pruned_loss=0.04803, over 19738.00 frames. ], tot_loss[loss=0.1568, simple_loss=0.2366, pruned_loss=0.03847, over 3947392.79 frames. ], batch size: 221, lr: 4.32e-03, grad_scale: 16.0
+2023-03-29 09:04:34,697 INFO [train.py:892] (3/4) Epoch 36, batch 1500, loss[loss=0.1693, simple_loss=0.2493, pruned_loss=0.04463, over 19414.00 frames. ], tot_loss[loss=0.1566, simple_loss=0.2363, pruned_loss=0.03842, over 3946859.79 frames. ], batch size: 40, lr: 4.32e-03, grad_scale: 16.0
+2023-03-29 09:04:40,372 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.359e+02 3.684e+02 4.435e+02 5.309e+02 1.081e+03, threshold=8.870e+02, percent-clipped=2.0
+2023-03-29 09:06:32,052 INFO [train.py:892] (3/4) Epoch 36, batch 1550, loss[loss=0.1262, simple_loss=0.2133, pruned_loss=0.01954, over 19864.00 frames. ], tot_loss[loss=0.1562, simple_loss=0.2364, pruned_loss=0.03798, over 3948502.10 frames. ], batch size: 104, lr: 4.32e-03, grad_scale: 16.0
+2023-03-29 09:07:11,669 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66494.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 09:07:25,650 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66500.0, num_to_drop=1, layers_to_drop={2}
+2023-03-29 09:08:01,315 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66516.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 09:08:11,874 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66521.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 09:08:20,347 INFO [train.py:892] (3/4) Epoch 36, batch 1600, loss[loss=0.1499, simple_loss=0.2303, pruned_loss=0.03474, over 19846.00 frames. ], tot_loss[loss=0.1567, simple_loss=0.2366, pruned_loss=0.0384, over 3948351.65 frames. ], batch size: 115, lr: 4.32e-03, grad_scale: 16.0
+2023-03-29 09:08:24,089 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.296e+02 3.591e+02 4.311e+02 5.202e+02 1.053e+03, threshold=8.622e+02, percent-clipped=1.0
+2023-03-29 09:08:57,609 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66542.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 09:08:57,975 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5256, 3.2446, 3.6342, 3.1109, 3.7707, 3.8045, 4.4789, 4.8556],
+       device='cuda:3'), covar=tensor([0.0598, 0.1608, 0.1500, 0.2219, 0.1680, 0.1357, 0.0536, 0.0587],
+       device='cuda:3'), in_proj_covar=tensor([0.0260, 0.0246, 0.0274, 0.0259, 0.0306, 0.0264, 0.0239, 0.0265],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 09:09:38,487 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6357, 3.9250, 4.1848, 4.7164, 2.9541, 3.2913, 2.7946, 2.7060],
+       device='cuda:3'), covar=tensor([0.0487, 0.2302, 0.0870, 0.0402, 0.2431, 0.1315, 0.1513, 0.1939],
+       device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0330, 0.0251, 0.0207, 0.0251, 0.0213, 0.0223, 0.0219],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 09:10:14,472 INFO [train.py:892] (3/4) Epoch 36, batch 1650, loss[loss=0.1347, simple_loss=0.2152, pruned_loss=0.02708, over 19897.00 frames. ], tot_loss[loss=0.1562, simple_loss=0.2361, pruned_loss=0.0381, over 3948175.38 frames. ], batch size: 113, lr: 4.32e-03, grad_scale: 16.0
+2023-03-29 09:10:29,328 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66582.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 09:11:42,685 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66614.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 09:12:09,539 INFO [train.py:892] (3/4) Epoch 36, batch 1700, loss[loss=0.1653, simple_loss=0.249, pruned_loss=0.04077, over 19737.00 frames. ], tot_loss[loss=0.1568, simple_loss=0.237, pruned_loss=0.03825, over 3947139.48 frames. ], batch size: 259, lr: 4.31e-03, grad_scale: 16.0
+2023-03-29 09:12:13,637 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.577e+02 3.609e+02 4.438e+02 5.504e+02 8.717e+02, threshold=8.877e+02, percent-clipped=1.0
+2023-03-29 09:13:19,146 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4966, 3.4807, 2.2584, 4.1411, 3.7843, 4.0630, 4.1890, 3.2792],
+       device='cuda:3'), covar=tensor([0.0557, 0.0639, 0.1489, 0.0613, 0.0554, 0.0415, 0.0580, 0.0790],
+       device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0148, 0.0146, 0.0158, 0.0137, 0.0141, 0.0152, 0.0149],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 09:13:48,666 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7670, 4.5395, 5.1059, 4.6118, 4.2343, 4.8336, 4.7141, 5.2616],
+       device='cuda:3'), covar=tensor([0.0804, 0.0380, 0.0361, 0.0389, 0.0823, 0.0485, 0.0457, 0.0287],
+       device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0226, 0.0224, 0.0237, 0.0207, 0.0246, 0.0238, 0.0220],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 09:13:57,362 INFO [train.py:892] (3/4) Epoch 36, batch 1750, loss[loss=0.1433, simple_loss=0.2167, pruned_loss=0.03493, over 19806.00 frames. ], tot_loss[loss=0.1578, simple_loss=0.238, pruned_loss=0.03875, over 3946507.83 frames. ], batch size: 114, lr: 4.31e-03, grad_scale: 16.0
+2023-03-29 09:15:32,718 INFO [train.py:892] (3/4) Epoch 36, batch 1800, loss[loss=0.1687, simple_loss=0.2516, pruned_loss=0.04289, over 19715.00 frames. ], tot_loss[loss=0.1579, simple_loss=0.2383, pruned_loss=0.03879, over 3947229.70 frames. ], batch size: 54, lr: 4.31e-03, grad_scale: 16.0
+2023-03-29 09:15:36,377 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.260e+02 3.634e+02 4.363e+02 5.062e+02 8.323e+02, threshold=8.726e+02, percent-clipped=0.0
+2023-03-29 09:15:38,839 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4408, 5.6561, 5.7305, 5.5839, 5.4169, 5.7075, 5.0906, 5.1035],
+       device='cuda:3'), covar=tensor([0.0409, 0.0450, 0.0464, 0.0409, 0.0532, 0.0419, 0.0633, 0.0923],
+       device='cuda:3'), in_proj_covar=tensor([0.0281, 0.0296, 0.0306, 0.0269, 0.0279, 0.0259, 0.0274, 0.0322],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 09:15:42,759 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66731.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 09:17:03,420 INFO [train.py:892] (3/4) Epoch 36, batch 1850, loss[loss=0.1621, simple_loss=0.2518, pruned_loss=0.03619, over 19577.00 frames. ], tot_loss[loss=0.1579, simple_loss=0.2392, pruned_loss=0.0383, over 3947537.64 frames. ], batch size: 53, lr: 4.31e-03, grad_scale: 16.0
+2023-03-29 09:18:08,596 INFO [train.py:892] (3/4) Epoch 37, batch 0, loss[loss=0.1364, simple_loss=0.2152, pruned_loss=0.02885, over 19744.00 frames. ], tot_loss[loss=0.1364, simple_loss=0.2152, pruned_loss=0.02885, over 19744.00 frames. ], batch size: 106, lr: 4.25e-03, grad_scale: 16.0
+2023-03-29 09:18:08,596 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-29 09:18:41,824 INFO [train.py:926] (3/4) Epoch 37, validation: loss=0.1834, simple_loss=0.2492, pruned_loss=0.05881, over 2883724.00 frames. 
+2023-03-29 09:18:41,826 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-29 09:18:58,827 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0822, 2.9484, 3.1011, 2.5445, 3.2252, 2.7377, 3.0762, 3.0864],
+       device='cuda:3'), covar=tensor([0.0534, 0.0490, 0.0495, 0.0775, 0.0347, 0.0461, 0.0457, 0.0402],
+       device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0091, 0.0087, 0.0114, 0.0083, 0.0086, 0.0084, 0.0078],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 09:19:11,070 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66792.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 09:19:27,990 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66800.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 09:20:04,086 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66816.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 09:20:31,170 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.683e+02 3.599e+02 4.155e+02 4.773e+02 9.045e+02, threshold=8.309e+02, percent-clipped=1.0
+2023-03-29 09:20:39,157 INFO [train.py:892] (3/4) Epoch 37, batch 50, loss[loss=0.1395, simple_loss=0.215, pruned_loss=0.03202, over 19809.00 frames. ], tot_loss[loss=0.1517, simple_loss=0.2299, pruned_loss=0.03668, over 892325.68 frames. ], batch size: 82, lr: 4.25e-03, grad_scale: 16.0
+2023-03-29 09:21:04,045 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9971, 3.8640, 4.2645, 3.8793, 3.6566, 4.1257, 3.9454, 4.3314],
+       device='cuda:3'), covar=tensor([0.0750, 0.0383, 0.0361, 0.0394, 0.1049, 0.0521, 0.0502, 0.0348],
+       device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0228, 0.0226, 0.0239, 0.0209, 0.0247, 0.0239, 0.0222],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 09:21:15,310 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66848.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 09:21:51,473 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66864.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 09:22:19,788 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66877.0, num_to_drop=1, layers_to_drop={3}
+2023-03-29 09:22:30,256 INFO [train.py:892] (3/4) Epoch 37, batch 100, loss[loss=0.1394, simple_loss=0.215, pruned_loss=0.0319, over 19743.00 frames. ], tot_loss[loss=0.1581, simple_loss=0.237, pruned_loss=0.03965, over 1568902.97 frames. ], batch size: 134, lr: 4.25e-03, grad_scale: 16.0
+2023-03-29 09:23:46,011 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66914.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 09:24:15,777 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.134e+02 3.769e+02 4.346e+02 5.384e+02 1.182e+03, threshold=8.692e+02, percent-clipped=4.0
+2023-03-29 09:24:22,077 INFO [train.py:892] (3/4) Epoch 37, batch 150, loss[loss=0.1551, simple_loss=0.2357, pruned_loss=0.03729, over 19708.00 frames. ], tot_loss[loss=0.1588, simple_loss=0.2387, pruned_loss=0.03946, over 2094471.66 frames. ], batch size: 81, lr: 4.25e-03, grad_scale: 16.0
+2023-03-29 09:24:46,043 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.75 vs. limit=5.0
+2023-03-29 09:25:24,983 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 09:25:27,323 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0869, 2.6834, 3.2941, 3.3205, 3.8695, 4.3158, 4.0958, 4.1515],
+       device='cuda:3'), covar=tensor([0.0933, 0.1629, 0.1179, 0.0671, 0.0367, 0.0235, 0.0417, 0.0462],
+       device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0168, 0.0178, 0.0153, 0.0138, 0.0134, 0.0127, 0.0119],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 09:25:35,522 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66962.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 09:26:13,811 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.3248, 5.7257, 5.8900, 5.6610, 5.5615, 5.5280, 5.5368, 5.4241],
+       device='cuda:3'), covar=tensor([0.1511, 0.1102, 0.0793, 0.1028, 0.0701, 0.0683, 0.1854, 0.1838],
+       device='cuda:3'), in_proj_covar=tensor([0.0301, 0.0339, 0.0374, 0.0304, 0.0280, 0.0287, 0.0368, 0.0392],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 09:26:19,096 INFO [train.py:892] (3/4) Epoch 37, batch 200, loss[loss=0.1453, simple_loss=0.2325, pruned_loss=0.02905, over 19838.00 frames. ], tot_loss[loss=0.1574, simple_loss=0.2374, pruned_loss=0.03864, over 2506332.59 frames. ], batch size: 90, lr: 4.24e-03, grad_scale: 16.0
+2023-03-29 09:27:15,553 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0
+2023-03-29 09:27:46,088 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67018.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 09:28:07,317 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.541e+02 3.516e+02 4.008e+02 4.634e+02 9.207e+02, threshold=8.015e+02, percent-clipped=1.0
+2023-03-29 09:28:13,945 INFO [train.py:892] (3/4) Epoch 37, batch 250, loss[loss=0.2111, simple_loss=0.2909, pruned_loss=0.06567, over 19797.00 frames. ], tot_loss[loss=0.1552, simple_loss=0.2357, pruned_loss=0.03734, over 2825182.41 frames. ], batch size: 65, lr: 4.24e-03, grad_scale: 16.0
+2023-03-29 09:30:07,631 INFO [train.py:892] (3/4) Epoch 37, batch 300, loss[loss=0.1431, simple_loss=0.2238, pruned_loss=0.03119, over 19662.00 frames. ], tot_loss[loss=0.158, simple_loss=0.238, pruned_loss=0.03901, over 3074357.38 frames. ], batch size: 43, lr: 4.24e-03, grad_scale: 16.0
+2023-03-29 09:30:15,044 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67083.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 09:30:23,109 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67087.0, num_to_drop=1, layers_to_drop={2}
+2023-03-29 09:31:58,057 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.283e+02 3.982e+02 4.417e+02 5.136e+02 1.288e+03, threshold=8.835e+02, percent-clipped=3.0
+2023-03-29 09:32:05,873 INFO [train.py:892] (3/4) Epoch 37, batch 350, loss[loss=0.1319, simple_loss=0.2183, pruned_loss=0.02269, over 19817.00 frames. ], tot_loss[loss=0.1584, simple_loss=0.2385, pruned_loss=0.0391, over 3268502.20 frames. ], batch size: 96, lr: 4.24e-03, grad_scale: 16.0
+2023-03-29 09:32:34,404 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 09:32:34,532 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0334, 2.8911, 3.2511, 2.7707, 3.3598, 3.2538, 3.9048, 4.3089],
+       device='cuda:3'), covar=tensor([0.0646, 0.1767, 0.1584, 0.2310, 0.1703, 0.1616, 0.0673, 0.0597],
+       device='cuda:3'), in_proj_covar=tensor([0.0257, 0.0243, 0.0271, 0.0257, 0.0303, 0.0262, 0.0237, 0.0263],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 09:32:44,801 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.0621, 2.0186, 2.1052, 2.1306, 2.0533, 2.1438, 2.0327, 2.1557],
+       device='cuda:3'), covar=tensor([0.0413, 0.0366, 0.0359, 0.0326, 0.0519, 0.0358, 0.0481, 0.0346],
+       device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0086, 0.0089, 0.0083, 0.0096, 0.0089, 0.0105, 0.0078],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 09:33:49,821 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67177.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 09:33:57,477 INFO [train.py:892] (3/4) Epoch 37, batch 400, loss[loss=0.142, simple_loss=0.2199, pruned_loss=0.03204, over 19807.00 frames. ], tot_loss[loss=0.1567, simple_loss=0.2367, pruned_loss=0.03836, over 3420348.55 frames. ], batch size: 65, lr: 4.24e-03, grad_scale: 16.0
+2023-03-29 09:34:42,374 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7640, 3.4881, 3.6392, 3.7977, 3.5981, 3.7417, 3.8432, 4.0283],
+       device='cuda:3'), covar=tensor([0.0696, 0.0498, 0.0553, 0.0458, 0.0770, 0.0605, 0.0468, 0.0339],
+       device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0183, 0.0206, 0.0182, 0.0181, 0.0164, 0.0156, 0.0203],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 09:34:51,069 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6067, 2.9119, 3.1232, 3.5217, 2.4214, 3.0765, 2.5111, 2.3954],
+       device='cuda:3'), covar=tensor([0.0652, 0.1594, 0.1109, 0.0534, 0.2182, 0.0890, 0.1348, 0.1646],
+       device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0328, 0.0250, 0.0207, 0.0250, 0.0212, 0.0223, 0.0219],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 09:35:34,655 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67225.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 09:35:41,244 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.424e+02 3.741e+02 4.233e+02 5.097e+02 1.151e+03, threshold=8.466e+02, percent-clipped=1.0
+2023-03-29 09:35:49,520 INFO [train.py:892] (3/4) Epoch 37, batch 450, loss[loss=0.1297, simple_loss=0.2142, pruned_loss=0.02263, over 19827.00 frames. ], tot_loss[loss=0.1565, simple_loss=0.2366, pruned_loss=0.03816, over 3537587.85 frames. ], batch size: 103, lr: 4.24e-03, grad_scale: 16.0
+2023-03-29 09:36:17,247 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3537, 3.7547, 4.0357, 4.4312, 2.8981, 3.5077, 2.7830, 2.7742],
+       device='cuda:3'), covar=tensor([0.0467, 0.1569, 0.0719, 0.0355, 0.1905, 0.0879, 0.1262, 0.1585],
+       device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0328, 0.0250, 0.0207, 0.0250, 0.0212, 0.0223, 0.0218],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 09:37:36,850 INFO [train.py:892] (3/4) Epoch 37, batch 500, loss[loss=0.1379, simple_loss=0.2211, pruned_loss=0.02738, over 19802.00 frames. ], tot_loss[loss=0.1565, simple_loss=0.2365, pruned_loss=0.03825, over 3627102.86 frames. ], batch size: 114, lr: 4.23e-03, grad_scale: 16.0
+2023-03-29 09:38:52,253 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67313.0, num_to_drop=1, layers_to_drop={2}
+2023-03-29 09:39:24,432 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.430e+02 3.820e+02 4.510e+02 5.565e+02 1.076e+03, threshold=9.020e+02, percent-clipped=2.0
+2023-03-29 09:39:30,777 INFO [train.py:892] (3/4) Epoch 37, batch 550, loss[loss=0.1407, simple_loss=0.2164, pruned_loss=0.03249, over 19947.00 frames. ], tot_loss[loss=0.1565, simple_loss=0.2367, pruned_loss=0.03814, over 3698241.29 frames. ], batch size: 46, lr: 4.23e-03, grad_scale: 16.0
+2023-03-29 09:41:25,508 INFO [train.py:892] (3/4) Epoch 37, batch 600, loss[loss=0.1318, simple_loss=0.2117, pruned_loss=0.02594, over 19733.00 frames. ], tot_loss[loss=0.1568, simple_loss=0.2366, pruned_loss=0.03843, over 3754536.05 frames. ], batch size: 118, lr: 4.23e-03, grad_scale: 32.0
+2023-03-29 09:41:39,475 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0
+2023-03-29 09:41:43,214 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67387.0, num_to_drop=1, layers_to_drop={2}
+2023-03-29 09:42:48,667 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9856, 4.0920, 2.4383, 4.2414, 4.4069, 1.9630, 3.6494, 3.2556],
+       device='cuda:3'), covar=tensor([0.0675, 0.0760, 0.2707, 0.0803, 0.0559, 0.2756, 0.1074, 0.0935],
+       device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0267, 0.0238, 0.0286, 0.0265, 0.0210, 0.0248, 0.0207],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 09:43:17,839 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.189e+02 3.534e+02 4.252e+02 5.287e+02 1.120e+03, threshold=8.504e+02, percent-clipped=1.0
+2023-03-29 09:43:24,004 INFO [train.py:892] (3/4) Epoch 37, batch 650, loss[loss=0.1455, simple_loss=0.2302, pruned_loss=0.03043, over 19743.00 frames. ], tot_loss[loss=0.1568, simple_loss=0.2369, pruned_loss=0.03839, over 3796148.03 frames. ], batch size: 89, lr: 4.23e-03, grad_scale: 32.0
+2023-03-29 09:43:33,733 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67435.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 09:43:42,893 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67439.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 09:44:16,911 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7156, 4.4526, 4.5394, 4.2653, 4.7256, 3.2163, 3.8963, 2.2572],
+       device='cuda:3'), covar=tensor([0.0179, 0.0241, 0.0159, 0.0208, 0.0138, 0.0931, 0.0756, 0.1567],
+       device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0149, 0.0115, 0.0136, 0.0120, 0.0136, 0.0144, 0.0129],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 09:45:16,659 INFO [train.py:892] (3/4) Epoch 37, batch 700, loss[loss=0.1597, simple_loss=0.2343, pruned_loss=0.04256, over 19748.00 frames. ], tot_loss[loss=0.1571, simple_loss=0.2376, pruned_loss=0.03836, over 3831116.39 frames. ], batch size: 259, lr: 4.23e-03, grad_scale: 32.0
+2023-03-29 09:45:52,573 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.5566, 5.9207, 5.9439, 5.8223, 5.6331, 5.9395, 5.3096, 5.2823],
+       device='cuda:3'), covar=tensor([0.0395, 0.0429, 0.0454, 0.0384, 0.0544, 0.0445, 0.0701, 0.1056],
+       device='cuda:3'), in_proj_covar=tensor([0.0278, 0.0296, 0.0304, 0.0266, 0.0278, 0.0258, 0.0273, 0.0319],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 09:46:31,455 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67513.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 09:46:36,266 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0
+2023-03-29 09:47:00,447 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5169, 3.6489, 2.2655, 3.7278, 3.8234, 1.8163, 3.2235, 3.0407],
+       device='cuda:3'), covar=tensor([0.0854, 0.0826, 0.2756, 0.0872, 0.0668, 0.2668, 0.1141, 0.0949],
+       device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0266, 0.0237, 0.0285, 0.0264, 0.0209, 0.0247, 0.0206],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 09:47:05,445 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.319e+02 3.696e+02 4.491e+02 5.364e+02 8.859e+02, threshold=8.982e+02, percent-clipped=1.0
+2023-03-29 09:47:12,720 INFO [train.py:892] (3/4) Epoch 37, batch 750, loss[loss=0.1564, simple_loss=0.2312, pruned_loss=0.04083, over 19751.00 frames. ], tot_loss[loss=0.1564, simple_loss=0.2369, pruned_loss=0.03789, over 3857186.20 frames. ], batch size: 205, lr: 4.23e-03, grad_scale: 32.0
+2023-03-29 09:47:45,175 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.90 vs. limit=5.0
+2023-03-29 09:48:32,735 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5996, 2.9206, 2.6357, 2.1537, 2.5958, 2.7709, 2.8722, 2.8370],
+       device='cuda:3'), covar=tensor([0.0368, 0.0305, 0.0316, 0.0552, 0.0408, 0.0364, 0.0266, 0.0248],
+       device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0103, 0.0104, 0.0105, 0.0109, 0.0092, 0.0093, 0.0092],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 09:48:51,666 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67574.0, num_to_drop=1, layers_to_drop={3}
+2023-03-29 09:49:07,695 INFO [train.py:892] (3/4) Epoch 37, batch 800, loss[loss=0.1538, simple_loss=0.2324, pruned_loss=0.03753, over 19795.00 frames. ], tot_loss[loss=0.1555, simple_loss=0.2364, pruned_loss=0.03732, over 3878035.76 frames. ], batch size: 185, lr: 4.22e-03, grad_scale: 32.0
+2023-03-29 09:49:18,244 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9997, 3.7190, 3.7984, 3.9974, 3.7870, 3.9473, 4.0908, 4.2563],
+       device='cuda:3'), covar=tensor([0.0693, 0.0506, 0.0581, 0.0441, 0.0760, 0.0591, 0.0430, 0.0343],
+       device='cuda:3'), in_proj_covar=tensor([0.0159, 0.0184, 0.0208, 0.0183, 0.0182, 0.0165, 0.0157, 0.0204],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 09:50:18,724 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 09:50:25,634 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0
+2023-03-29 09:50:51,537 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.125e+02 3.592e+02 4.129e+02 4.799e+02 7.012e+02, threshold=8.258e+02, percent-clipped=0.0
+2023-03-29 09:50:55,515 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.22 vs. limit=5.0
+2023-03-29 09:50:58,193 INFO [train.py:892] (3/4) Epoch 37, batch 850, loss[loss=0.1665, simple_loss=0.2443, pruned_loss=0.04431, over 19795.00 frames. ], tot_loss[loss=0.1549, simple_loss=0.2357, pruned_loss=0.03704, over 3894486.11 frames. ], batch size: 211, lr: 4.22e-03, grad_scale: 32.0
+2023-03-29 09:51:45,832 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3266, 4.5166, 2.7137, 4.7699, 5.0165, 2.3029, 4.3280, 3.6796],
+       device='cuda:3'), covar=tensor([0.0727, 0.0734, 0.2599, 0.0665, 0.0472, 0.2685, 0.0864, 0.0897],
+       device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0266, 0.0237, 0.0286, 0.0265, 0.0209, 0.0247, 0.0207],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 09:52:10,917 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 09:52:54,012 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67680.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 09:52:54,937 INFO [train.py:892] (3/4) Epoch 37, batch 900, loss[loss=0.1787, simple_loss=0.2552, pruned_loss=0.0511, over 19657.00 frames. ], tot_loss[loss=0.1561, simple_loss=0.2365, pruned_loss=0.03785, over 3905566.12 frames. ], batch size: 343, lr: 4.22e-03, grad_scale: 16.0
+2023-03-29 09:53:21,618 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 09:54:45,460 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.385e+02 3.765e+02 4.292e+02 5.023e+02 7.542e+02, threshold=8.583e+02, percent-clipped=0.0
+2023-03-29 09:54:49,593 INFO [train.py:892] (3/4) Epoch 37, batch 950, loss[loss=0.2383, simple_loss=0.3165, pruned_loss=0.08003, over 19422.00 frames. ], tot_loss[loss=0.1562, simple_loss=0.2366, pruned_loss=0.03791, over 3914427.60 frames. ], batch size: 412, lr: 4.22e-03, grad_scale: 16.0
+2023-03-29 09:55:09,851 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67739.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 09:55:15,601 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67741.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 09:55:41,698 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67753.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 09:56:23,548 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8146, 3.4668, 3.7563, 2.7996, 4.0473, 3.1975, 3.5920, 3.8340],
+       device='cuda:3'), covar=tensor([0.0667, 0.0441, 0.0546, 0.0843, 0.0338, 0.0389, 0.0493, 0.0395],
+       device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0091, 0.0088, 0.0115, 0.0084, 0.0087, 0.0085, 0.0078],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 09:56:41,366 INFO [train.py:892] (3/4) Epoch 37, batch 1000, loss[loss=0.1715, simple_loss=0.2385, pruned_loss=0.0523, over 19815.00 frames. ], tot_loss[loss=0.1562, simple_loss=0.2366, pruned_loss=0.03792, over 3922250.44 frames. ], batch size: 202, lr: 4.22e-03, grad_scale: 16.0
+2023-03-29 09:56:55,006 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67787.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 09:58:28,697 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.309e+02 3.606e+02 4.244e+02 4.990e+02 9.256e+02, threshold=8.487e+02, percent-clipped=3.0
+2023-03-29 09:58:33,851 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5395, 2.7689, 4.0297, 3.1162, 3.3728, 3.1196, 2.4049, 2.5368],
+       device='cuda:3'), covar=tensor([0.1228, 0.3064, 0.0592, 0.1121, 0.1671, 0.1620, 0.2578, 0.2709],
+       device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0393, 0.0352, 0.0290, 0.0377, 0.0386, 0.0381, 0.0352],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 09:58:34,568 INFO [train.py:892] (3/4) Epoch 37, batch 1050, loss[loss=0.1573, simple_loss=0.221, pruned_loss=0.04685, over 19800.00 frames. ], tot_loss[loss=0.1562, simple_loss=0.2366, pruned_loss=0.03786, over 3927325.79 frames. ], batch size: 149, lr: 4.22e-03, grad_scale: 16.0
+2023-03-29 09:58:42,241 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7691, 3.4950, 3.7567, 3.0112, 4.0633, 3.2647, 3.5733, 3.9557],
+       device='cuda:3'), covar=tensor([0.0701, 0.0411, 0.0750, 0.0718, 0.0390, 0.0402, 0.0510, 0.0339],
+       device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0091, 0.0087, 0.0114, 0.0084, 0.0086, 0.0084, 0.0078],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 09:59:14,074 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0777, 2.7082, 3.0855, 3.2628, 3.8221, 4.3518, 4.1555, 4.2200],
+       device='cuda:3'), covar=tensor([0.0944, 0.1550, 0.1332, 0.0689, 0.0366, 0.0216, 0.0297, 0.0464],
+       device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0171, 0.0181, 0.0155, 0.0140, 0.0136, 0.0129, 0.0120],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 10:00:01,293 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67869.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 10:00:25,320 INFO [train.py:892] (3/4) Epoch 37, batch 1100, loss[loss=0.1419, simple_loss=0.2123, pruned_loss=0.03572, over 19789.00 frames. ], tot_loss[loss=0.1556, simple_loss=0.2361, pruned_loss=0.03757, over 3932066.69 frames. ], batch size: 172, lr: 4.22e-03, grad_scale: 16.0
+2023-03-29 10:01:30,583 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.2674, 2.3335, 2.3179, 1.9812, 2.4380, 2.0179, 2.4167, 2.3405],
+       device='cuda:3'), covar=tensor([0.0556, 0.0541, 0.0553, 0.0962, 0.0447, 0.0562, 0.0492, 0.0434],
+       device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0091, 0.0088, 0.0114, 0.0084, 0.0087, 0.0085, 0.0078],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 10:01:45,633 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0
+2023-03-29 10:02:13,752 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.252e+02 3.748e+02 4.417e+02 5.113e+02 1.132e+03, threshold=8.834e+02, percent-clipped=3.0
+2023-03-29 10:02:18,098 INFO [train.py:892] (3/4) Epoch 37, batch 1150, loss[loss=0.1549, simple_loss=0.2438, pruned_loss=0.03302, over 19726.00 frames. ], tot_loss[loss=0.1562, simple_loss=0.2365, pruned_loss=0.03801, over 3934696.65 frames. ], batch size: 52, lr: 4.21e-03, grad_scale: 16.0
+2023-03-29 10:04:11,718 INFO [train.py:892] (3/4) Epoch 37, batch 1200, loss[loss=0.1594, simple_loss=0.2321, pruned_loss=0.04338, over 19821.00 frames. ], tot_loss[loss=0.1571, simple_loss=0.2373, pruned_loss=0.03846, over 3938834.39 frames. ], batch size: 204, lr: 4.21e-03, grad_scale: 16.0
+2023-03-29 10:04:12,835 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:06:05,471 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.592e+02 3.677e+02 4.252e+02 5.217e+02 1.129e+03, threshold=8.505e+02, percent-clipped=2.0
+2023-03-29 10:06:09,669 INFO [train.py:892] (3/4) Epoch 37, batch 1250, loss[loss=0.1708, simple_loss=0.2511, pruned_loss=0.04523, over 19682.00 frames. ], tot_loss[loss=0.1572, simple_loss=0.2372, pruned_loss=0.03859, over 3941439.54 frames. ], batch size: 56, lr: 4.21e-03, grad_scale: 16.0
+2023-03-29 10:06:20,208 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68036.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:06:35,664 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68042.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:06:51,024 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68048.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 10:08:04,302 INFO [train.py:892] (3/4) Epoch 37, batch 1300, loss[loss=0.1418, simple_loss=0.2217, pruned_loss=0.03092, over 19833.00 frames. ], tot_loss[loss=0.1574, simple_loss=0.2372, pruned_loss=0.0388, over 3941620.11 frames. ], batch size: 76, lr: 4.21e-03, grad_scale: 16.0
+2023-03-29 10:08:18,921 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68087.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 10:09:54,436 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.204e+02 3.685e+02 4.439e+02 5.105e+02 9.466e+02, threshold=8.879e+02, percent-clipped=1.0
+2023-03-29 10:09:58,214 INFO [train.py:892] (3/4) Epoch 37, batch 1350, loss[loss=0.1452, simple_loss=0.2251, pruned_loss=0.03263, over 19801.00 frames. ], tot_loss[loss=0.157, simple_loss=0.2372, pruned_loss=0.03838, over 3942333.31 frames. ], batch size: 107, lr: 4.21e-03, grad_scale: 16.0
+2023-03-29 10:10:07,746 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.47 vs. limit=5.0
+2023-03-29 10:10:35,467 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68148.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 10:11:23,992 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68169.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:11:48,282 INFO [train.py:892] (3/4) Epoch 37, batch 1400, loss[loss=0.1554, simple_loss=0.2395, pruned_loss=0.03572, over 19696.00 frames. ], tot_loss[loss=0.1564, simple_loss=0.2363, pruned_loss=0.03825, over 3945236.73 frames. ], batch size: 265, lr: 4.21e-03, grad_scale: 16.0
+2023-03-29 10:13:09,715 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68217.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:13:19,162 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5068, 3.6591, 2.1691, 3.7404, 3.8615, 1.7691, 3.2118, 2.9319],
+       device='cuda:3'), covar=tensor([0.0818, 0.0846, 0.2879, 0.0897, 0.0616, 0.2798, 0.1156, 0.0988],
+       device='cuda:3'), in_proj_covar=tensor([0.0238, 0.0261, 0.0233, 0.0280, 0.0260, 0.0205, 0.0241, 0.0202],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 10:13:35,439 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.616e+02 3.628e+02 4.259e+02 5.104e+02 1.150e+03, threshold=8.519e+02, percent-clipped=2.0
+2023-03-29 10:13:41,096 INFO [train.py:892] (3/4) Epoch 37, batch 1450, loss[loss=0.1832, simple_loss=0.2676, pruned_loss=0.04937, over 19656.00 frames. ], tot_loss[loss=0.1566, simple_loss=0.2367, pruned_loss=0.03821, over 3945923.66 frames. ], batch size: 79, lr: 4.20e-03, grad_scale: 16.0
+2023-03-29 10:14:36,683 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0
+2023-03-29 10:15:37,218 INFO [train.py:892] (3/4) Epoch 37, batch 1500, loss[loss=0.1689, simple_loss=0.2508, pruned_loss=0.04348, over 19525.00 frames. ], tot_loss[loss=0.1567, simple_loss=0.2372, pruned_loss=0.03809, over 3943745.46 frames. ], batch size: 54, lr: 4.20e-03, grad_scale: 16.0
+2023-03-29 10:16:42,976 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6514, 4.7344, 2.8325, 5.0325, 5.1696, 2.2785, 4.4206, 3.7462],
+       device='cuda:3'), covar=tensor([0.0600, 0.0787, 0.2606, 0.0626, 0.0430, 0.2759, 0.0865, 0.0840],
+       device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0263, 0.0235, 0.0283, 0.0262, 0.0206, 0.0243, 0.0204],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 10:17:23,971 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.350e+02 3.594e+02 4.207e+02 5.059e+02 8.265e+02, threshold=8.414e+02, percent-clipped=0.0
+2023-03-29 10:17:29,465 INFO [train.py:892] (3/4) Epoch 37, batch 1550, loss[loss=0.1418, simple_loss=0.2167, pruned_loss=0.03347, over 19875.00 frames. ], tot_loss[loss=0.1559, simple_loss=0.2362, pruned_loss=0.03777, over 3946081.28 frames. ], batch size: 136, lr: 4.20e-03, grad_scale: 16.0
+2023-03-29 10:17:40,411 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:17:42,331 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:18:08,115 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:19:17,876 INFO [train.py:892] (3/4) Epoch 37, batch 1600, loss[loss=0.1388, simple_loss=0.2236, pruned_loss=0.02701, over 19637.00 frames. ], tot_loss[loss=0.1548, simple_loss=0.2349, pruned_loss=0.03732, over 3947441.89 frames. ], batch size: 68, lr: 4.20e-03, grad_scale: 16.0
+2023-03-29 10:19:26,679 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68384.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:19:54,865 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68396.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:20:01,566 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:21:10,223 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.336e+02 3.362e+02 4.148e+02 5.414e+02 8.370e+02, threshold=8.296e+02, percent-clipped=0.0
+2023-03-29 10:21:15,819 INFO [train.py:892] (3/4) Epoch 37, batch 1650, loss[loss=0.2426, simple_loss=0.3277, pruned_loss=0.07875, over 19395.00 frames. ], tot_loss[loss=0.1563, simple_loss=0.2368, pruned_loss=0.03791, over 3946120.10 frames. ], batch size: 412, lr: 4.20e-03, grad_scale: 16.0
+2023-03-29 10:21:35,495 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0
+2023-03-29 10:21:44,695 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68443.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 10:22:24,439 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:23:09,600 INFO [train.py:892] (3/4) Epoch 37, batch 1700, loss[loss=0.1348, simple_loss=0.2094, pruned_loss=0.03007, over 19762.00 frames. ], tot_loss[loss=0.1563, simple_loss=0.237, pruned_loss=0.0378, over 3946963.55 frames. ], batch size: 155, lr: 4.20e-03, grad_scale: 16.0
+2023-03-29 10:23:51,813 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4288, 3.4928, 3.3587, 3.0989, 3.4793, 2.5865, 2.7227, 1.6376],
+       device='cuda:3'), covar=tensor([0.0357, 0.0314, 0.0268, 0.0333, 0.0273, 0.1649, 0.0963, 0.2371],
+       device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0149, 0.0115, 0.0137, 0.0122, 0.0137, 0.0144, 0.0130],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 10:24:24,183 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6446, 4.6924, 5.0455, 4.7851, 4.9588, 4.5994, 4.7745, 4.5761],
+       device='cuda:3'), covar=tensor([0.1465, 0.1776, 0.0875, 0.1289, 0.0746, 0.0874, 0.1785, 0.1987],
+       device='cuda:3'), in_proj_covar=tensor([0.0304, 0.0342, 0.0378, 0.0311, 0.0285, 0.0291, 0.0371, 0.0398],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 10:24:53,967 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.100e+02 3.371e+02 4.024e+02 4.531e+02 9.295e+02, threshold=8.049e+02, percent-clipped=2.0
+2023-03-29 10:24:57,922 INFO [train.py:892] (3/4) Epoch 37, batch 1750, loss[loss=0.1537, simple_loss=0.2273, pruned_loss=0.04007, over 19868.00 frames. ], tot_loss[loss=0.1564, simple_loss=0.2371, pruned_loss=0.03784, over 3946031.32 frames. ], batch size: 165, lr: 4.20e-03, grad_scale: 16.0
+2023-03-29 10:26:31,431 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68578.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:26:35,976 INFO [train.py:892] (3/4) Epoch 37, batch 1800, loss[loss=0.1678, simple_loss=0.2516, pruned_loss=0.04203, over 19764.00 frames. ], tot_loss[loss=0.1571, simple_loss=0.238, pruned_loss=0.03813, over 3945906.69 frames. ], batch size: 244, lr: 4.19e-03, grad_scale: 16.0
+2023-03-29 10:27:34,664 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9259, 4.8240, 5.3554, 4.8208, 4.3034, 5.0717, 4.9765, 5.5029],
+       device='cuda:3'), covar=tensor([0.0846, 0.0389, 0.0334, 0.0385, 0.0730, 0.0449, 0.0467, 0.0301],
+       device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0230, 0.0228, 0.0241, 0.0210, 0.0251, 0.0242, 0.0227],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 10:28:03,170 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.060e+02 3.457e+02 4.221e+02 5.089e+02 1.107e+03, threshold=8.442e+02, percent-clipped=2.0
+2023-03-29 10:28:06,928 INFO [train.py:892] (3/4) Epoch 37, batch 1850, loss[loss=0.1598, simple_loss=0.2434, pruned_loss=0.03805, over 19677.00 frames. ], tot_loss[loss=0.1572, simple_loss=0.2386, pruned_loss=0.03793, over 3946236.43 frames. ], batch size: 56, lr: 4.19e-03, grad_scale: 16.0
+2023-03-29 10:29:11,064 INFO [train.py:892] (3/4) Epoch 38, batch 0, loss[loss=0.1566, simple_loss=0.2409, pruned_loss=0.03619, over 19710.00 frames. ], tot_loss[loss=0.1566, simple_loss=0.2409, pruned_loss=0.03619, over 19710.00 frames. ], batch size: 78, lr: 4.14e-03, grad_scale: 16.0
+2023-03-29 10:29:11,065 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-29 10:29:46,257 INFO [train.py:926] (3/4) Epoch 38, validation: loss=0.1847, simple_loss=0.2497, pruned_loss=0.05979, over 2883724.00 frames. 
+2023-03-29 10:29:46,259 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-29 10:29:50,044 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68637.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:29:56,719 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:31:45,462 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:31:45,625 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68685.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:31:46,760 INFO [train.py:892] (3/4) Epoch 38, batch 50, loss[loss=0.1474, simple_loss=0.2261, pruned_loss=0.03436, over 19863.00 frames. ], tot_loss[loss=0.1542, simple_loss=0.2327, pruned_loss=0.0378, over 890744.07 frames. ], batch size: 64, lr: 4.13e-03, grad_scale: 16.0
+2023-03-29 10:33:14,581 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3460, 4.1922, 4.1962, 3.9136, 4.3808, 3.0282, 3.6381, 2.0221],
+       device='cuda:3'), covar=tensor([0.0229, 0.0232, 0.0169, 0.0211, 0.0155, 0.1061, 0.0780, 0.1700],
+       device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0150, 0.0116, 0.0138, 0.0122, 0.0138, 0.0145, 0.0131],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 10:33:21,861 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.214e+02 3.660e+02 4.318e+02 5.204e+02 7.882e+02, threshold=8.636e+02, percent-clipped=0.0
+2023-03-29 10:33:36,613 INFO [train.py:892] (3/4) Epoch 38, batch 100, loss[loss=0.1254, simple_loss=0.1962, pruned_loss=0.02731, over 19825.00 frames. ], tot_loss[loss=0.1551, simple_loss=0.2346, pruned_loss=0.03774, over 1569544.46 frames. ], batch size: 202, lr: 4.13e-03, grad_scale: 16.0
+2023-03-29 10:33:54,489 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68743.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 10:34:01,626 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68746.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:34:22,595 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68755.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:35:24,029 INFO [train.py:892] (3/4) Epoch 38, batch 150, loss[loss=0.1376, simple_loss=0.2164, pruned_loss=0.02942, over 19778.00 frames. ], tot_loss[loss=0.1543, simple_loss=0.2334, pruned_loss=0.03761, over 2097647.51 frames. ], batch size: 108, lr: 4.13e-03, grad_scale: 16.0
+2023-03-29 10:35:36,051 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3359, 3.1221, 3.3736, 2.6488, 3.4396, 2.8964, 3.2546, 3.4135],
+       device='cuda:3'), covar=tensor([0.0641, 0.0520, 0.0575, 0.0860, 0.0467, 0.0533, 0.0480, 0.0386],
+       device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0092, 0.0088, 0.0115, 0.0084, 0.0087, 0.0085, 0.0079],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 10:35:38,035 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68791.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 10:37:03,404 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.249e+02 3.607e+02 4.338e+02 5.110e+02 1.083e+03, threshold=8.676e+02, percent-clipped=2.0
+2023-03-29 10:37:20,041 INFO [train.py:892] (3/4) Epoch 38, batch 200, loss[loss=0.1466, simple_loss=0.2249, pruned_loss=0.0342, over 19840.00 frames. ], tot_loss[loss=0.1543, simple_loss=0.2338, pruned_loss=0.03739, over 2508377.06 frames. ], batch size: 161, lr: 4.13e-03, grad_scale: 16.0
+2023-03-29 10:37:20,962 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4400, 5.9422, 6.0550, 5.8041, 5.6981, 5.6919, 5.7032, 5.4922],
+       device='cuda:3'), covar=tensor([0.1486, 0.1364, 0.0718, 0.1172, 0.0607, 0.0680, 0.1738, 0.1979],
+       device='cuda:3'), in_proj_covar=tensor([0.0304, 0.0341, 0.0377, 0.0309, 0.0284, 0.0288, 0.0367, 0.0397],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 10:37:55,599 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6297, 4.7340, 5.0308, 4.8192, 4.9213, 4.5335, 4.7582, 4.5429],
+       device='cuda:3'), covar=tensor([0.1571, 0.1557, 0.0841, 0.1227, 0.0780, 0.0958, 0.1885, 0.2048],
+       device='cuda:3'), in_proj_covar=tensor([0.0304, 0.0341, 0.0377, 0.0309, 0.0284, 0.0289, 0.0368, 0.0398],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 10:39:13,003 INFO [train.py:892] (3/4) Epoch 38, batch 250, loss[loss=0.1798, simple_loss=0.2612, pruned_loss=0.04917, over 19714.00 frames. ], tot_loss[loss=0.1549, simple_loss=0.2343, pruned_loss=0.03775, over 2827492.43 frames. ], batch size: 305, lr: 4.13e-03, grad_scale: 16.0
+2023-03-29 10:40:13,565 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0
+2023-03-29 10:40:53,758 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.221e+02 3.576e+02 4.212e+02 4.903e+02 9.274e+02, threshold=8.425e+02, percent-clipped=1.0
+2023-03-29 10:41:06,946 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:41:10,024 INFO [train.py:892] (3/4) Epoch 38, batch 300, loss[loss=0.1447, simple_loss=0.234, pruned_loss=0.0277, over 19808.00 frames. ], tot_loss[loss=0.1551, simple_loss=0.2351, pruned_loss=0.03758, over 3076401.31 frames. ], batch size: 82, lr: 4.13e-03, grad_scale: 16.0
+2023-03-29 10:42:28,309 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.81 vs. limit=5.0
+2023-03-29 10:43:04,369 INFO [train.py:892] (3/4) Epoch 38, batch 350, loss[loss=0.1298, simple_loss=0.2133, pruned_loss=0.02316, over 19739.00 frames. ], tot_loss[loss=0.1552, simple_loss=0.2356, pruned_loss=0.03741, over 3269844.10 frames. ], batch size: 95, lr: 4.13e-03, grad_scale: 16.0
+2023-03-29 10:43:19,849 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4879, 5.8756, 5.9959, 5.7095, 5.6752, 5.7455, 5.7129, 5.4512],
+       device='cuda:3'), covar=tensor([0.1358, 0.1315, 0.0773, 0.1226, 0.0661, 0.0720, 0.1681, 0.2075],
+       device='cuda:3'), in_proj_covar=tensor([0.0305, 0.0345, 0.0380, 0.0312, 0.0286, 0.0291, 0.0370, 0.0401],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 10:44:39,949 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.577e+02 3.628e+02 4.401e+02 5.124e+02 1.005e+03, threshold=8.801e+02, percent-clipped=1.0
+2023-03-29 10:44:56,363 INFO [train.py:892] (3/4) Epoch 38, batch 400, loss[loss=0.1521, simple_loss=0.2346, pruned_loss=0.03484, over 19821.00 frames. ], tot_loss[loss=0.1564, simple_loss=0.2367, pruned_loss=0.03802, over 3421184.35 frames. ], batch size: 72, lr: 4.12e-03, grad_scale: 16.0
+2023-03-29 10:45:08,762 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69041.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:45:13,806 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69043.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:45:28,225 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9960, 3.8999, 4.2578, 3.9160, 3.6651, 4.1032, 3.9473, 4.3102],
+       device='cuda:3'), covar=tensor([0.0768, 0.0375, 0.0355, 0.0406, 0.1086, 0.0566, 0.0531, 0.0343],
+       device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0229, 0.0228, 0.0240, 0.0209, 0.0251, 0.0242, 0.0226],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 10:45:34,782 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.95 vs. limit=5.0
+2023-03-29 10:45:41,728 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:46:50,029 INFO [train.py:892] (3/4) Epoch 38, batch 450, loss[loss=0.145, simple_loss=0.2193, pruned_loss=0.03533, over 19883.00 frames. ], tot_loss[loss=0.1564, simple_loss=0.2364, pruned_loss=0.03817, over 3539392.90 frames. ], batch size: 88, lr: 4.12e-03, grad_scale: 16.0
+2023-03-29 10:47:28,965 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69103.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:47:31,224 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69104.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:48:23,827 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.342e+02 3.756e+02 4.232e+02 5.064e+02 8.468e+02, threshold=8.465e+02, percent-clipped=0.0
+2023-03-29 10:48:24,745 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0153, 3.9115, 4.3463, 3.9482, 3.7531, 4.2623, 4.0775, 4.4431],
+       device='cuda:3'), covar=tensor([0.0929, 0.0468, 0.0455, 0.0482, 0.1150, 0.0610, 0.0560, 0.0433],
+       device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0230, 0.0228, 0.0240, 0.0210, 0.0252, 0.0242, 0.0226],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 10:48:30,128 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1740, 4.2630, 2.5298, 4.5065, 4.7075, 1.9901, 3.8816, 3.4930],
+       device='cuda:3'), covar=tensor([0.0669, 0.0746, 0.2693, 0.0743, 0.0530, 0.2836, 0.0990, 0.0872],
+       device='cuda:3'), in_proj_covar=tensor([0.0239, 0.0263, 0.0234, 0.0283, 0.0262, 0.0206, 0.0243, 0.0204],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 10:48:39,811 INFO [train.py:892] (3/4) Epoch 38, batch 500, loss[loss=0.1744, simple_loss=0.2512, pruned_loss=0.04879, over 19710.00 frames. ], tot_loss[loss=0.1574, simple_loss=0.2373, pruned_loss=0.03872, over 3629217.76 frames. ], batch size: 61, lr: 4.12e-03, grad_scale: 16.0
+2023-03-29 10:49:38,580 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69162.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:49:52,881 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69168.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:50:22,716 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8796, 5.0515, 5.0915, 5.0839, 4.7362, 5.0061, 4.6258, 4.2950],
+       device='cuda:3'), covar=tensor([0.1142, 0.1009, 0.0939, 0.0767, 0.1121, 0.1073, 0.1408, 0.2439],
+       device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0302, 0.0311, 0.0273, 0.0283, 0.0264, 0.0279, 0.0328],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 10:50:32,782 INFO [train.py:892] (3/4) Epoch 38, batch 550, loss[loss=0.1417, simple_loss=0.2319, pruned_loss=0.02574, over 19653.00 frames. ], tot_loss[loss=0.1575, simple_loss=0.2375, pruned_loss=0.03879, over 3699457.57 frames. ], batch size: 69, lr: 4.12e-03, grad_scale: 16.0
+2023-03-29 10:50:42,776 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.85 vs. limit=5.0
+2023-03-29 10:50:54,516 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6379, 4.9097, 4.9644, 4.8463, 4.6182, 4.9286, 4.5092, 4.4862],
+       device='cuda:3'), covar=tensor([0.0550, 0.0520, 0.0494, 0.0445, 0.0652, 0.0504, 0.0699, 0.1036],
+       device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0303, 0.0311, 0.0273, 0.0283, 0.0264, 0.0279, 0.0328],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 10:51:40,927 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5953, 2.6600, 2.7062, 2.7170, 2.6538, 2.6762, 2.7057, 2.8147],
+       device='cuda:3'), covar=tensor([0.0368, 0.0392, 0.0402, 0.0331, 0.0457, 0.0386, 0.0410, 0.0374],
+       device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0087, 0.0090, 0.0084, 0.0096, 0.0089, 0.0106, 0.0079],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 10:51:57,257 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69223.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:52:02,706 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:52:11,497 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.978e+02 3.709e+02 4.324e+02 5.706e+02 8.766e+02, threshold=8.649e+02, percent-clipped=1.0
+2023-03-29 10:52:13,008 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:52:13,492 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.80 vs. limit=5.0
+2023-03-29 10:52:24,172 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:52:27,225 INFO [train.py:892] (3/4) Epoch 38, batch 600, loss[loss=0.1581, simple_loss=0.2383, pruned_loss=0.03894, over 19753.00 frames. ], tot_loss[loss=0.1573, simple_loss=0.2374, pruned_loss=0.03859, over 3753823.42 frames. ], batch size: 188, lr: 4.12e-03, grad_scale: 16.0
+2023-03-29 10:52:49,615 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69245.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 10:53:54,384 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:53:58,575 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7462, 2.9973, 3.2013, 3.6059, 2.6506, 3.0877, 2.4715, 2.4457],
+       device='cuda:3'), covar=tensor([0.0609, 0.1779, 0.1133, 0.0556, 0.2071, 0.0996, 0.1489, 0.1778],
+       device='cuda:3'), in_proj_covar=tensor([0.0248, 0.0327, 0.0251, 0.0208, 0.0248, 0.0212, 0.0222, 0.0219],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 10:54:14,566 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69282.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:54:14,625 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.2048, 5.3833, 5.6673, 5.3420, 5.4368, 5.2609, 5.3785, 5.1243],
+       device='cuda:3'), covar=tensor([0.1446, 0.1501, 0.0756, 0.1274, 0.0692, 0.0728, 0.1803, 0.2020],
+       device='cuda:3'), in_proj_covar=tensor([0.0303, 0.0344, 0.0379, 0.0311, 0.0287, 0.0290, 0.0370, 0.0399],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 10:54:22,233 INFO [train.py:892] (3/4) Epoch 38, batch 650, loss[loss=0.138, simple_loss=0.2162, pruned_loss=0.02987, over 19901.00 frames. ], tot_loss[loss=0.1551, simple_loss=0.2355, pruned_loss=0.03741, over 3798437.93 frames. ], batch size: 113, lr: 4.12e-03, grad_scale: 16.0
+2023-03-29 10:54:24,225 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69286.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:55:09,601 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1867, 4.0506, 4.4501, 4.0663, 3.7651, 4.2967, 4.1364, 4.4914],
+       device='cuda:3'), covar=tensor([0.0715, 0.0353, 0.0321, 0.0347, 0.1048, 0.0493, 0.0433, 0.0326],
+       device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0231, 0.0228, 0.0241, 0.0211, 0.0252, 0.0242, 0.0226],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 10:55:11,901 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69306.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 10:55:42,185 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:55:59,560 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.644e+02 3.609e+02 4.195e+02 5.280e+02 9.691e+02, threshold=8.389e+02, percent-clipped=2.0
+2023-03-29 10:56:15,743 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69334.0, num_to_drop=1, layers_to_drop={3}
+2023-03-29 10:56:18,462 INFO [train.py:892] (3/4) Epoch 38, batch 700, loss[loss=0.1726, simple_loss=0.2527, pruned_loss=0.04625, over 19734.00 frames. ], tot_loss[loss=0.1562, simple_loss=0.2369, pruned_loss=0.03772, over 3831408.83 frames. ], batch size: 291, lr: 4.12e-03, grad_scale: 16.0
+2023-03-29 10:56:30,350 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:57:30,773 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69368.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:57:36,659 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0
+2023-03-29 10:58:00,784 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:58:10,783 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4485, 4.1366, 4.2448, 4.4394, 4.1772, 4.5205, 4.5401, 4.7100],
+       device='cuda:3'), covar=tensor([0.0686, 0.0476, 0.0519, 0.0379, 0.0739, 0.0483, 0.0449, 0.0312],
+       device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0184, 0.0207, 0.0182, 0.0181, 0.0165, 0.0158, 0.0205],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 10:58:11,800 INFO [train.py:892] (3/4) Epoch 38, batch 750, loss[loss=0.1326, simple_loss=0.2166, pruned_loss=0.02434, over 19739.00 frames. ], tot_loss[loss=0.1563, simple_loss=0.2372, pruned_loss=0.03772, over 3857433.50 frames. ], batch size: 118, lr: 4.11e-03, grad_scale: 16.0
+2023-03-29 10:58:19,055 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69389.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:58:30,067 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:58:41,199 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69399.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 10:59:10,870 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0
+2023-03-29 10:59:33,008 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0
+2023-03-29 10:59:43,863 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9279, 4.9841, 5.2945, 4.9878, 5.1936, 4.7899, 5.0176, 4.7635],
+       device='cuda:3'), covar=tensor([0.1361, 0.1750, 0.0878, 0.1388, 0.0726, 0.0877, 0.1888, 0.2151],
+       device='cuda:3'), in_proj_covar=tensor([0.0303, 0.0345, 0.0379, 0.0312, 0.0288, 0.0291, 0.0371, 0.0399],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 10:59:51,962 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.082e+02 3.634e+02 4.546e+02 5.374e+02 9.782e+02, threshold=9.091e+02, percent-clipped=2.0
+2023-03-29 10:59:53,079 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:00:07,019 INFO [train.py:892] (3/4) Epoch 38, batch 800, loss[loss=0.1589, simple_loss=0.2341, pruned_loss=0.04192, over 19755.00 frames. ], tot_loss[loss=0.156, simple_loss=0.237, pruned_loss=0.03747, over 3878281.28 frames. ], batch size: 84, lr: 4.11e-03, grad_scale: 16.0
+2023-03-29 11:00:53,778 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69455.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:02:01,795 INFO [train.py:892] (3/4) Epoch 38, batch 850, loss[loss=0.1626, simple_loss=0.2453, pruned_loss=0.03989, over 19750.00 frames. ], tot_loss[loss=0.1559, simple_loss=0.2367, pruned_loss=0.03755, over 3893403.89 frames. ], batch size: 250, lr: 4.11e-03, grad_scale: 16.0
+2023-03-29 11:03:16,104 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:03:19,206 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0
+2023-03-29 11:03:30,081 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:03:39,264 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.319e+02 3.651e+02 4.238e+02 5.380e+02 9.885e+02, threshold=8.476e+02, percent-clipped=3.0
+2023-03-29 11:03:54,830 INFO [train.py:892] (3/4) Epoch 38, batch 900, loss[loss=0.1546, simple_loss=0.2418, pruned_loss=0.03373, over 19764.00 frames. ], tot_loss[loss=0.1556, simple_loss=0.2366, pruned_loss=0.03733, over 3906762.72 frames. ], batch size: 244, lr: 4.11e-03, grad_scale: 16.0
+2023-03-29 11:03:57,959 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.37 vs. limit=5.0
+2023-03-29 11:04:37,616 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5515, 2.0176, 2.3431, 2.7605, 3.2194, 3.3071, 3.1105, 3.1662],
+       device='cuda:3'), covar=tensor([0.1106, 0.1844, 0.1462, 0.0832, 0.0530, 0.0402, 0.0500, 0.0653],
+       device='cuda:3'), in_proj_covar=tensor([0.0163, 0.0169, 0.0179, 0.0154, 0.0140, 0.0135, 0.0130, 0.0120],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 11:05:35,979 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69581.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:05:45,030 INFO [train.py:892] (3/4) Epoch 38, batch 950, loss[loss=0.1336, simple_loss=0.2162, pruned_loss=0.02547, over 19768.00 frames. ], tot_loss[loss=0.155, simple_loss=0.2358, pruned_loss=0.03705, over 3917682.73 frames. ], batch size: 113, lr: 4.11e-03, grad_scale: 16.0
+2023-03-29 11:06:21,774 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69601.0, num_to_drop=1, layers_to_drop={3}
+2023-03-29 11:07:25,066 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.229e+02 3.525e+02 4.148e+02 4.911e+02 8.887e+02, threshold=8.297e+02, percent-clipped=1.0
+2023-03-29 11:07:26,060 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69629.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 11:07:42,208 INFO [train.py:892] (3/4) Epoch 38, batch 1000, loss[loss=0.1599, simple_loss=0.2393, pruned_loss=0.04022, over 19840.00 frames. ], tot_loss[loss=0.155, simple_loss=0.2362, pruned_loss=0.03693, over 3923990.44 frames. ], batch size: 58, lr: 4.11e-03, grad_scale: 16.0
+2023-03-29 11:08:08,041 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0
+2023-03-29 11:09:10,137 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69674.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 11:09:14,576 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69676.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:09:41,321 INFO [train.py:892] (3/4) Epoch 38, batch 1050, loss[loss=0.131, simple_loss=0.2039, pruned_loss=0.02904, over 19767.00 frames. ], tot_loss[loss=0.1555, simple_loss=0.2366, pruned_loss=0.03718, over 3929906.25 frames. ], batch size: 130, lr: 4.10e-03, grad_scale: 32.0
+2023-03-29 11:09:53,807 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0
+2023-03-29 11:10:13,104 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69699.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:11:13,942 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69724.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:11:23,831 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.661e+02 3.738e+02 4.279e+02 5.183e+02 8.031e+02, threshold=8.559e+02, percent-clipped=0.0
+2023-03-29 11:11:41,316 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69735.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 11:11:42,237 INFO [train.py:892] (3/4) Epoch 38, batch 1100, loss[loss=0.1387, simple_loss=0.2228, pruned_loss=0.02726, over 19875.00 frames. ], tot_loss[loss=0.1558, simple_loss=0.2369, pruned_loss=0.03737, over 3933699.27 frames. ], batch size: 46, lr: 4.10e-03, grad_scale: 32.0
+2023-03-29 11:12:07,918 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69747.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:12:15,508 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69750.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:13:36,956 INFO [train.py:892] (3/4) Epoch 38, batch 1150, loss[loss=0.1497, simple_loss=0.2318, pruned_loss=0.03385, over 19823.00 frames. ], tot_loss[loss=0.1546, simple_loss=0.2355, pruned_loss=0.03681, over 3938077.79 frames. ], batch size: 288, lr: 4.10e-03, grad_scale: 32.0
+2023-03-29 11:14:51,213 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69818.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:14:53,425 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8631, 2.8869, 1.8236, 3.2792, 3.0595, 3.2059, 3.3292, 2.6834],
+       device='cuda:3'), covar=tensor([0.0696, 0.0764, 0.1796, 0.0695, 0.0718, 0.0595, 0.0641, 0.0887],
+       device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0147, 0.0146, 0.0157, 0.0137, 0.0141, 0.0153, 0.0150],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 11:15:03,644 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69824.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:15:13,289 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.448e+02 3.531e+02 4.116e+02 4.879e+02 7.988e+02, threshold=8.232e+02, percent-clipped=0.0
+2023-03-29 11:15:27,953 INFO [train.py:892] (3/4) Epoch 38, batch 1200, loss[loss=0.1452, simple_loss=0.2195, pruned_loss=0.03551, over 19553.00 frames. ], tot_loss[loss=0.1546, simple_loss=0.2356, pruned_loss=0.03679, over 3940158.60 frames. ], batch size: 41, lr: 4.10e-03, grad_scale: 32.0
+2023-03-29 11:16:39,760 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69866.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:16:53,622 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69872.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:17:15,100 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:17:26,328 INFO [train.py:892] (3/4) Epoch 38, batch 1250, loss[loss=0.1376, simple_loss=0.2129, pruned_loss=0.03113, over 19739.00 frames. ], tot_loss[loss=0.1544, simple_loss=0.2354, pruned_loss=0.03675, over 3942342.35 frames. ], batch size: 71, lr: 4.10e-03, grad_scale: 32.0
+2023-03-29 11:17:44,332 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2862, 3.2812, 2.0526, 3.8573, 3.5079, 3.7975, 3.9100, 3.0033],
+       device='cuda:3'), covar=tensor([0.0655, 0.0646, 0.1790, 0.0735, 0.0731, 0.0519, 0.0577, 0.0876],
+       device='cuda:3'), in_proj_covar=tensor([0.0147, 0.0148, 0.0146, 0.0158, 0.0137, 0.0141, 0.0153, 0.0150],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 11:18:00,529 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69901.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 11:18:19,793 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9080, 4.1155, 4.4495, 4.0061, 3.8696, 4.3310, 4.1231, 4.5805],
+       device='cuda:3'), covar=tensor([0.1256, 0.0423, 0.0532, 0.0496, 0.1224, 0.0683, 0.0598, 0.0496],
+       device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0229, 0.0229, 0.0241, 0.0211, 0.0252, 0.0242, 0.0226],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 11:19:06,685 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:19:06,840 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69929.0, num_to_drop=1, layers_to_drop={2}
+2023-03-29 11:19:08,042 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.501e+02 3.622e+02 4.531e+02 5.693e+02 1.011e+03, threshold=9.061e+02, percent-clipped=2.0
+2023-03-29 11:19:20,442 INFO [train.py:892] (3/4) Epoch 38, batch 1300, loss[loss=0.1371, simple_loss=0.2191, pruned_loss=0.02751, over 19743.00 frames. ], tot_loss[loss=0.1551, simple_loss=0.2364, pruned_loss=0.03695, over 3942566.18 frames. ], batch size: 89, lr: 4.10e-03, grad_scale: 16.0
+2023-03-29 11:19:38,936 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1048, 3.2591, 3.2428, 3.3096, 3.1572, 3.4071, 3.0870, 3.3638],
+       device='cuda:3'), covar=tensor([0.0259, 0.0395, 0.0362, 0.0258, 0.0339, 0.0251, 0.0336, 0.0302],
+       device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0087, 0.0090, 0.0085, 0.0096, 0.0090, 0.0106, 0.0079],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 11:19:46,479 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69949.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 11:20:47,659 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69976.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:20:49,545 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:20:55,763 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4274, 4.1260, 4.2022, 4.4039, 4.1625, 4.4604, 4.5154, 4.7182],
+       device='cuda:3'), covar=tensor([0.0664, 0.0444, 0.0546, 0.0364, 0.0652, 0.0555, 0.0430, 0.0316],
+       device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0185, 0.0209, 0.0184, 0.0184, 0.0168, 0.0159, 0.0208],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 11:21:07,939 INFO [train.py:892] (3/4) Epoch 38, batch 1350, loss[loss=0.1374, simple_loss=0.2261, pruned_loss=0.02435, over 19732.00 frames. ], tot_loss[loss=0.1544, simple_loss=0.2356, pruned_loss=0.03666, over 3944975.99 frames. ], batch size: 80, lr: 4.10e-03, grad_scale: 16.0
+2023-03-29 11:21:41,280 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([6.0515, 6.3115, 6.3155, 6.1709, 6.1165, 6.3168, 5.6472, 5.6852],
+       device='cuda:3'), covar=tensor([0.0390, 0.0443, 0.0485, 0.0410, 0.0527, 0.0454, 0.0655, 0.0941],
+       device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0300, 0.0311, 0.0272, 0.0281, 0.0262, 0.0276, 0.0324],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 11:22:40,546 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:22:40,690 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:22:51,531 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.339e+02 3.460e+02 3.863e+02 4.574e+02 6.887e+02, threshold=7.727e+02, percent-clipped=0.0
+2023-03-29 11:22:53,028 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70030.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 11:23:06,003 INFO [train.py:892] (3/4) Epoch 38, batch 1400, loss[loss=0.1641, simple_loss=0.2383, pruned_loss=0.04496, over 19871.00 frames. ], tot_loss[loss=0.1542, simple_loss=0.2352, pruned_loss=0.03665, over 3945961.72 frames. ], batch size: 46, lr: 4.09e-03, grad_scale: 16.0
+2023-03-29 11:23:41,286 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:24:30,449 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70072.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:25:00,839 INFO [train.py:892] (3/4) Epoch 38, batch 1450, loss[loss=0.1601, simple_loss=0.2352, pruned_loss=0.0425, over 19878.00 frames. ], tot_loss[loss=0.1552, simple_loss=0.2362, pruned_loss=0.03708, over 3944979.43 frames. ], batch size: 158, lr: 4.09e-03, grad_scale: 16.0
+2023-03-29 11:25:26,291 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70098.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:25:31,463 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4620, 3.3913, 5.0011, 3.7334, 4.0275, 3.8446, 2.8307, 3.0117],
+       device='cuda:3'), covar=tensor([0.0851, 0.2752, 0.0395, 0.1022, 0.1634, 0.1323, 0.2477, 0.2432],
+       device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0399, 0.0356, 0.0295, 0.0381, 0.0392, 0.0386, 0.0357],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 11:25:53,099 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70110.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:26:07,483 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7308, 2.1593, 2.2660, 2.0672, 2.3280, 2.4711, 2.6934, 2.8589],
+       device='cuda:3'), covar=tensor([0.0927, 0.1594, 0.1738, 0.2071, 0.1226, 0.1273, 0.0859, 0.0728],
+       device='cuda:3'), in_proj_covar=tensor([0.0261, 0.0246, 0.0275, 0.0260, 0.0307, 0.0264, 0.0239, 0.0267],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 11:26:38,225 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.553e+02 3.599e+02 4.475e+02 5.242e+02 1.040e+03, threshold=8.950e+02, percent-clipped=4.0
+2023-03-29 11:26:50,807 INFO [train.py:892] (3/4) Epoch 38, batch 1500, loss[loss=0.1481, simple_loss=0.2241, pruned_loss=0.036, over 19780.00 frames. ], tot_loss[loss=0.1546, simple_loss=0.2355, pruned_loss=0.03691, over 3945512.72 frames. ], batch size: 217, lr: 4.09e-03, grad_scale: 16.0
+2023-03-29 11:28:04,438 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.2263, 5.5101, 5.5570, 5.4124, 5.2117, 5.5379, 4.9666, 5.0235],
+       device='cuda:3'), covar=tensor([0.0446, 0.0442, 0.0461, 0.0414, 0.0570, 0.0474, 0.0667, 0.0941],
+       device='cuda:3'), in_proj_covar=tensor([0.0282, 0.0299, 0.0310, 0.0271, 0.0280, 0.0261, 0.0275, 0.0323],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 11:28:12,797 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70171.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:28:47,251 INFO [train.py:892] (3/4) Epoch 38, batch 1550, loss[loss=0.1292, simple_loss=0.2089, pruned_loss=0.02471, over 19771.00 frames. ], tot_loss[loss=0.1543, simple_loss=0.2353, pruned_loss=0.03664, over 3946930.87 frames. ], batch size: 70, lr: 4.09e-03, grad_scale: 16.0
+2023-03-29 11:29:15,051 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9541, 3.3139, 3.4664, 3.9179, 2.7182, 3.3185, 2.5327, 2.6103],
+       device='cuda:3'), covar=tensor([0.0563, 0.1639, 0.0949, 0.0461, 0.1946, 0.0854, 0.1439, 0.1580],
+       device='cuda:3'), in_proj_covar=tensor([0.0254, 0.0331, 0.0253, 0.0210, 0.0253, 0.0215, 0.0225, 0.0222],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 11:30:25,216 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.741e+02 3.595e+02 4.117e+02 4.902e+02 9.648e+02, threshold=8.233e+02, percent-clipped=1.0
+2023-03-29 11:30:39,805 INFO [train.py:892] (3/4) Epoch 38, batch 1600, loss[loss=0.1854, simple_loss=0.2683, pruned_loss=0.05124, over 19746.00 frames. ], tot_loss[loss=0.154, simple_loss=0.235, pruned_loss=0.0365, over 3948373.69 frames. ], batch size: 209, lr: 4.09e-03, grad_scale: 16.0
+2023-03-29 11:31:34,784 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:32:29,351 INFO [train.py:892] (3/4) Epoch 38, batch 1650, loss[loss=0.1337, simple_loss=0.2077, pruned_loss=0.02985, over 19869.00 frames. ], tot_loss[loss=0.1536, simple_loss=0.2341, pruned_loss=0.03652, over 3948850.53 frames. ], batch size: 136, lr: 4.09e-03, grad_scale: 16.0
+2023-03-29 11:33:05,205 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0
+2023-03-29 11:33:52,960 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:34:12,141 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.483e+02 3.682e+02 4.356e+02 5.206e+02 1.027e+03, threshold=8.712e+02, percent-clipped=2.0
+2023-03-29 11:34:13,222 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70330.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 11:34:27,899 INFO [train.py:892] (3/4) Epoch 38, batch 1700, loss[loss=0.141, simple_loss=0.2182, pruned_loss=0.03193, over 19818.00 frames. ], tot_loss[loss=0.1552, simple_loss=0.2353, pruned_loss=0.0375, over 3948655.40 frames. ], batch size: 127, lr: 4.09e-03, grad_scale: 16.0
+2023-03-29 11:35:09,095 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0
+2023-03-29 11:35:59,748 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70378.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 11:36:12,498 INFO [train.py:892] (3/4) Epoch 38, batch 1750, loss[loss=0.1393, simple_loss=0.2244, pruned_loss=0.02703, over 19774.00 frames. ], tot_loss[loss=0.1556, simple_loss=0.2356, pruned_loss=0.03781, over 3950021.16 frames. ], batch size: 113, lr: 4.08e-03, grad_scale: 16.0
+2023-03-29 11:36:30,554 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0
+2023-03-29 11:37:39,938 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.846e+02 3.811e+02 4.551e+02 5.592e+02 1.733e+03, threshold=9.102e+02, percent-clipped=3.0
+2023-03-29 11:37:40,841 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5842, 3.3341, 3.6332, 2.8115, 3.8086, 2.9601, 3.2818, 3.5627],
+       device='cuda:3'), covar=tensor([0.0714, 0.0479, 0.0567, 0.0827, 0.0468, 0.0548, 0.0597, 0.0502],
+       device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0092, 0.0089, 0.0115, 0.0084, 0.0087, 0.0085, 0.0079],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 11:37:51,332 INFO [train.py:892] (3/4) Epoch 38, batch 1800, loss[loss=0.1475, simple_loss=0.2224, pruned_loss=0.03633, over 19741.00 frames. ], tot_loss[loss=0.1556, simple_loss=0.236, pruned_loss=0.03761, over 3949717.15 frames. ], batch size: 140, lr: 4.08e-03, grad_scale: 16.0
+2023-03-29 11:38:50,111 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:39:26,488 INFO [train.py:892] (3/4) Epoch 38, batch 1850, loss[loss=0.1492, simple_loss=0.2448, pruned_loss=0.02685, over 19850.00 frames. ], tot_loss[loss=0.1565, simple_loss=0.2377, pruned_loss=0.03766, over 3946379.33 frames. ], batch size: 58, lr: 4.08e-03, grad_scale: 16.0
+2023-03-29 11:40:29,560 INFO [train.py:892] (3/4) Epoch 39, batch 0, loss[loss=0.1499, simple_loss=0.2333, pruned_loss=0.03321, over 19757.00 frames. ], tot_loss[loss=0.1499, simple_loss=0.2333, pruned_loss=0.03321, over 19757.00 frames. ], batch size: 256, lr: 4.03e-03, grad_scale: 16.0
+2023-03-29 11:40:29,560 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-29 11:41:05,167 INFO [train.py:926] (3/4) Epoch 39, validation: loss=0.1858, simple_loss=0.25, pruned_loss=0.06079, over 2883724.00 frames. 
+2023-03-29 11:41:05,168 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-29 11:42:37,526 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.196e+02 3.428e+02 3.863e+02 4.713e+02 6.861e+02, threshold=7.726e+02, percent-clipped=0.0
+2023-03-29 11:42:50,276 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3109, 3.3066, 3.3103, 3.4447, 3.3178, 3.4686, 3.3608, 3.5142],
+       device='cuda:3'), covar=tensor([0.1122, 0.0729, 0.0806, 0.0656, 0.0966, 0.0880, 0.0788, 0.0793],
+       device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0186, 0.0210, 0.0185, 0.0185, 0.0169, 0.0160, 0.0209],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 11:43:01,762 INFO [train.py:892] (3/4) Epoch 39, batch 50, loss[loss=0.1357, simple_loss=0.2142, pruned_loss=0.02859, over 19855.00 frames. ], tot_loss[loss=0.152, simple_loss=0.2305, pruned_loss=0.03677, over 891544.05 frames. ], batch size: 122, lr: 4.03e-03, grad_scale: 16.0
+2023-03-29 11:44:59,632 INFO [train.py:892] (3/4) Epoch 39, batch 100, loss[loss=0.1405, simple_loss=0.2242, pruned_loss=0.02837, over 19835.00 frames. ], tot_loss[loss=0.1512, simple_loss=0.2308, pruned_loss=0.03577, over 1568634.19 frames. ], batch size: 143, lr: 4.02e-03, grad_scale: 16.0
+2023-03-29 11:45:57,286 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:46:29,523 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.491e+02 3.857e+02 4.460e+02 4.985e+02 1.010e+03, threshold=8.919e+02, percent-clipped=5.0
+2023-03-29 11:46:43,730 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9041, 4.2091, 2.4095, 4.4718, 4.6440, 1.9991, 3.4961, 3.1079],
+       device='cuda:3'), covar=tensor([0.0922, 0.0799, 0.3071, 0.0724, 0.0515, 0.3089, 0.1401, 0.1179],
+       device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0265, 0.0235, 0.0285, 0.0266, 0.0207, 0.0245, 0.0207],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 11:46:53,596 INFO [train.py:892] (3/4) Epoch 39, batch 150, loss[loss=0.1656, simple_loss=0.2464, pruned_loss=0.04238, over 19780.00 frames. ], tot_loss[loss=0.1508, simple_loss=0.2308, pruned_loss=0.03535, over 2095891.94 frames. ], batch size: 215, lr: 4.02e-03, grad_scale: 16.0
+2023-03-29 11:48:51,186 INFO [train.py:892] (3/4) Epoch 39, batch 200, loss[loss=0.1395, simple_loss=0.2238, pruned_loss=0.02757, over 19795.00 frames. ], tot_loss[loss=0.1522, simple_loss=0.2319, pruned_loss=0.0363, over 2507504.88 frames. ], batch size: 185, lr: 4.02e-03, grad_scale: 16.0
+2023-03-29 11:49:58,025 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6893, 2.7683, 2.7883, 2.7642, 2.7635, 2.7077, 2.7619, 2.9360],
+       device='cuda:3'), covar=tensor([0.0383, 0.0339, 0.0413, 0.0398, 0.0416, 0.0426, 0.0428, 0.0373],
+       device='cuda:3'), in_proj_covar=tensor([0.0093, 0.0088, 0.0090, 0.0085, 0.0097, 0.0090, 0.0106, 0.0079],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 11:50:22,153 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.699e+02 3.379e+02 4.143e+02 5.094e+02 9.478e+02, threshold=8.286e+02, percent-clipped=1.0
+2023-03-29 11:50:49,108 INFO [train.py:892] (3/4) Epoch 39, batch 250, loss[loss=0.1491, simple_loss=0.2261, pruned_loss=0.03606, over 19750.00 frames. ], tot_loss[loss=0.152, simple_loss=0.2319, pruned_loss=0.03607, over 2828304.67 frames. ], batch size: 110, lr: 4.02e-03, grad_scale: 16.0
+2023-03-29 11:51:47,979 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70766.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:52:29,888 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1806, 3.3622, 3.3013, 3.3399, 3.1835, 3.3453, 3.1121, 3.3769],
+       device='cuda:3'), covar=tensor([0.0263, 0.0308, 0.0295, 0.0311, 0.0383, 0.0261, 0.0380, 0.0347],
+       device='cuda:3'), in_proj_covar=tensor([0.0092, 0.0087, 0.0090, 0.0084, 0.0097, 0.0090, 0.0106, 0.0079],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 11:52:42,443 INFO [train.py:892] (3/4) Epoch 39, batch 300, loss[loss=0.1395, simple_loss=0.2226, pruned_loss=0.02817, over 19790.00 frames. ], tot_loss[loss=0.1512, simple_loss=0.231, pruned_loss=0.03567, over 3078036.87 frames. ], batch size: 94, lr: 4.02e-03, grad_scale: 16.0
+2023-03-29 11:52:55,528 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1754, 4.0164, 3.9863, 3.7661, 4.1776, 2.9899, 3.4820, 2.0586],
+       device='cuda:3'), covar=tensor([0.0190, 0.0219, 0.0152, 0.0194, 0.0145, 0.1017, 0.0657, 0.1539],
+       device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0150, 0.0116, 0.0138, 0.0122, 0.0137, 0.0145, 0.0130],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 11:53:37,736 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:53:44,299 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.13 vs. limit=2.0
+2023-03-29 11:54:14,756 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.577e+02 3.476e+02 3.994e+02 4.897e+02 8.420e+02, threshold=7.989e+02, percent-clipped=1.0
+2023-03-29 11:54:43,428 INFO [train.py:892] (3/4) Epoch 39, batch 350, loss[loss=0.149, simple_loss=0.2362, pruned_loss=0.03089, over 19481.00 frames. ], tot_loss[loss=0.1513, simple_loss=0.2319, pruned_loss=0.03536, over 3271671.95 frames. ], batch size: 43, lr: 4.02e-03, grad_scale: 16.0
+2023-03-29 11:56:39,302 INFO [train.py:892] (3/4) Epoch 39, batch 400, loss[loss=0.1494, simple_loss=0.2352, pruned_loss=0.03181, over 19662.00 frames. ], tot_loss[loss=0.152, simple_loss=0.2326, pruned_loss=0.03572, over 3422330.42 frames. ], batch size: 55, lr: 4.02e-03, grad_scale: 16.0
+2023-03-29 11:57:38,308 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70916.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 11:58:08,796 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.576e+02 3.611e+02 4.279e+02 5.158e+02 1.128e+03, threshold=8.559e+02, percent-clipped=5.0
+2023-03-29 11:58:31,830 INFO [train.py:892] (3/4) Epoch 39, batch 450, loss[loss=0.1592, simple_loss=0.2291, pruned_loss=0.0446, over 19862.00 frames. ], tot_loss[loss=0.1521, simple_loss=0.2326, pruned_loss=0.03577, over 3539297.19 frames. ], batch size: 122, lr: 4.02e-03, grad_scale: 16.0
+2023-03-29 11:59:27,834 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:00:29,090 INFO [train.py:892] (3/4) Epoch 39, batch 500, loss[loss=0.1783, simple_loss=0.2651, pruned_loss=0.04579, over 19692.00 frames. ], tot_loss[loss=0.1517, simple_loss=0.2321, pruned_loss=0.03563, over 3631817.24 frames. ], batch size: 337, lr: 4.01e-03, grad_scale: 16.0
+2023-03-29 12:00:32,541 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:02:01,503 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.739e+02 3.809e+02 4.349e+02 5.400e+02 1.199e+03, threshold=8.697e+02, percent-clipped=2.0
+2023-03-29 12:02:02,618 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:02:27,798 INFO [train.py:892] (3/4) Epoch 39, batch 550, loss[loss=0.1515, simple_loss=0.2315, pruned_loss=0.03568, over 19901.00 frames. ], tot_loss[loss=0.1533, simple_loss=0.2338, pruned_loss=0.03634, over 3700964.02 frames. ], batch size: 50, lr: 4.01e-03, grad_scale: 16.0
+2023-03-29 12:02:54,034 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71053.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:04:21,973 INFO [train.py:892] (3/4) Epoch 39, batch 600, loss[loss=0.1801, simple_loss=0.2724, pruned_loss=0.04391, over 19537.00 frames. ], tot_loss[loss=0.154, simple_loss=0.2348, pruned_loss=0.0366, over 3756226.52 frames. ], batch size: 54, lr: 4.01e-03, grad_scale: 16.0
+2023-03-29 12:04:23,410 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71091.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:04:30,168 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.77 vs. limit=5.0
+2023-03-29 12:05:55,696 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.519e+02 3.574e+02 4.295e+02 5.387e+02 1.048e+03, threshold=8.591e+02, percent-clipped=4.0
+2023-03-29 12:06:00,016 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0073, 3.3298, 3.7248, 3.2669, 4.0788, 4.0307, 4.6490, 5.2729],
+       device='cuda:3'), covar=tensor([0.0361, 0.1531, 0.1377, 0.2085, 0.1532, 0.1200, 0.0549, 0.0348],
+       device='cuda:3'), in_proj_covar=tensor([0.0259, 0.0244, 0.0271, 0.0258, 0.0305, 0.0262, 0.0236, 0.0266],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 12:06:26,057 INFO [train.py:892] (3/4) Epoch 39, batch 650, loss[loss=0.1536, simple_loss=0.2352, pruned_loss=0.03605, over 19776.00 frames. ], tot_loss[loss=0.1541, simple_loss=0.2348, pruned_loss=0.03673, over 3799173.66 frames. ], batch size: 247, lr: 4.01e-03, grad_scale: 16.0
+2023-03-29 12:07:21,416 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0
+2023-03-29 12:07:52,684 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9586, 3.3464, 3.7128, 3.2163, 4.0756, 4.0984, 4.6094, 5.2037],
+       device='cuda:3'), covar=tensor([0.0436, 0.1578, 0.1467, 0.2180, 0.1623, 0.1175, 0.0581, 0.0369],
+       device='cuda:3'), in_proj_covar=tensor([0.0258, 0.0243, 0.0270, 0.0257, 0.0303, 0.0261, 0.0235, 0.0265],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 12:07:53,133 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.30 vs. limit=5.0
+2023-03-29 12:08:25,800 INFO [train.py:892] (3/4) Epoch 39, batch 700, loss[loss=0.1406, simple_loss=0.2224, pruned_loss=0.02937, over 19830.00 frames. ], tot_loss[loss=0.1549, simple_loss=0.2356, pruned_loss=0.03713, over 3832973.58 frames. ], batch size: 166, lr: 4.01e-03, grad_scale: 16.0
+2023-03-29 12:08:29,990 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.75 vs. limit=5.0
+2023-03-29 12:09:56,521 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.218e+02 3.479e+02 4.128e+02 5.224e+02 8.144e+02, threshold=8.256e+02, percent-clipped=0.0
+2023-03-29 12:10:19,506 INFO [train.py:892] (3/4) Epoch 39, batch 750, loss[loss=0.1606, simple_loss=0.239, pruned_loss=0.04111, over 19724.00 frames. ], tot_loss[loss=0.1539, simple_loss=0.235, pruned_loss=0.03637, over 3859087.51 frames. ], batch size: 85, lr: 4.01e-03, grad_scale: 16.0
+2023-03-29 12:12:12,842 INFO [train.py:892] (3/4) Epoch 39, batch 800, loss[loss=0.1406, simple_loss=0.2261, pruned_loss=0.02751, over 19810.00 frames. ], tot_loss[loss=0.1528, simple_loss=0.2342, pruned_loss=0.03571, over 3879064.80 frames. ], batch size: 72, lr: 4.01e-03, grad_scale: 16.0
+2023-03-29 12:13:43,657 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.446e+02 3.401e+02 4.193e+02 5.158e+02 9.890e+02, threshold=8.386e+02, percent-clipped=3.0
+2023-03-29 12:14:08,920 INFO [train.py:892] (3/4) Epoch 39, batch 850, loss[loss=0.1549, simple_loss=0.2377, pruned_loss=0.0361, over 19781.00 frames. ], tot_loss[loss=0.1548, simple_loss=0.2359, pruned_loss=0.03681, over 3892229.16 frames. ], batch size: 46, lr: 4.00e-03, grad_scale: 16.0
+2023-03-29 12:14:25,338 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71348.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:15:12,608 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1010, 2.7094, 3.2983, 3.2814, 3.8055, 4.3601, 4.2192, 4.2109],
+       device='cuda:3'), covar=tensor([0.0995, 0.1646, 0.1265, 0.0725, 0.0439, 0.0218, 0.0329, 0.0440],
+       device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0169, 0.0182, 0.0155, 0.0140, 0.0136, 0.0130, 0.0121],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 12:15:33,261 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.0292, 1.4888, 1.6051, 2.2101, 2.3829, 2.5455, 2.3626, 2.4781],
+       device='cuda:3'), covar=tensor([0.1191, 0.2020, 0.1892, 0.0929, 0.0650, 0.0483, 0.0570, 0.0570],
+       device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0169, 0.0181, 0.0155, 0.0140, 0.0136, 0.0130, 0.0121],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 12:15:53,688 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71386.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:16:03,925 INFO [train.py:892] (3/4) Epoch 39, batch 900, loss[loss=0.1487, simple_loss=0.2316, pruned_loss=0.03292, over 19625.00 frames. ], tot_loss[loss=0.1549, simple_loss=0.236, pruned_loss=0.03693, over 3904575.10 frames. ], batch size: 65, lr: 4.00e-03, grad_scale: 16.0
+2023-03-29 12:17:36,399 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.339e+02 3.295e+02 3.866e+02 4.930e+02 9.721e+02, threshold=7.731e+02, percent-clipped=3.0
+2023-03-29 12:18:00,863 INFO [train.py:892] (3/4) Epoch 39, batch 950, loss[loss=0.1345, simple_loss=0.2118, pruned_loss=0.02859, over 19708.00 frames. ], tot_loss[loss=0.1544, simple_loss=0.2353, pruned_loss=0.03673, over 3915151.88 frames. ], batch size: 85, lr: 4.00e-03, grad_scale: 8.0
+2023-03-29 12:19:56,184 INFO [train.py:892] (3/4) Epoch 39, batch 1000, loss[loss=0.1833, simple_loss=0.2575, pruned_loss=0.05451, over 19750.00 frames. ], tot_loss[loss=0.1546, simple_loss=0.2354, pruned_loss=0.03694, over 3923104.48 frames. ], batch size: 250, lr: 4.00e-03, grad_scale: 8.0
+2023-03-29 12:21:17,944 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0
+2023-03-29 12:21:28,810 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.429e+02 3.687e+02 4.367e+02 5.594e+02 8.534e+02, threshold=8.735e+02, percent-clipped=3.0
+2023-03-29 12:21:52,950 INFO [train.py:892] (3/4) Epoch 39, batch 1050, loss[loss=0.1456, simple_loss=0.2275, pruned_loss=0.03184, over 19871.00 frames. ], tot_loss[loss=0.1545, simple_loss=0.2351, pruned_loss=0.03696, over 3929934.87 frames. ], batch size: 158, lr: 4.00e-03, grad_scale: 8.0
+2023-03-29 12:22:01,438 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.6466, 5.7225, 5.8540, 5.7727, 5.6516, 5.8019, 5.0415, 4.9427],
+       device='cuda:3'), covar=tensor([0.0922, 0.1131, 0.0852, 0.0678, 0.0877, 0.0903, 0.1453, 0.2327],
+       device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0305, 0.0314, 0.0275, 0.0284, 0.0265, 0.0281, 0.0329],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 12:23:51,084 INFO [train.py:892] (3/4) Epoch 39, batch 1100, loss[loss=0.1575, simple_loss=0.2423, pruned_loss=0.03637, over 19667.00 frames. ], tot_loss[loss=0.1539, simple_loss=0.2347, pruned_loss=0.03654, over 3933734.75 frames. ], batch size: 73, lr: 4.00e-03, grad_scale: 8.0
+2023-03-29 12:24:48,585 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:25:24,241 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.877e+02 3.477e+02 4.222e+02 5.368e+02 8.846e+02, threshold=8.443e+02, percent-clipped=1.0
+2023-03-29 12:25:46,045 INFO [train.py:892] (3/4) Epoch 39, batch 1150, loss[loss=0.1564, simple_loss=0.2385, pruned_loss=0.03719, over 19791.00 frames. ], tot_loss[loss=0.1535, simple_loss=0.2343, pruned_loss=0.03636, over 3936673.41 frames. ], batch size: 193, lr: 4.00e-03, grad_scale: 8.0
+2023-03-29 12:26:02,318 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71648.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:26:42,212 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3047, 3.5353, 3.0735, 2.7057, 3.1187, 3.4180, 3.3794, 3.4537],
+       device='cuda:3'), covar=tensor([0.0235, 0.0277, 0.0260, 0.0420, 0.0306, 0.0304, 0.0227, 0.0188],
+       device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0106, 0.0108, 0.0108, 0.0111, 0.0095, 0.0096, 0.0095],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 12:26:58,681 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7983, 4.4919, 4.5769, 4.3353, 4.7836, 3.2018, 4.0296, 2.3549],
+       device='cuda:3'), covar=tensor([0.0168, 0.0219, 0.0138, 0.0182, 0.0141, 0.0977, 0.0647, 0.1471],
+       device='cuda:3'), in_proj_covar=tensor([0.0107, 0.0150, 0.0115, 0.0137, 0.0122, 0.0136, 0.0145, 0.0129],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 12:27:08,323 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:27:26,115 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71686.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:27:38,377 INFO [train.py:892] (3/4) Epoch 39, batch 1200, loss[loss=0.1651, simple_loss=0.2426, pruned_loss=0.0438, over 19784.00 frames. ], tot_loss[loss=0.1533, simple_loss=0.2341, pruned_loss=0.03626, over 3940164.64 frames. ], batch size: 241, lr: 3.99e-03, grad_scale: 8.0
+2023-03-29 12:27:53,257 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=71696.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:29:12,325 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.421e+02 3.452e+02 4.248e+02 4.889e+02 8.041e+02, threshold=8.496e+02, percent-clipped=0.0
+2023-03-29 12:29:21,365 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=71734.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:29:35,338 INFO [train.py:892] (3/4) Epoch 39, batch 1250, loss[loss=0.1454, simple_loss=0.2224, pruned_loss=0.03417, over 19778.00 frames. ], tot_loss[loss=0.1538, simple_loss=0.2344, pruned_loss=0.03664, over 3941642.78 frames. ], batch size: 108, lr: 3.99e-03, grad_scale: 8.0
+2023-03-29 12:31:04,860 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4546, 2.4779, 1.6652, 2.6691, 2.5276, 2.5485, 2.6806, 2.1675],
+       device='cuda:3'), covar=tensor([0.0701, 0.0773, 0.1416, 0.0774, 0.0696, 0.0695, 0.0673, 0.1028],
+       device='cuda:3'), in_proj_covar=tensor([0.0148, 0.0149, 0.0146, 0.0159, 0.0137, 0.0142, 0.0152, 0.0151],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 12:31:23,179 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0
+2023-03-29 12:31:24,892 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9571, 3.9820, 2.4503, 4.1669, 4.3958, 1.8993, 3.6381, 3.2576],
+       device='cuda:3'), covar=tensor([0.0743, 0.0870, 0.2746, 0.0795, 0.0503, 0.2821, 0.1107, 0.0970],
+       device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0266, 0.0235, 0.0285, 0.0265, 0.0207, 0.0244, 0.0207],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 12:31:32,697 INFO [train.py:892] (3/4) Epoch 39, batch 1300, loss[loss=0.1942, simple_loss=0.265, pruned_loss=0.06176, over 19872.00 frames. ], tot_loss[loss=0.153, simple_loss=0.2335, pruned_loss=0.03627, over 3945108.09 frames. ], batch size: 138, lr: 3.99e-03, grad_scale: 8.0
+2023-03-29 12:33:04,692 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.312e+02 3.787e+02 4.345e+02 5.084e+02 1.281e+03, threshold=8.689e+02, percent-clipped=1.0
+2023-03-29 12:33:29,442 INFO [train.py:892] (3/4) Epoch 39, batch 1350, loss[loss=0.1246, simple_loss=0.2064, pruned_loss=0.02142, over 19595.00 frames. ], tot_loss[loss=0.153, simple_loss=0.2335, pruned_loss=0.03625, over 3945501.15 frames. ], batch size: 45, lr: 3.99e-03, grad_scale: 8.0
+2023-03-29 12:35:29,342 INFO [train.py:892] (3/4) Epoch 39, batch 1400, loss[loss=0.1333, simple_loss=0.2083, pruned_loss=0.02921, over 19866.00 frames. ], tot_loss[loss=0.1522, simple_loss=0.2327, pruned_loss=0.03582, over 3946920.99 frames. ], batch size: 154, lr: 3.99e-03, grad_scale: 8.0
+2023-03-29 12:35:51,988 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3412, 2.6911, 4.5806, 3.9171, 4.3868, 4.4791, 4.3233, 4.1638],
+       device='cuda:3'), covar=tensor([0.0611, 0.1024, 0.0120, 0.0805, 0.0164, 0.0231, 0.0177, 0.0202],
+       device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0106, 0.0091, 0.0154, 0.0089, 0.0102, 0.0092, 0.0089],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 12:36:11,835 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6479, 3.0456, 3.7039, 3.1325, 3.8508, 3.8596, 4.4887, 5.0321],
+       device='cuda:3'), covar=tensor([0.0510, 0.1802, 0.1475, 0.2204, 0.1774, 0.1381, 0.0620, 0.0469],
+       device='cuda:3'), in_proj_covar=tensor([0.0261, 0.0245, 0.0273, 0.0260, 0.0306, 0.0265, 0.0239, 0.0268],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 12:36:38,265 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71919.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:37:03,098 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.539e+02 3.590e+02 4.169e+02 5.056e+02 8.405e+02, threshold=8.337e+02, percent-clipped=0.0
+2023-03-29 12:37:24,619 INFO [train.py:892] (3/4) Epoch 39, batch 1450, loss[loss=0.1772, simple_loss=0.2698, pruned_loss=0.04235, over 19657.00 frames. ], tot_loss[loss=0.1525, simple_loss=0.2333, pruned_loss=0.03586, over 3948641.80 frames. ], batch size: 67, lr: 3.99e-03, grad_scale: 8.0
+2023-03-29 12:38:39,481 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:38:58,441 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71980.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:39:24,752 INFO [train.py:892] (3/4) Epoch 39, batch 1500, loss[loss=0.159, simple_loss=0.2399, pruned_loss=0.03902, over 19731.00 frames. ], tot_loss[loss=0.152, simple_loss=0.2326, pruned_loss=0.0357, over 3949739.46 frames. ], batch size: 291, lr: 3.99e-03, grad_scale: 8.0
+2023-03-29 12:40:13,286 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2965, 3.0675, 3.3886, 2.3995, 3.4579, 2.8645, 3.1408, 3.3445],
+       device='cuda:3'), covar=tensor([0.0556, 0.0507, 0.0511, 0.1015, 0.0351, 0.0528, 0.0527, 0.0336],
+       device='cuda:3'), in_proj_covar=tensor([0.0082, 0.0091, 0.0088, 0.0114, 0.0084, 0.0087, 0.0084, 0.0079],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 12:41:04,306 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.372e+02 3.694e+02 4.331e+02 4.992e+02 8.331e+02, threshold=8.661e+02, percent-clipped=0.0
+2023-03-29 12:41:29,066 INFO [train.py:892] (3/4) Epoch 39, batch 1550, loss[loss=0.16, simple_loss=0.2392, pruned_loss=0.04039, over 19877.00 frames. ], tot_loss[loss=0.1515, simple_loss=0.2324, pruned_loss=0.03527, over 3950713.14 frames. ], batch size: 139, lr: 3.98e-03, grad_scale: 8.0
+2023-03-29 12:42:32,475 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6722, 2.7123, 4.1443, 3.2112, 3.3659, 3.1398, 2.3411, 2.4985],
+       device='cuda:3'), covar=tensor([0.1142, 0.3458, 0.0566, 0.1182, 0.1985, 0.1749, 0.2882, 0.2985],
+       device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0394, 0.0353, 0.0293, 0.0378, 0.0389, 0.0384, 0.0356],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 12:43:25,411 INFO [train.py:892] (3/4) Epoch 39, batch 1600, loss[loss=0.1311, simple_loss=0.2045, pruned_loss=0.02886, over 19870.00 frames. ], tot_loss[loss=0.1526, simple_loss=0.2334, pruned_loss=0.03592, over 3949662.72 frames. ], batch size: 157, lr: 3.98e-03, grad_scale: 8.0
+2023-03-29 12:44:50,305 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:44:53,656 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.272e+02 3.496e+02 4.151e+02 4.701e+02 9.677e+02, threshold=8.302e+02, percent-clipped=1.0
+2023-03-29 12:45:19,180 INFO [train.py:892] (3/4) Epoch 39, batch 1650, loss[loss=0.1413, simple_loss=0.2234, pruned_loss=0.02959, over 19679.00 frames. ], tot_loss[loss=0.152, simple_loss=0.2328, pruned_loss=0.03557, over 3949727.64 frames. ], batch size: 64, lr: 3.98e-03, grad_scale: 8.0
+2023-03-29 12:47:10,101 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72190.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:47:11,005 INFO [train.py:892] (3/4) Epoch 39, batch 1700, loss[loss=0.1375, simple_loss=0.2238, pruned_loss=0.02555, over 19736.00 frames. ], tot_loss[loss=0.1528, simple_loss=0.2339, pruned_loss=0.03583, over 3948924.52 frames. ], batch size: 77, lr: 3.98e-03, grad_scale: 8.0
+2023-03-29 12:48:15,729 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8353, 2.4277, 4.0473, 3.6237, 4.0223, 4.0180, 3.8327, 3.8420],
+       device='cuda:3'), covar=tensor([0.0831, 0.1214, 0.0156, 0.0646, 0.0181, 0.0284, 0.0230, 0.0227],
+       device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0107, 0.0092, 0.0154, 0.0089, 0.0102, 0.0092, 0.0090],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 12:48:41,498 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.323e+02 3.706e+02 4.408e+02 5.454e+02 1.029e+03, threshold=8.816e+02, percent-clipped=3.0
+2023-03-29 12:49:01,634 INFO [train.py:892] (3/4) Epoch 39, batch 1750, loss[loss=0.1508, simple_loss=0.23, pruned_loss=0.0358, over 19769.00 frames. ], tot_loss[loss=0.1531, simple_loss=0.2343, pruned_loss=0.03599, over 3948865.23 frames. ], batch size: 241, lr: 3.98e-03, grad_scale: 8.0
+2023-03-29 12:49:42,097 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4789, 5.8527, 6.0511, 5.7333, 5.6751, 5.7312, 5.7463, 5.5028],
+       device='cuda:3'), covar=tensor([0.1549, 0.1448, 0.0826, 0.1241, 0.0678, 0.0779, 0.1942, 0.1870],
+       device='cuda:3'), in_proj_covar=tensor([0.0305, 0.0348, 0.0384, 0.0314, 0.0288, 0.0295, 0.0374, 0.0403],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 12:50:03,361 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:50:09,146 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72275.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:50:25,458 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4002, 3.7762, 3.9625, 4.4721, 3.1248, 3.4916, 2.8363, 2.6974],
+       device='cuda:3'), covar=tensor([0.0478, 0.1632, 0.0760, 0.0382, 0.1875, 0.0943, 0.1262, 0.1617],
+       device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0329, 0.0253, 0.0211, 0.0253, 0.0216, 0.0224, 0.0222],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 12:50:40,149 INFO [train.py:892] (3/4) Epoch 39, batch 1800, loss[loss=0.1365, simple_loss=0.2241, pruned_loss=0.02448, over 19622.00 frames. ], tot_loss[loss=0.1527, simple_loss=0.2337, pruned_loss=0.03587, over 3949501.12 frames. ], batch size: 52, lr: 3.98e-03, grad_scale: 8.0
+2023-03-29 12:50:41,880 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1681, 2.5148, 3.0176, 3.2876, 3.8932, 4.3911, 4.1991, 4.2837],
+       device='cuda:3'), covar=tensor([0.0898, 0.1916, 0.1590, 0.0729, 0.0428, 0.0260, 0.0374, 0.0342],
+       device='cuda:3'), in_proj_covar=tensor([0.0164, 0.0169, 0.0182, 0.0155, 0.0141, 0.0136, 0.0130, 0.0120],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 12:51:36,796 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:51:47,529 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:51:57,377 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.312e+02 3.449e+02 3.996e+02 5.050e+02 9.624e+02, threshold=7.991e+02, percent-clipped=1.0
+2023-03-29 12:52:15,292 INFO [train.py:892] (3/4) Epoch 39, batch 1850, loss[loss=0.1585, simple_loss=0.2515, pruned_loss=0.03278, over 19579.00 frames. ], tot_loss[loss=0.154, simple_loss=0.2359, pruned_loss=0.03599, over 3947470.53 frames. ], batch size: 53, lr: 3.98e-03, grad_scale: 8.0
+2023-03-29 12:53:18,932 INFO [train.py:892] (3/4) Epoch 40, batch 0, loss[loss=0.1686, simple_loss=0.2539, pruned_loss=0.04161, over 19628.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2539, pruned_loss=0.04161, over 19628.00 frames. ], batch size: 343, lr: 3.93e-03, grad_scale: 8.0
+2023-03-29 12:53:18,932 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-29 12:53:43,009 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5818, 2.6077, 4.0360, 3.0643, 3.3596, 3.0089, 2.3215, 2.3767],
+       device='cuda:3'), covar=tensor([0.1351, 0.3508, 0.0619, 0.1224, 0.2028, 0.1855, 0.3024, 0.3368],
+       device='cuda:3'), in_proj_covar=tensor([0.0354, 0.0394, 0.0352, 0.0294, 0.0377, 0.0388, 0.0384, 0.0357],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 12:53:45,033 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7787, 3.3360, 3.6246, 3.1085, 3.7643, 3.7537, 4.5332, 5.0124],
+       device='cuda:3'), covar=tensor([0.0410, 0.1714, 0.1357, 0.2277, 0.1613, 0.1574, 0.0503, 0.0391],
+       device='cuda:3'), in_proj_covar=tensor([0.0260, 0.0245, 0.0274, 0.0260, 0.0307, 0.0265, 0.0239, 0.0267],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 12:53:52,920 INFO [train.py:926] (3/4) Epoch 40, validation: loss=0.1851, simple_loss=0.2491, pruned_loss=0.0605, over 2883724.00 frames. 
+2023-03-29 12:53:52,921 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-29 12:54:43,683 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.60 vs. limit=5.0
+2023-03-29 12:55:01,287 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6893, 4.7111, 5.0616, 4.7867, 4.9422, 4.5782, 4.8066, 4.5468],
+       device='cuda:3'), covar=tensor([0.1422, 0.1491, 0.0838, 0.1257, 0.0789, 0.0823, 0.1879, 0.2028],
+       device='cuda:3'), in_proj_covar=tensor([0.0304, 0.0346, 0.0381, 0.0313, 0.0287, 0.0294, 0.0372, 0.0401],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 12:55:30,739 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72387.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:55:53,216 INFO [train.py:892] (3/4) Epoch 40, batch 50, loss[loss=0.1685, simple_loss=0.2499, pruned_loss=0.04358, over 19627.00 frames. ], tot_loss[loss=0.1534, simple_loss=0.2344, pruned_loss=0.03619, over 889948.48 frames. ], batch size: 299, lr: 3.92e-03, grad_scale: 8.0
+2023-03-29 12:57:04,511 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9954, 3.8597, 3.8300, 3.6193, 3.9902, 2.8358, 3.2649, 1.8770],
+       device='cuda:3'), covar=tensor([0.0208, 0.0231, 0.0157, 0.0214, 0.0154, 0.1117, 0.0623, 0.1715],
+       device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0151, 0.0116, 0.0137, 0.0122, 0.0137, 0.0145, 0.0130],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 12:57:12,956 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.328e+02 3.429e+02 4.154e+02 4.820e+02 1.115e+03, threshold=8.307e+02, percent-clipped=2.0
+2023-03-29 12:57:46,548 INFO [train.py:892] (3/4) Epoch 40, batch 100, loss[loss=0.1956, simple_loss=0.309, pruned_loss=0.04108, over 18717.00 frames. ], tot_loss[loss=0.1518, simple_loss=0.2324, pruned_loss=0.03557, over 1568384.41 frames. ], batch size: 564, lr: 3.92e-03, grad_scale: 8.0
+2023-03-29 12:58:35,872 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0
+2023-03-29 12:59:12,950 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0
+2023-03-29 12:59:14,485 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72485.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 12:59:37,310 INFO [train.py:892] (3/4) Epoch 40, batch 150, loss[loss=0.1552, simple_loss=0.243, pruned_loss=0.0337, over 19730.00 frames. ], tot_loss[loss=0.1526, simple_loss=0.2336, pruned_loss=0.03578, over 2095762.60 frames. ], batch size: 51, lr: 3.92e-03, grad_scale: 8.0
+2023-03-29 13:00:58,607 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.053e+02 3.352e+02 3.945e+02 4.917e+02 7.870e+02, threshold=7.891e+02, percent-clipped=0.0
+2023-03-29 13:01:02,169 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3148, 3.5248, 2.2049, 4.1605, 3.7438, 4.0574, 4.1365, 3.1880],
+       device='cuda:3'), covar=tensor([0.0752, 0.0598, 0.1627, 0.0681, 0.0586, 0.0435, 0.0594, 0.0838],
+       device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0151, 0.0147, 0.0161, 0.0139, 0.0144, 0.0155, 0.0153],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 13:01:36,357 INFO [train.py:892] (3/4) Epoch 40, batch 200, loss[loss=0.171, simple_loss=0.2481, pruned_loss=0.04692, over 19864.00 frames. ], tot_loss[loss=0.1513, simple_loss=0.232, pruned_loss=0.03534, over 2508412.00 frames. ], batch size: 64, lr: 3.92e-03, grad_scale: 8.0
+2023-03-29 13:02:44,380 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72575.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:03:24,765 INFO [train.py:892] (3/4) Epoch 40, batch 250, loss[loss=0.1687, simple_loss=0.2438, pruned_loss=0.04673, over 19764.00 frames. ], tot_loss[loss=0.1513, simple_loss=0.2319, pruned_loss=0.0354, over 2827225.37 frames. ], batch size: 244, lr: 3.92e-03, grad_scale: 8.0
+2023-03-29 13:04:28,623 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72623.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:04:44,506 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.269e+02 3.429e+02 3.885e+02 4.899e+02 8.561e+02, threshold=7.769e+02, percent-clipped=1.0
+2023-03-29 13:05:16,280 INFO [train.py:892] (3/4) Epoch 40, batch 300, loss[loss=0.1725, simple_loss=0.2573, pruned_loss=0.04382, over 19701.00 frames. ], tot_loss[loss=0.1515, simple_loss=0.2324, pruned_loss=0.0353, over 3075562.52 frames. ], batch size: 305, lr: 3.92e-03, grad_scale: 8.0
+2023-03-29 13:06:40,893 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72682.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:07:13,652 INFO [train.py:892] (3/4) Epoch 40, batch 350, loss[loss=0.1408, simple_loss=0.2235, pruned_loss=0.02905, over 19828.00 frames. ], tot_loss[loss=0.152, simple_loss=0.2329, pruned_loss=0.03553, over 3269637.20 frames. ], batch size: 127, lr: 3.92e-03, grad_scale: 8.0
+2023-03-29 13:08:29,574 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.535e+02 3.481e+02 4.095e+02 5.199e+02 1.202e+03, threshold=8.190e+02, percent-clipped=3.0
+2023-03-29 13:08:53,343 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0
+2023-03-29 13:09:02,736 INFO [train.py:892] (3/4) Epoch 40, batch 400, loss[loss=0.1315, simple_loss=0.2091, pruned_loss=0.02697, over 19864.00 frames. ], tot_loss[loss=0.1519, simple_loss=0.2327, pruned_loss=0.03548, over 3420533.53 frames. ], batch size: 46, lr: 3.91e-03, grad_scale: 8.0
+2023-03-29 13:09:03,780 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8926, 3.7552, 4.1117, 3.7672, 3.5768, 4.0209, 3.8711, 4.1811],
+       device='cuda:3'), covar=tensor([0.0729, 0.0383, 0.0358, 0.0411, 0.1140, 0.0593, 0.0496, 0.0362],
+       device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0228, 0.0228, 0.0241, 0.0211, 0.0253, 0.0242, 0.0226],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 13:09:05,949 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1253, 2.7637, 3.1070, 3.2093, 3.7136, 4.2220, 4.0150, 4.0817],
+       device='cuda:3'), covar=tensor([0.0886, 0.1417, 0.1246, 0.0690, 0.0484, 0.0211, 0.0405, 0.0395],
+       device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0171, 0.0184, 0.0156, 0.0143, 0.0137, 0.0131, 0.0121],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 13:10:20,586 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8925, 4.5094, 4.5779, 4.3406, 4.8773, 3.1375, 3.9416, 2.4523],
+       device='cuda:3'), covar=tensor([0.0184, 0.0221, 0.0152, 0.0195, 0.0134, 0.1039, 0.0802, 0.1525],
+       device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0151, 0.0117, 0.0139, 0.0123, 0.0138, 0.0146, 0.0131],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003],
+       device='cuda:3')
+2023-03-29 13:10:34,508 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72785.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:10:56,588 INFO [train.py:892] (3/4) Epoch 40, batch 450, loss[loss=0.2101, simple_loss=0.2893, pruned_loss=0.06541, over 19617.00 frames. ], tot_loss[loss=0.1524, simple_loss=0.2334, pruned_loss=0.03572, over 3537548.00 frames. ], batch size: 387, lr: 3.91e-03, grad_scale: 8.0
+2023-03-29 13:11:21,166 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3492, 3.6408, 3.8537, 4.4213, 2.8364, 3.3558, 2.9593, 2.7273],
+       device='cuda:3'), covar=tensor([0.0562, 0.1751, 0.0935, 0.0459, 0.2166, 0.1164, 0.1335, 0.1731],
+       device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0327, 0.0251, 0.0208, 0.0250, 0.0213, 0.0222, 0.0219],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 13:12:05,340 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4524, 5.6884, 5.7602, 5.5838, 5.4966, 5.6942, 5.1324, 5.1149],
+       device='cuda:3'), covar=tensor([0.0448, 0.0458, 0.0428, 0.0450, 0.0594, 0.0513, 0.0642, 0.0976],
+       device='cuda:3'), in_proj_covar=tensor([0.0284, 0.0303, 0.0312, 0.0273, 0.0281, 0.0262, 0.0279, 0.0325],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 13:12:17,043 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.344e+02 3.525e+02 4.138e+02 5.025e+02 1.485e+03, threshold=8.275e+02, percent-clipped=3.0
+2023-03-29 13:12:23,798 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72833.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:12:56,061 INFO [train.py:892] (3/4) Epoch 40, batch 500, loss[loss=0.1641, simple_loss=0.2526, pruned_loss=0.0378, over 19843.00 frames. ], tot_loss[loss=0.1536, simple_loss=0.2346, pruned_loss=0.03633, over 3627794.51 frames. ], batch size: 58, lr: 3.91e-03, grad_scale: 8.0
+2023-03-29 13:14:11,350 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.2141, 2.1898, 2.3210, 2.3077, 2.2957, 2.3002, 2.2227, 2.3403],
+       device='cuda:3'), covar=tensor([0.0446, 0.0377, 0.0397, 0.0341, 0.0499, 0.0394, 0.0514, 0.0382],
+       device='cuda:3'), in_proj_covar=tensor([0.0094, 0.0088, 0.0090, 0.0085, 0.0098, 0.0090, 0.0107, 0.0079],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 13:14:44,554 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72894.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:14:47,547 INFO [train.py:892] (3/4) Epoch 40, batch 550, loss[loss=0.1605, simple_loss=0.2393, pruned_loss=0.04086, over 19879.00 frames. ], tot_loss[loss=0.1533, simple_loss=0.2341, pruned_loss=0.03626, over 3700677.42 frames. ], batch size: 88, lr: 3.91e-03, grad_scale: 8.0
+2023-03-29 13:16:09,048 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.263e+02 3.703e+02 4.460e+02 5.204e+02 8.287e+02, threshold=8.920e+02, percent-clipped=1.0
+2023-03-29 13:16:45,764 INFO [train.py:892] (3/4) Epoch 40, batch 600, loss[loss=0.1677, simple_loss=0.2551, pruned_loss=0.04016, over 19645.00 frames. ], tot_loss[loss=0.1531, simple_loss=0.2338, pruned_loss=0.03616, over 3754184.79 frames. ], batch size: 79, lr: 3.91e-03, grad_scale: 8.0
+2023-03-29 13:17:07,338 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:17:26,641 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72964.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 13:18:08,986 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72982.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:18:38,511 INFO [train.py:892] (3/4) Epoch 40, batch 650, loss[loss=0.169, simple_loss=0.2465, pruned_loss=0.04578, over 19800.00 frames. ], tot_loss[loss=0.1539, simple_loss=0.235, pruned_loss=0.03637, over 3796813.10 frames. ], batch size: 67, lr: 3.91e-03, grad_scale: 8.0
+2023-03-29 13:19:43,362 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73025.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 13:19:47,088 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9925, 4.0718, 2.3860, 4.2157, 4.4225, 2.0614, 3.6824, 3.3059],
+       device='cuda:3'), covar=tensor([0.0756, 0.0873, 0.2936, 0.0786, 0.0541, 0.2735, 0.1051, 0.0902],
+       device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0268, 0.0238, 0.0286, 0.0268, 0.0209, 0.0246, 0.0208],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 13:19:55,889 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73030.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:19:57,060 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.367e+02 3.602e+02 4.054e+02 4.696e+02 9.081e+02, threshold=8.108e+02, percent-clipped=1.0
+2023-03-29 13:20:13,306 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4398, 3.2414, 3.3186, 3.4744, 3.3698, 3.4070, 3.5248, 3.7188],
+       device='cuda:3'), covar=tensor([0.0770, 0.0564, 0.0606, 0.0484, 0.0782, 0.0763, 0.0572, 0.0384],
+       device='cuda:3'), in_proj_covar=tensor([0.0158, 0.0184, 0.0206, 0.0182, 0.0181, 0.0165, 0.0157, 0.0204],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 13:20:29,528 INFO [train.py:892] (3/4) Epoch 40, batch 700, loss[loss=0.1416, simple_loss=0.2213, pruned_loss=0.031, over 19798.00 frames. ], tot_loss[loss=0.1526, simple_loss=0.2339, pruned_loss=0.03569, over 3830083.86 frames. ], batch size: 83, lr: 3.91e-03, grad_scale: 8.0
+2023-03-29 13:22:25,644 INFO [train.py:892] (3/4) Epoch 40, batch 750, loss[loss=0.1498, simple_loss=0.2328, pruned_loss=0.03346, over 19805.00 frames. ], tot_loss[loss=0.1524, simple_loss=0.2336, pruned_loss=0.03562, over 3856481.30 frames. ], batch size: 82, lr: 3.91e-03, grad_scale: 8.0
+2023-03-29 13:22:52,643 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4601, 2.4797, 4.0267, 2.9857, 3.1448, 2.9306, 2.2698, 2.4360],
+       device='cuda:3'), covar=tensor([0.1410, 0.3826, 0.0605, 0.1314, 0.2415, 0.2058, 0.2992, 0.3082],
+       device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0399, 0.0356, 0.0297, 0.0382, 0.0391, 0.0387, 0.0360],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 13:23:43,994 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.740e+02 3.647e+02 4.174e+02 4.953e+02 1.017e+03, threshold=8.348e+02, percent-clipped=2.0
+2023-03-29 13:24:07,546 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.10 vs. limit=2.0
+2023-03-29 13:24:18,100 INFO [train.py:892] (3/4) Epoch 40, batch 800, loss[loss=0.1405, simple_loss=0.2227, pruned_loss=0.0291, over 19814.00 frames. ], tot_loss[loss=0.1524, simple_loss=0.2338, pruned_loss=0.03549, over 3877737.58 frames. ], batch size: 98, lr: 3.90e-03, grad_scale: 8.0
+2023-03-29 13:25:14,401 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6335, 2.8980, 2.6012, 2.1066, 2.6791, 2.8441, 2.8401, 2.9149],
+       device='cuda:3'), covar=tensor([0.0448, 0.0340, 0.0359, 0.0627, 0.0430, 0.0352, 0.0291, 0.0264],
+       device='cuda:3'), in_proj_covar=tensor([0.0113, 0.0106, 0.0107, 0.0107, 0.0110, 0.0095, 0.0096, 0.0095],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 13:25:54,991 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0154, 2.4769, 2.9618, 3.2313, 3.6391, 4.0488, 3.9383, 3.9911],
+       device='cuda:3'), covar=tensor([0.0964, 0.1620, 0.1335, 0.0686, 0.0466, 0.0276, 0.0381, 0.0380],
+       device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0171, 0.0184, 0.0156, 0.0143, 0.0138, 0.0131, 0.0122],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 13:26:08,404 INFO [train.py:892] (3/4) Epoch 40, batch 850, loss[loss=0.2095, simple_loss=0.2886, pruned_loss=0.06524, over 19627.00 frames. ], tot_loss[loss=0.1525, simple_loss=0.2339, pruned_loss=0.03556, over 3894179.29 frames. ], batch size: 359, lr: 3.90e-03, grad_scale: 8.0
+2023-03-29 13:26:20,073 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2712, 3.2610, 2.0980, 3.8780, 3.5803, 3.8682, 3.8989, 3.0799],
+       device='cuda:3'), covar=tensor([0.0668, 0.0688, 0.1685, 0.0626, 0.0579, 0.0421, 0.0609, 0.0825],
+       device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0151, 0.0147, 0.0161, 0.0139, 0.0144, 0.0155, 0.0153],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 13:27:30,163 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.270e+02 3.428e+02 4.036e+02 4.943e+02 8.954e+02, threshold=8.072e+02, percent-clipped=2.0
+2023-03-29 13:27:35,386 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3173, 3.0275, 3.3597, 2.9175, 3.5205, 3.4146, 4.0535, 4.4862],
+       device='cuda:3'), covar=tensor([0.0510, 0.1621, 0.1487, 0.2190, 0.1533, 0.1491, 0.0632, 0.0557],
+       device='cuda:3'), in_proj_covar=tensor([0.0261, 0.0247, 0.0275, 0.0260, 0.0305, 0.0265, 0.0240, 0.0267],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 13:28:06,258 INFO [train.py:892] (3/4) Epoch 40, batch 900, loss[loss=0.1372, simple_loss=0.2239, pruned_loss=0.02528, over 19705.00 frames. ], tot_loss[loss=0.1524, simple_loss=0.234, pruned_loss=0.03539, over 3906468.45 frames. ], batch size: 78, lr: 3.90e-03, grad_scale: 8.0
+2023-03-29 13:28:17,293 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73250.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:28:32,452 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3043, 4.1355, 4.5603, 4.1062, 3.8631, 4.4249, 4.2361, 4.6583],
+       device='cuda:3'), covar=tensor([0.0763, 0.0389, 0.0371, 0.0408, 0.1076, 0.0590, 0.0509, 0.0355],
+       device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0230, 0.0229, 0.0242, 0.0212, 0.0254, 0.0243, 0.0228],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 13:28:36,974 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8400, 3.9322, 2.4048, 4.0497, 4.2401, 1.9534, 3.5063, 3.2087],
+       device='cuda:3'), covar=tensor([0.0793, 0.0860, 0.2774, 0.0846, 0.0693, 0.2902, 0.1124, 0.0986],
+       device='cuda:3'), in_proj_covar=tensor([0.0240, 0.0266, 0.0236, 0.0285, 0.0265, 0.0207, 0.0244, 0.0207],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 13:29:54,218 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73293.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:29:59,519 INFO [train.py:892] (3/4) Epoch 40, batch 950, loss[loss=0.1431, simple_loss=0.2292, pruned_loss=0.0285, over 19481.00 frames. ], tot_loss[loss=0.1516, simple_loss=0.233, pruned_loss=0.03516, over 3917079.18 frames. ], batch size: 43, lr: 3.90e-03, grad_scale: 8.0
+2023-03-29 13:30:56,555 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73320.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 13:31:21,661 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.304e+02 3.673e+02 4.396e+02 5.131e+02 1.124e+03, threshold=8.792e+02, percent-clipped=1.0
+2023-03-29 13:31:53,828 INFO [train.py:892] (3/4) Epoch 40, batch 1000, loss[loss=0.1448, simple_loss=0.2188, pruned_loss=0.03543, over 19796.00 frames. ], tot_loss[loss=0.1514, simple_loss=0.2328, pruned_loss=0.03498, over 3925386.29 frames. ], batch size: 149, lr: 3.90e-03, grad_scale: 8.0
+2023-03-29 13:32:15,086 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73354.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:32:23,184 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8247, 2.8151, 2.9539, 2.5203, 3.0298, 2.5401, 2.9379, 2.9168],
+       device='cuda:3'), covar=tensor([0.0595, 0.0540, 0.0500, 0.0758, 0.0464, 0.0562, 0.0488, 0.0405],
+       device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0092, 0.0089, 0.0114, 0.0085, 0.0088, 0.0085, 0.0079],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 13:33:10,921 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4679, 5.7663, 5.8411, 5.7258, 5.4981, 5.8267, 5.1740, 5.2321],
+       device='cuda:3'), covar=tensor([0.0485, 0.0486, 0.0510, 0.0476, 0.0681, 0.0512, 0.0754, 0.0989],
+       device='cuda:3'), in_proj_covar=tensor([0.0285, 0.0305, 0.0312, 0.0274, 0.0283, 0.0262, 0.0279, 0.0325],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 13:33:45,741 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1428, 3.1202, 3.2020, 2.7524, 3.3667, 2.8319, 3.2331, 3.2735],
+       device='cuda:3'), covar=tensor([0.0632, 0.0495, 0.0656, 0.0729, 0.0462, 0.0492, 0.0422, 0.0411],
+       device='cuda:3'), in_proj_covar=tensor([0.0083, 0.0092, 0.0089, 0.0115, 0.0085, 0.0088, 0.0085, 0.0080],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 13:33:49,229 INFO [train.py:892] (3/4) Epoch 40, batch 1050, loss[loss=0.1339, simple_loss=0.2111, pruned_loss=0.02838, over 19897.00 frames. ], tot_loss[loss=0.1515, simple_loss=0.2327, pruned_loss=0.03517, over 3931902.66 frames. ], batch size: 116, lr: 3.90e-03, grad_scale: 8.0
+2023-03-29 13:35:09,643 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.437e+02 3.473e+02 4.249e+02 4.984e+02 7.164e+02, threshold=8.499e+02, percent-clipped=0.0
+2023-03-29 13:35:15,801 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1432, 2.6195, 4.3807, 3.8133, 4.1288, 4.3472, 4.1234, 4.0636],
+       device='cuda:3'), covar=tensor([0.0639, 0.1039, 0.0120, 0.0708, 0.0184, 0.0219, 0.0205, 0.0190],
+       device='cuda:3'), in_proj_covar=tensor([0.0103, 0.0106, 0.0091, 0.0153, 0.0089, 0.0102, 0.0092, 0.0089],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 13:35:39,507 INFO [train.py:892] (3/4) Epoch 40, batch 1100, loss[loss=0.1364, simple_loss=0.2172, pruned_loss=0.02781, over 19849.00 frames. ], tot_loss[loss=0.1517, simple_loss=0.233, pruned_loss=0.03527, over 3936929.31 frames. ], batch size: 104, lr: 3.90e-03, grad_scale: 16.0
+2023-03-29 13:36:38,347 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8852, 4.0780, 4.4489, 4.9869, 3.2698, 3.4888, 3.1321, 3.1237],
+       device='cuda:3'), covar=tensor([0.0427, 0.1961, 0.0684, 0.0311, 0.1939, 0.1131, 0.1197, 0.1491],
+       device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0329, 0.0253, 0.0209, 0.0252, 0.0214, 0.0224, 0.0221],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 13:36:53,384 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2966, 3.1924, 4.9211, 3.7881, 3.8409, 3.6919, 2.6503, 2.9229],
+       device='cuda:3'), covar=tensor([0.0992, 0.3618, 0.0472, 0.1039, 0.1862, 0.1621, 0.2896, 0.2616],
+       device='cuda:3'), in_proj_covar=tensor([0.0359, 0.0400, 0.0357, 0.0297, 0.0382, 0.0393, 0.0389, 0.0361],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 13:37:30,940 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0
+2023-03-29 13:37:31,524 INFO [train.py:892] (3/4) Epoch 40, batch 1150, loss[loss=0.1502, simple_loss=0.2219, pruned_loss=0.03922, over 19747.00 frames. ], tot_loss[loss=0.1523, simple_loss=0.2331, pruned_loss=0.03577, over 3939702.45 frames. ], batch size: 139, lr: 3.89e-03, grad_scale: 16.0
+2023-03-29 13:38:04,420 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8845, 3.3123, 3.6855, 3.2090, 4.0448, 4.0315, 4.6530, 5.1827],
+       device='cuda:3'), covar=tensor([0.0465, 0.1554, 0.1441, 0.2157, 0.1461, 0.1282, 0.0521, 0.0362],
+       device='cuda:3'), in_proj_covar=tensor([0.0263, 0.0249, 0.0277, 0.0262, 0.0308, 0.0267, 0.0242, 0.0268],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 13:38:51,538 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.496e+02 3.768e+02 4.402e+02 5.430e+02 9.903e+02, threshold=8.803e+02, percent-clipped=3.0
+2023-03-29 13:39:27,961 INFO [train.py:892] (3/4) Epoch 40, batch 1200, loss[loss=0.1362, simple_loss=0.2071, pruned_loss=0.03267, over 19802.00 frames. ], tot_loss[loss=0.1519, simple_loss=0.2331, pruned_loss=0.03533, over 3939773.97 frames. ], batch size: 149, lr: 3.89e-03, grad_scale: 16.0
+2023-03-29 13:39:37,195 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73550.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:40:59,125 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7668, 2.9069, 4.2841, 3.3473, 3.4403, 3.3201, 2.4302, 2.5956],
+       device='cuda:3'), covar=tensor([0.1047, 0.2904, 0.0501, 0.1019, 0.1804, 0.1442, 0.2499, 0.2700],
+       device='cuda:3'), in_proj_covar=tensor([0.0356, 0.0397, 0.0355, 0.0296, 0.0380, 0.0391, 0.0386, 0.0359],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 13:41:06,771 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73590.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:41:19,028 INFO [train.py:892] (3/4) Epoch 40, batch 1250, loss[loss=0.1348, simple_loss=0.2099, pruned_loss=0.02987, over 19769.00 frames. ], tot_loss[loss=0.1515, simple_loss=0.2326, pruned_loss=0.03524, over 3943165.63 frames. ], batch size: 152, lr: 3.89e-03, grad_scale: 16.0
+2023-03-29 13:41:24,269 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73598.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:42:14,474 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73620.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 13:42:38,880 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.357e+02 3.537e+02 4.071e+02 5.247e+02 9.229e+02, threshold=8.143e+02, percent-clipped=1.0
+2023-03-29 13:43:11,853 INFO [train.py:892] (3/4) Epoch 40, batch 1300, loss[loss=0.1472, simple_loss=0.2332, pruned_loss=0.03058, over 19674.00 frames. ], tot_loss[loss=0.1516, simple_loss=0.2328, pruned_loss=0.03515, over 3944395.38 frames. ], batch size: 55, lr: 3.89e-03, grad_scale: 16.0
+2023-03-29 13:43:19,019 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73649.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:43:23,453 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73651.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:44:03,520 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73668.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 13:45:04,583 INFO [train.py:892] (3/4) Epoch 40, batch 1350, loss[loss=0.1544, simple_loss=0.2241, pruned_loss=0.04237, over 19833.00 frames. ], tot_loss[loss=0.1522, simple_loss=0.233, pruned_loss=0.03568, over 3946189.25 frames. ], batch size: 128, lr: 3.89e-03, grad_scale: 16.0
+2023-03-29 13:45:19,552 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.3520, 2.3472, 2.4427, 2.4682, 2.4424, 2.4093, 2.3799, 2.5164],
+       device='cuda:3'), covar=tensor([0.0459, 0.0371, 0.0433, 0.0343, 0.0463, 0.0394, 0.0518, 0.0382],
+       device='cuda:3'), in_proj_covar=tensor([0.0095, 0.0089, 0.0091, 0.0086, 0.0099, 0.0092, 0.0108, 0.0080],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 13:45:48,904 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9709, 2.6890, 3.1453, 3.1812, 3.6151, 4.0703, 3.9231, 3.9231],
+       device='cuda:3'), covar=tensor([0.0982, 0.1497, 0.1130, 0.0703, 0.0451, 0.0247, 0.0411, 0.0425],
+       device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0171, 0.0183, 0.0156, 0.0143, 0.0137, 0.0131, 0.0121],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 13:46:21,787 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.112e+02 3.360e+02 4.018e+02 4.695e+02 9.172e+02, threshold=8.036e+02, percent-clipped=1.0
+2023-03-29 13:46:24,952 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73732.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:46:56,753 INFO [train.py:892] (3/4) Epoch 40, batch 1400, loss[loss=0.1394, simple_loss=0.224, pruned_loss=0.02745, over 19736.00 frames. ], tot_loss[loss=0.1514, simple_loss=0.2316, pruned_loss=0.03554, over 3947950.81 frames. ], batch size: 95, lr: 3.89e-03, grad_scale: 16.0
+2023-03-29 13:48:43,272 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.1848, 4.7527, 4.8364, 4.5463, 5.1239, 3.2527, 4.1724, 2.4906],
+       device='cuda:3'), covar=tensor([0.0178, 0.0224, 0.0150, 0.0205, 0.0145, 0.1003, 0.0847, 0.1587],
+       device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0152, 0.0117, 0.0138, 0.0122, 0.0138, 0.0145, 0.0131],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 13:48:43,434 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73793.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:48:49,095 INFO [train.py:892] (3/4) Epoch 40, batch 1450, loss[loss=0.1502, simple_loss=0.2273, pruned_loss=0.03657, over 19722.00 frames. ], tot_loss[loss=0.1516, simple_loss=0.232, pruned_loss=0.03559, over 3949185.17 frames. ], batch size: 61, lr: 3.89e-03, grad_scale: 16.0
+2023-03-29 13:49:11,355 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0
+2023-03-29 13:50:11,139 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.549e+02 3.468e+02 4.023e+02 4.768e+02 7.367e+02, threshold=8.047e+02, percent-clipped=0.0
+2023-03-29 13:50:45,943 INFO [train.py:892] (3/4) Epoch 40, batch 1500, loss[loss=0.1492, simple_loss=0.2302, pruned_loss=0.03415, over 19820.00 frames. ], tot_loss[loss=0.1529, simple_loss=0.2333, pruned_loss=0.03623, over 3949074.60 frames. ], batch size: 103, lr: 3.89e-03, grad_scale: 16.0
+2023-03-29 13:50:59,151 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0
+2023-03-29 13:52:34,844 INFO [train.py:892] (3/4) Epoch 40, batch 1550, loss[loss=0.1667, simple_loss=0.2405, pruned_loss=0.04643, over 19785.00 frames. ], tot_loss[loss=0.1529, simple_loss=0.2332, pruned_loss=0.03626, over 3950153.37 frames. ], batch size: 236, lr: 3.88e-03, grad_scale: 16.0
+2023-03-29 13:53:51,997 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.479e+02 3.617e+02 4.156e+02 5.212e+02 8.023e+02, threshold=8.312e+02, percent-clipped=0.0
+2023-03-29 13:54:26,933 INFO [train.py:892] (3/4) Epoch 40, batch 1600, loss[loss=0.1539, simple_loss=0.24, pruned_loss=0.03396, over 19664.00 frames. ], tot_loss[loss=0.1526, simple_loss=0.233, pruned_loss=0.03612, over 3950429.02 frames. ], batch size: 57, lr: 3.88e-03, grad_scale: 16.0
+2023-03-29 13:54:27,713 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73946.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:54:33,886 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:54:49,128 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0
+2023-03-29 13:56:21,356 INFO [train.py:892] (3/4) Epoch 40, batch 1650, loss[loss=0.1568, simple_loss=0.2382, pruned_loss=0.03768, over 19683.00 frames. ], tot_loss[loss=0.1518, simple_loss=0.2323, pruned_loss=0.0357, over 3950339.62 frames. ], batch size: 59, lr: 3.88e-03, grad_scale: 16.0
+2023-03-29 13:56:27,280 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73997.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:57:48,765 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.164e+02 3.493e+02 4.073e+02 5.082e+02 8.769e+02, threshold=8.147e+02, percent-clipped=1.0
+2023-03-29 13:58:18,972 INFO [train.py:892] (3/4) Epoch 40, batch 1700, loss[loss=0.135, simple_loss=0.22, pruned_loss=0.02502, over 19756.00 frames. ], tot_loss[loss=0.1519, simple_loss=0.2324, pruned_loss=0.0357, over 3949765.85 frames. ], batch size: 88, lr: 3.88e-03, grad_scale: 16.0
+2023-03-29 13:59:01,498 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74064.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 13:59:11,412 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0778, 4.9597, 5.5166, 4.9387, 4.3917, 5.2029, 5.1032, 5.6516],
+       device='cuda:3'), covar=tensor([0.0839, 0.0378, 0.0359, 0.0407, 0.0795, 0.0468, 0.0452, 0.0269],
+       device='cuda:3'), in_proj_covar=tensor([0.0288, 0.0230, 0.0228, 0.0242, 0.0211, 0.0253, 0.0243, 0.0227],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 13:59:50,981 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74088.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:00:06,298 INFO [train.py:892] (3/4) Epoch 40, batch 1750, loss[loss=0.1404, simple_loss=0.2234, pruned_loss=0.02865, over 19783.00 frames. ], tot_loss[loss=0.1526, simple_loss=0.2329, pruned_loss=0.03609, over 3948278.80 frames. ], batch size: 65, lr: 3.88e-03, grad_scale: 16.0
+2023-03-29 14:00:32,289 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.52 vs. limit=5.0
+2023-03-29 14:00:58,648 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4429, 2.7622, 4.8202, 4.0839, 4.5588, 4.7236, 4.6069, 4.4344],
+       device='cuda:3'), covar=tensor([0.0648, 0.1103, 0.0110, 0.0952, 0.0151, 0.0223, 0.0175, 0.0159],
+       device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0106, 0.0091, 0.0154, 0.0090, 0.0103, 0.0093, 0.0090],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 14:00:59,201 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0
+2023-03-29 14:01:04,076 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74125.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:01:13,864 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.104e+02 3.603e+02 4.226e+02 5.080e+02 8.016e+02, threshold=8.451e+02, percent-clipped=0.0
+2023-03-29 14:01:43,823 INFO [train.py:892] (3/4) Epoch 40, batch 1800, loss[loss=0.1562, simple_loss=0.2457, pruned_loss=0.03334, over 19853.00 frames. ], tot_loss[loss=0.152, simple_loss=0.2324, pruned_loss=0.03579, over 3949227.84 frames. ], batch size: 81, lr: 3.88e-03, grad_scale: 16.0
+2023-03-29 14:01:59,223 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4399, 3.7728, 3.8917, 4.5118, 3.0289, 3.4205, 2.8622, 2.7559],
+       device='cuda:3'), covar=tensor([0.0467, 0.1778, 0.0831, 0.0358, 0.1869, 0.1008, 0.1257, 0.1576],
+       device='cuda:3'), in_proj_covar=tensor([0.0251, 0.0327, 0.0252, 0.0210, 0.0251, 0.0214, 0.0223, 0.0220],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 14:02:56,873 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9847, 4.8405, 5.3331, 4.8309, 4.3130, 5.0916, 4.9631, 5.4832],
+       device='cuda:3'), covar=tensor([0.0791, 0.0394, 0.0362, 0.0372, 0.0817, 0.0501, 0.0483, 0.0332],
+       device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0229, 0.0227, 0.0241, 0.0211, 0.0253, 0.0242, 0.0227],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 14:03:18,047 INFO [train.py:892] (3/4) Epoch 40, batch 1850, loss[loss=0.1693, simple_loss=0.2599, pruned_loss=0.0394, over 19849.00 frames. ], tot_loss[loss=0.1527, simple_loss=0.2342, pruned_loss=0.03558, over 3949072.50 frames. ], batch size: 58, lr: 3.88e-03, grad_scale: 16.0
+2023-03-29 14:04:24,221 INFO [train.py:892] (3/4) Epoch 41, batch 0, loss[loss=0.137, simple_loss=0.2156, pruned_loss=0.02923, over 19829.00 frames. ], tot_loss[loss=0.137, simple_loss=0.2156, pruned_loss=0.02923, over 19829.00 frames. ], batch size: 121, lr: 3.83e-03, grad_scale: 16.0
+2023-03-29 14:04:24,222 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-29 14:04:45,621 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7558, 2.4158, 2.9570, 2.6058, 3.1499, 3.2762, 2.9465, 3.1517],
+       device='cuda:3'), covar=tensor([0.0579, 0.0869, 0.0138, 0.0350, 0.0155, 0.0242, 0.0217, 0.0224],
+       device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0106, 0.0091, 0.0153, 0.0090, 0.0102, 0.0093, 0.0090],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 14:04:57,765 INFO [train.py:926] (3/4) Epoch 41, validation: loss=0.1869, simple_loss=0.2502, pruned_loss=0.06181, over 2883724.00 frames. 
+2023-03-29 14:04:57,767 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-29 14:06:10,082 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.348e+02 3.328e+02 3.806e+02 4.731e+02 7.088e+02, threshold=7.612e+02, percent-clipped=0.0
+2023-03-29 14:06:31,513 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74240.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:06:46,822 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:06:57,521 INFO [train.py:892] (3/4) Epoch 41, batch 50, loss[loss=0.1441, simple_loss=0.2274, pruned_loss=0.03041, over 19889.00 frames. ], tot_loss[loss=0.1521, simple_loss=0.2336, pruned_loss=0.03533, over 890043.26 frames. ], batch size: 92, lr: 3.83e-03, grad_scale: 16.0
+2023-03-29 14:07:16,134 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0
+2023-03-29 14:08:04,850 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0
+2023-03-29 14:08:22,697 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8009, 5.0591, 5.1012, 4.9789, 4.7844, 5.0681, 4.6189, 4.6033],
+       device='cuda:3'), covar=tensor([0.0447, 0.0433, 0.0422, 0.0426, 0.0593, 0.0452, 0.0622, 0.0910],
+       device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0307, 0.0316, 0.0276, 0.0284, 0.0267, 0.0281, 0.0327],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 14:08:35,542 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:08:49,403 INFO [train.py:892] (3/4) Epoch 41, batch 100, loss[loss=0.1243, simple_loss=0.2025, pruned_loss=0.02309, over 19782.00 frames. ], tot_loss[loss=0.1509, simple_loss=0.2327, pruned_loss=0.0345, over 1568566.89 frames. ], batch size: 94, lr: 3.83e-03, grad_scale: 16.0
+2023-03-29 14:08:50,563 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:09:54,213 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.051e+02 3.488e+02 4.125e+02 4.651e+02 7.521e+02, threshold=8.249e+02, percent-clipped=0.0
+2023-03-29 14:10:05,038 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.83 vs. limit=5.0
+2023-03-29 14:10:38,031 INFO [train.py:892] (3/4) Epoch 41, batch 150, loss[loss=0.1352, simple_loss=0.2115, pruned_loss=0.02942, over 19797.00 frames. ], tot_loss[loss=0.1514, simple_loss=0.2331, pruned_loss=0.03479, over 2094704.59 frames. ], batch size: 126, lr: 3.82e-03, grad_scale: 16.0
+2023-03-29 14:10:41,454 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6203, 3.7093, 2.3169, 3.8200, 3.9375, 1.8866, 3.2584, 3.0760],
+       device='cuda:3'), covar=tensor([0.0747, 0.0840, 0.2594, 0.0765, 0.0598, 0.2638, 0.1135, 0.0896],
+       device='cuda:3'), in_proj_covar=tensor([0.0241, 0.0267, 0.0236, 0.0286, 0.0265, 0.0207, 0.0245, 0.0208],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 14:12:03,289 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74388.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:12:35,706 INFO [train.py:892] (3/4) Epoch 41, batch 200, loss[loss=0.1399, simple_loss=0.2178, pruned_loss=0.03095, over 19491.00 frames. ], tot_loss[loss=0.1515, simple_loss=0.2333, pruned_loss=0.03484, over 2506531.80 frames. ], batch size: 43, lr: 3.82e-03, grad_scale: 16.0
+2023-03-29 14:13:13,310 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0
+2023-03-29 14:13:19,277 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74420.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:13:44,025 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.532e+02 3.589e+02 4.027e+02 4.762e+02 8.323e+02, threshold=8.054e+02, percent-clipped=1.0
+2023-03-29 14:13:55,054 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74436.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:14:17,732 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4564, 5.9016, 5.9759, 5.7658, 5.5897, 5.6516, 5.6868, 5.4990],
+       device='cuda:3'), covar=tensor([0.1269, 0.1125, 0.0693, 0.1084, 0.0557, 0.0709, 0.1551, 0.1687],
+       device='cuda:3'), in_proj_covar=tensor([0.0305, 0.0346, 0.0378, 0.0315, 0.0287, 0.0294, 0.0374, 0.0404],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 14:14:27,266 INFO [train.py:892] (3/4) Epoch 41, batch 250, loss[loss=0.164, simple_loss=0.2439, pruned_loss=0.04202, over 19807.00 frames. ], tot_loss[loss=0.1508, simple_loss=0.2327, pruned_loss=0.03444, over 2827279.31 frames. ], batch size: 86, lr: 3.82e-03, grad_scale: 16.0
+2023-03-29 14:16:20,810 INFO [train.py:892] (3/4) Epoch 41, batch 300, loss[loss=0.1401, simple_loss=0.2207, pruned_loss=0.02979, over 19872.00 frames. ], tot_loss[loss=0.1508, simple_loss=0.2322, pruned_loss=0.03472, over 3075372.16 frames. ], batch size: 138, lr: 3.82e-03, grad_scale: 16.0
+2023-03-29 14:17:27,573 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.493e+02 3.756e+02 4.487e+02 5.432e+02 9.829e+02, threshold=8.974e+02, percent-clipped=6.0
+2023-03-29 14:18:10,549 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0
+2023-03-29 14:18:14,846 INFO [train.py:892] (3/4) Epoch 41, batch 350, loss[loss=0.1536, simple_loss=0.2291, pruned_loss=0.03904, over 19820.00 frames. ], tot_loss[loss=0.1509, simple_loss=0.2326, pruned_loss=0.03458, over 3269083.15 frames. ], batch size: 127, lr: 3.82e-03, grad_scale: 16.0
+2023-03-29 14:19:57,750 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74595.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:19:59,648 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74596.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:20:06,635 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0
+2023-03-29 14:20:09,538 INFO [train.py:892] (3/4) Epoch 41, batch 400, loss[loss=0.1422, simple_loss=0.2159, pruned_loss=0.03424, over 19817.00 frames. ], tot_loss[loss=0.1511, simple_loss=0.2327, pruned_loss=0.03476, over 3420801.80 frames. ], batch size: 133, lr: 3.82e-03, grad_scale: 16.0
+2023-03-29 14:20:45,000 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2835, 3.5972, 3.8208, 4.3002, 2.9355, 3.2304, 2.5745, 2.7003],
+       device='cuda:3'), covar=tensor([0.0515, 0.1818, 0.0839, 0.0404, 0.1904, 0.1100, 0.1435, 0.1601],
+       device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0328, 0.0253, 0.0211, 0.0252, 0.0215, 0.0223, 0.0219],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 14:20:56,884 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-29 14:21:17,920 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.495e+02 3.791e+02 4.431e+02 5.113e+02 8.392e+02, threshold=8.862e+02, percent-clipped=0.0
+2023-03-29 14:21:54,862 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4924, 2.5959, 2.7842, 2.4460, 2.9176, 2.9396, 3.3344, 3.6783],
+       device='cuda:3'), covar=tensor([0.0753, 0.1807, 0.1826, 0.2353, 0.1600, 0.1527, 0.0764, 0.0710],
+       device='cuda:3'), in_proj_covar=tensor([0.0261, 0.0246, 0.0274, 0.0261, 0.0306, 0.0264, 0.0240, 0.0267],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 14:22:06,126 INFO [train.py:892] (3/4) Epoch 41, batch 450, loss[loss=0.1414, simple_loss=0.2304, pruned_loss=0.02617, over 19785.00 frames. ], tot_loss[loss=0.1516, simple_loss=0.2335, pruned_loss=0.03485, over 3537096.37 frames. ], batch size: 94, lr: 3.82e-03, grad_scale: 16.0
+2023-03-29 14:22:18,101 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74656.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:24:01,236 INFO [train.py:892] (3/4) Epoch 41, batch 500, loss[loss=0.1801, simple_loss=0.2945, pruned_loss=0.03282, over 18786.00 frames. ], tot_loss[loss=0.1509, simple_loss=0.2324, pruned_loss=0.03466, over 3627812.11 frames. ], batch size: 564, lr: 3.81e-03, grad_scale: 16.0
+2023-03-29 14:24:47,970 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:25:12,465 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.085e+02 3.392e+02 3.989e+02 4.649e+02 9.968e+02, threshold=7.978e+02, percent-clipped=1.0
+2023-03-29 14:25:55,776 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0354, 4.8580, 5.4354, 4.9152, 4.2811, 5.1413, 5.0810, 5.5931],
+       device='cuda:3'), covar=tensor([0.0810, 0.0390, 0.0361, 0.0384, 0.0857, 0.0467, 0.0447, 0.0293],
+       device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0229, 0.0228, 0.0240, 0.0212, 0.0253, 0.0241, 0.0227],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 14:25:56,899 INFO [train.py:892] (3/4) Epoch 41, batch 550, loss[loss=0.1355, simple_loss=0.2198, pruned_loss=0.02561, over 19820.00 frames. ], tot_loss[loss=0.1502, simple_loss=0.2318, pruned_loss=0.03435, over 3700409.81 frames. ], batch size: 50, lr: 3.81e-03, grad_scale: 16.0
+2023-03-29 14:26:36,050 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74768.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:27:02,491 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5854, 3.3082, 3.5377, 3.0734, 3.7699, 3.7652, 4.3763, 4.8490],
+       device='cuda:3'), covar=tensor([0.0498, 0.1534, 0.1413, 0.2285, 0.1636, 0.1452, 0.0562, 0.0619],
+       device='cuda:3'), in_proj_covar=tensor([0.0263, 0.0248, 0.0276, 0.0263, 0.0309, 0.0266, 0.0241, 0.0269],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 14:27:48,999 INFO [train.py:892] (3/4) Epoch 41, batch 600, loss[loss=0.1457, simple_loss=0.2344, pruned_loss=0.02853, over 19792.00 frames. ], tot_loss[loss=0.1509, simple_loss=0.2321, pruned_loss=0.03479, over 3756423.38 frames. ], batch size: 73, lr: 3.81e-03, grad_scale: 16.0
+2023-03-29 14:28:17,344 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8710, 2.7110, 5.0635, 4.2922, 4.6757, 4.9279, 4.8288, 4.5792],
+       device='cuda:3'), covar=tensor([0.0544, 0.1097, 0.0088, 0.0846, 0.0176, 0.0204, 0.0142, 0.0168],
+       device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0106, 0.0091, 0.0153, 0.0090, 0.0103, 0.0094, 0.0090],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 14:28:53,133 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.538e+02 3.664e+02 4.346e+02 5.320e+02 9.628e+02, threshold=8.692e+02, percent-clipped=3.0
+2023-03-29 14:29:38,244 INFO [train.py:892] (3/4) Epoch 41, batch 650, loss[loss=0.1629, simple_loss=0.2532, pruned_loss=0.03632, over 19661.00 frames. ], tot_loss[loss=0.1514, simple_loss=0.2326, pruned_loss=0.03511, over 3799275.88 frames. ], batch size: 55, lr: 3.81e-03, grad_scale: 16.0
+2023-03-29 14:30:02,670 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8777, 4.0076, 2.3984, 4.1103, 4.3279, 1.9420, 3.5356, 3.2742],
+       device='cuda:3'), covar=tensor([0.0780, 0.0768, 0.2821, 0.0789, 0.0513, 0.2716, 0.1124, 0.0944],
+       device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0268, 0.0237, 0.0288, 0.0266, 0.0209, 0.0246, 0.0208],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 14:31:22,021 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74896.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:31:27,820 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9253, 2.2702, 2.8329, 3.0556, 3.5960, 3.9022, 3.8312, 3.8828],
+       device='cuda:3'), covar=tensor([0.1055, 0.1871, 0.1426, 0.0840, 0.0506, 0.0291, 0.0375, 0.0399],
+       device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0171, 0.0183, 0.0158, 0.0142, 0.0138, 0.0131, 0.0122],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 14:31:33,013 INFO [train.py:892] (3/4) Epoch 41, batch 700, loss[loss=0.1759, simple_loss=0.262, pruned_loss=0.0449, over 19819.00 frames. ], tot_loss[loss=0.1529, simple_loss=0.234, pruned_loss=0.03586, over 3831393.26 frames. ], batch size: 204, lr: 3.81e-03, grad_scale: 16.0
+2023-03-29 14:31:52,062 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74909.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:32:42,151 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.552e+02 3.676e+02 4.188e+02 5.143e+02 8.176e+02, threshold=8.375e+02, percent-clipped=0.0
+2023-03-29 14:33:13,909 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74944.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:33:30,568 INFO [train.py:892] (3/4) Epoch 41, batch 750, loss[loss=0.1464, simple_loss=0.2301, pruned_loss=0.03139, over 19653.00 frames. ], tot_loss[loss=0.152, simple_loss=0.2335, pruned_loss=0.03528, over 3857940.93 frames. ], batch size: 66, lr: 3.81e-03, grad_scale: 16.0
+2023-03-29 14:33:32,200 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:34:02,605 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74964.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:34:15,124 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74970.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:35:26,065 INFO [train.py:892] (3/4) Epoch 41, batch 800, loss[loss=0.179, simple_loss=0.2632, pruned_loss=0.04743, over 19636.00 frames. ], tot_loss[loss=0.1515, simple_loss=0.233, pruned_loss=0.03502, over 3879693.80 frames. ], batch size: 343, lr: 3.81e-03, grad_scale: 16.0
+2023-03-29 14:36:21,801 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75025.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:36:36,274 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.639e+02 3.669e+02 4.402e+02 5.474e+02 1.379e+03, threshold=8.804e+02, percent-clipped=2.0
+2023-03-29 14:37:20,408 INFO [train.py:892] (3/4) Epoch 41, batch 850, loss[loss=0.1811, simple_loss=0.2965, pruned_loss=0.03282, over 18791.00 frames. ], tot_loss[loss=0.1527, simple_loss=0.2338, pruned_loss=0.03581, over 3895157.21 frames. ], batch size: 564, lr: 3.81e-03, grad_scale: 16.0
+2023-03-29 14:39:13,370 INFO [train.py:892] (3/4) Epoch 41, batch 900, loss[loss=0.1441, simple_loss=0.226, pruned_loss=0.03112, over 19805.00 frames. ], tot_loss[loss=0.1526, simple_loss=0.2338, pruned_loss=0.03574, over 3906565.48 frames. ], batch size: 148, lr: 3.80e-03, grad_scale: 16.0
+2023-03-29 14:40:18,832 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8319, 4.0736, 4.4276, 4.9839, 3.1175, 3.6207, 3.1011, 2.9447],
+       device='cuda:3'), covar=tensor([0.0435, 0.1738, 0.0696, 0.0332, 0.2022, 0.1088, 0.1213, 0.1569],
+       device='cuda:3'), in_proj_covar=tensor([0.0253, 0.0330, 0.0255, 0.0212, 0.0252, 0.0216, 0.0224, 0.0221],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 14:40:21,665 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.262e+02 3.745e+02 4.337e+02 5.272e+02 8.502e+02, threshold=8.673e+02, percent-clipped=0.0
+2023-03-29 14:41:05,614 INFO [train.py:892] (3/4) Epoch 41, batch 950, loss[loss=0.1403, simple_loss=0.2256, pruned_loss=0.0275, over 19667.00 frames. ], tot_loss[loss=0.1525, simple_loss=0.2335, pruned_loss=0.03573, over 3916220.95 frames. ], batch size: 55, lr: 3.80e-03, grad_scale: 16.0
+2023-03-29 14:42:59,288 INFO [train.py:892] (3/4) Epoch 41, batch 1000, loss[loss=0.1409, simple_loss=0.2201, pruned_loss=0.03086, over 19678.00 frames. ], tot_loss[loss=0.1517, simple_loss=0.2329, pruned_loss=0.03527, over 3920681.39 frames. ], batch size: 52, lr: 3.80e-03, grad_scale: 16.0
+2023-03-29 14:43:29,316 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6752, 3.8708, 2.4923, 4.3791, 4.0795, 4.3873, 4.4078, 3.3808],
+       device='cuda:3'), covar=tensor([0.0602, 0.0529, 0.1489, 0.0637, 0.0467, 0.0326, 0.0563, 0.0831],
+       device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0151, 0.0146, 0.0161, 0.0139, 0.0144, 0.0156, 0.0154],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 14:44:01,354 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75228.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:44:07,916 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.478e+02 3.475e+02 3.874e+02 4.705e+02 1.241e+03, threshold=7.748e+02, percent-clipped=2.0
+2023-03-29 14:44:46,109 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.1140, 4.6945, 4.7463, 4.4941, 5.0534, 3.1622, 4.0111, 2.6605],
+       device='cuda:3'), covar=tensor([0.0170, 0.0211, 0.0147, 0.0198, 0.0160, 0.1075, 0.0917, 0.1465],
+       device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0153, 0.0118, 0.0140, 0.0123, 0.0140, 0.0146, 0.0132],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003],
+       device='cuda:3')
+2023-03-29 14:44:53,972 INFO [train.py:892] (3/4) Epoch 41, batch 1050, loss[loss=0.1396, simple_loss=0.2274, pruned_loss=0.02586, over 19702.00 frames. ], tot_loss[loss=0.1518, simple_loss=0.233, pruned_loss=0.03531, over 3927239.22 frames. ], batch size: 54, lr: 3.80e-03, grad_scale: 16.0
+2023-03-29 14:44:54,885 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75251.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:45:26,194 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75265.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:45:38,174 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75270.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:46:19,815 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:46:40,146 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:46:43,191 INFO [train.py:892] (3/4) Epoch 41, batch 1100, loss[loss=0.1355, simple_loss=0.2064, pruned_loss=0.03225, over 19790.00 frames. ], tot_loss[loss=0.1516, simple_loss=0.2328, pruned_loss=0.0352, over 3932737.29 frames. ], batch size: 120, lr: 3.80e-03, grad_scale: 16.0
+2023-03-29 14:47:27,614 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:47:28,271 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0
+2023-03-29 14:47:51,239 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.168e+02 3.537e+02 4.080e+02 4.857e+02 1.316e+03, threshold=8.160e+02, percent-clipped=3.0
+2023-03-29 14:47:53,820 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75331.0, num_to_drop=1, layers_to_drop={3}
+2023-03-29 14:47:58,427 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0
+2023-03-29 14:48:07,233 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9221, 3.1771, 2.7606, 2.3790, 2.8589, 3.1351, 3.0851, 3.1287],
+       device='cuda:3'), covar=tensor([0.0313, 0.0339, 0.0335, 0.0549, 0.0345, 0.0293, 0.0250, 0.0226],
+       device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0107, 0.0108, 0.0108, 0.0111, 0.0097, 0.0098, 0.0096],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 14:48:39,022 INFO [train.py:892] (3/4) Epoch 41, batch 1150, loss[loss=0.1868, simple_loss=0.2712, pruned_loss=0.05122, over 19629.00 frames. ], tot_loss[loss=0.1529, simple_loss=0.2343, pruned_loss=0.03577, over 3934085.78 frames. ], batch size: 351, lr: 3.80e-03, grad_scale: 16.0
+2023-03-29 14:50:04,798 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0
+2023-03-29 14:50:16,274 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6997, 4.8059, 2.7873, 5.0612, 5.2590, 2.1715, 4.4389, 3.8317],
+       device='cuda:3'), covar=tensor([0.0545, 0.0589, 0.2454, 0.0504, 0.0435, 0.2629, 0.0823, 0.0810],
+       device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0267, 0.0236, 0.0288, 0.0265, 0.0208, 0.0245, 0.0209],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 14:50:32,218 INFO [train.py:892] (3/4) Epoch 41, batch 1200, loss[loss=0.1593, simple_loss=0.2357, pruned_loss=0.04149, over 19875.00 frames. ], tot_loss[loss=0.1521, simple_loss=0.2333, pruned_loss=0.03541, over 3937118.02 frames. ], batch size: 125, lr: 3.80e-03, grad_scale: 16.0
+2023-03-29 14:50:41,336 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8178, 2.8941, 4.3908, 3.2543, 3.4947, 3.2520, 2.4286, 2.5287],
+       device='cuda:3'), covar=tensor([0.1149, 0.3184, 0.0480, 0.1117, 0.1881, 0.1715, 0.2849, 0.2992],
+       device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0401, 0.0355, 0.0296, 0.0381, 0.0395, 0.0387, 0.0360],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 14:51:41,598 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.627e+02 3.559e+02 4.233e+02 5.176e+02 1.064e+03, threshold=8.467e+02, percent-clipped=2.0
+2023-03-29 14:52:27,172 INFO [train.py:892] (3/4) Epoch 41, batch 1250, loss[loss=0.1392, simple_loss=0.2248, pruned_loss=0.02683, over 19885.00 frames. ], tot_loss[loss=0.1521, simple_loss=0.2336, pruned_loss=0.03531, over 3939443.52 frames. ], batch size: 88, lr: 3.80e-03, grad_scale: 32.0
+2023-03-29 14:54:20,138 INFO [train.py:892] (3/4) Epoch 41, batch 1300, loss[loss=0.1839, simple_loss=0.2753, pruned_loss=0.04629, over 19679.00 frames. ], tot_loss[loss=0.1517, simple_loss=0.2334, pruned_loss=0.03503, over 3940895.93 frames. ], batch size: 55, lr: 3.79e-03, grad_scale: 32.0
+2023-03-29 14:54:52,541 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2099, 4.0948, 4.4616, 4.0633, 3.7938, 4.3233, 4.1561, 4.5153],
+       device='cuda:3'), covar=tensor([0.0735, 0.0326, 0.0326, 0.0371, 0.0978, 0.0487, 0.0475, 0.0343],
+       device='cuda:3'), in_proj_covar=tensor([0.0283, 0.0226, 0.0227, 0.0238, 0.0209, 0.0251, 0.0239, 0.0225],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 14:55:27,414 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.667e+02 3.652e+02 4.285e+02 4.954e+02 9.983e+02, threshold=8.569e+02, percent-clipped=1.0
+2023-03-29 14:56:10,916 INFO [train.py:892] (3/4) Epoch 41, batch 1350, loss[loss=0.1475, simple_loss=0.2341, pruned_loss=0.03044, over 19772.00 frames. ], tot_loss[loss=0.1515, simple_loss=0.2329, pruned_loss=0.03501, over 3943062.73 frames. ], batch size: 247, lr: 3.79e-03, grad_scale: 32.0
+2023-03-29 14:56:43,497 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0
+2023-03-29 14:56:46,565 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75565.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:57:28,137 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:58:05,437 INFO [train.py:892] (3/4) Epoch 41, batch 1400, loss[loss=0.1479, simple_loss=0.2245, pruned_loss=0.03559, over 19768.00 frames. ], tot_loss[loss=0.1515, simple_loss=0.2327, pruned_loss=0.03518, over 3943642.09 frames. ], batch size: 152, lr: 3.79e-03, grad_scale: 32.0
+2023-03-29 14:58:20,262 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5982, 2.0828, 2.4484, 2.9070, 3.2448, 3.3874, 3.2889, 3.3156],
+       device='cuda:3'), covar=tensor([0.1097, 0.1785, 0.1443, 0.0740, 0.0497, 0.0379, 0.0509, 0.0579],
+       device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0170, 0.0184, 0.0157, 0.0142, 0.0139, 0.0132, 0.0123],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 14:58:33,229 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:58:49,454 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75620.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 14:59:04,016 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75626.0, num_to_drop=1, layers_to_drop={2}
+2023-03-29 14:59:14,846 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.057e+02 3.782e+02 4.383e+02 5.452e+02 9.029e+02, threshold=8.766e+02, percent-clipped=1.0
+2023-03-29 15:00:00,084 INFO [train.py:892] (3/4) Epoch 41, batch 1450, loss[loss=0.1522, simple_loss=0.2288, pruned_loss=0.03785, over 19779.00 frames. ], tot_loss[loss=0.1522, simple_loss=0.2332, pruned_loss=0.03553, over 3945519.79 frames. ], batch size: 217, lr: 3.79e-03, grad_scale: 32.0
+2023-03-29 15:00:16,607 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8927, 2.3191, 2.6904, 3.1163, 3.5834, 3.9108, 3.7837, 3.8828],
+       device='cuda:3'), covar=tensor([0.0987, 0.1733, 0.1495, 0.0734, 0.0464, 0.0296, 0.0438, 0.0365],
+       device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0170, 0.0184, 0.0157, 0.0142, 0.0139, 0.0132, 0.0123],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 15:00:40,611 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75668.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 15:01:52,760 INFO [train.py:892] (3/4) Epoch 41, batch 1500, loss[loss=0.1471, simple_loss=0.2315, pruned_loss=0.03139, over 19798.00 frames. ], tot_loss[loss=0.1521, simple_loss=0.2335, pruned_loss=0.03538, over 3946927.72 frames. ], batch size: 79, lr: 3.79e-03, grad_scale: 32.0
+2023-03-29 15:03:00,955 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.571e+02 3.753e+02 4.206e+02 5.021e+02 8.487e+02, threshold=8.413e+02, percent-clipped=0.0
+2023-03-29 15:03:47,903 INFO [train.py:892] (3/4) Epoch 41, batch 1550, loss[loss=0.1322, simple_loss=0.2114, pruned_loss=0.02646, over 19867.00 frames. ], tot_loss[loss=0.1515, simple_loss=0.2332, pruned_loss=0.03493, over 3947478.36 frames. ], batch size: 64, lr: 3.79e-03, grad_scale: 32.0
+2023-03-29 15:05:36,423 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.5658, 5.0545, 5.1808, 4.8541, 5.4669, 3.5681, 4.2685, 2.7104],
+       device='cuda:3'), covar=tensor([0.0157, 0.0212, 0.0144, 0.0200, 0.0136, 0.0843, 0.1013, 0.1470],
+       device='cuda:3'), in_proj_covar=tensor([0.0108, 0.0153, 0.0118, 0.0139, 0.0122, 0.0139, 0.0146, 0.0132],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 15:05:37,510 INFO [train.py:892] (3/4) Epoch 41, batch 1600, loss[loss=0.1402, simple_loss=0.2221, pruned_loss=0.02913, over 19598.00 frames. ], tot_loss[loss=0.1521, simple_loss=0.2337, pruned_loss=0.03525, over 3947268.04 frames. ], batch size: 44, lr: 3.79e-03, grad_scale: 32.0
+2023-03-29 15:06:45,413 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.043e+02 3.453e+02 4.213e+02 4.846e+02 1.121e+03, threshold=8.426e+02, percent-clipped=1.0
+2023-03-29 15:07:30,022 INFO [train.py:892] (3/4) Epoch 41, batch 1650, loss[loss=0.1467, simple_loss=0.2371, pruned_loss=0.02813, over 19851.00 frames. ], tot_loss[loss=0.1512, simple_loss=0.2328, pruned_loss=0.03475, over 3948598.03 frames. ], batch size: 49, lr: 3.79e-03, grad_scale: 32.0
+2023-03-29 15:08:28,883 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.11 vs. limit=2.0
+2023-03-29 15:08:44,838 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75884.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 15:09:22,815 INFO [train.py:892] (3/4) Epoch 41, batch 1700, loss[loss=0.1233, simple_loss=0.2033, pruned_loss=0.02167, over 19756.00 frames. ], tot_loss[loss=0.1519, simple_loss=0.2336, pruned_loss=0.03515, over 3948287.31 frames. ], batch size: 100, lr: 3.78e-03, grad_scale: 32.0
+2023-03-29 15:10:20,802 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75926.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 15:10:32,265 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.390e+02 3.529e+02 4.151e+02 4.845e+02 7.414e+02, threshold=8.303e+02, percent-clipped=0.0
+2023-03-29 15:10:35,834 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75932.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 15:11:15,200 INFO [train.py:892] (3/4) Epoch 41, batch 1750, loss[loss=0.1688, simple_loss=0.27, pruned_loss=0.03387, over 19684.00 frames. ], tot_loss[loss=0.152, simple_loss=0.2339, pruned_loss=0.03508, over 3949186.43 frames. ], batch size: 55, lr: 3.78e-03, grad_scale: 32.0
+2023-03-29 15:12:02,394 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75974.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 15:13:01,539 INFO [train.py:892] (3/4) Epoch 41, batch 1800, loss[loss=0.1476, simple_loss=0.2206, pruned_loss=0.03728, over 19671.00 frames. ], tot_loss[loss=0.1518, simple_loss=0.2335, pruned_loss=0.0351, over 3948246.22 frames. ], batch size: 73, lr: 3.78e-03, grad_scale: 32.0
+2023-03-29 15:13:08,015 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7948, 3.5783, 3.8868, 3.2427, 4.0349, 3.3467, 3.6437, 3.9293],
+       device='cuda:3'), covar=tensor([0.0656, 0.0431, 0.0615, 0.0635, 0.0361, 0.0446, 0.0481, 0.0371],
+       device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0094, 0.0090, 0.0115, 0.0086, 0.0090, 0.0086, 0.0081],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 15:13:57,891 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.757e+02 3.667e+02 4.191e+02 5.134e+02 9.957e+02, threshold=8.381e+02, percent-clipped=3.0
+2023-03-29 15:14:33,402 INFO [train.py:892] (3/4) Epoch 41, batch 1850, loss[loss=0.1502, simple_loss=0.237, pruned_loss=0.03171, over 19834.00 frames. ], tot_loss[loss=0.1522, simple_loss=0.2344, pruned_loss=0.03499, over 3947419.46 frames. ], batch size: 57, lr: 3.78e-03, grad_scale: 32.0
+2023-03-29 15:15:35,916 INFO [train.py:892] (3/4) Epoch 42, batch 0, loss[loss=0.1412, simple_loss=0.213, pruned_loss=0.03467, over 19839.00 frames. ], tot_loss[loss=0.1412, simple_loss=0.213, pruned_loss=0.03467, over 19839.00 frames. ], batch size: 161, lr: 3.73e-03, grad_scale: 32.0
+2023-03-29 15:15:35,916 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-29 15:16:09,088 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.2785, 2.2494, 2.4214, 2.3138, 2.3541, 2.3488, 2.3333, 2.3943],
+       device='cuda:3'), covar=tensor([0.0414, 0.0401, 0.0350, 0.0385, 0.0495, 0.0417, 0.0499, 0.0425],
+       device='cuda:3'), in_proj_covar=tensor([0.0096, 0.0090, 0.0093, 0.0088, 0.0101, 0.0093, 0.0109, 0.0081],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 15:16:12,640 INFO [train.py:926] (3/4) Epoch 42, validation: loss=0.1864, simple_loss=0.2496, pruned_loss=0.06163, over 2883724.00 frames. 
+2023-03-29 15:16:12,642 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-29 15:17:12,453 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7416, 3.5170, 3.7406, 3.0339, 4.0108, 3.2501, 3.4660, 3.8697],
+       device='cuda:3'), covar=tensor([0.0715, 0.0434, 0.0508, 0.0753, 0.0437, 0.0467, 0.0538, 0.0359],
+       device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0093, 0.0090, 0.0114, 0.0085, 0.0089, 0.0086, 0.0081],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 15:18:08,451 INFO [train.py:892] (3/4) Epoch 42, batch 50, loss[loss=0.1472, simple_loss=0.2242, pruned_loss=0.03513, over 19847.00 frames. ], tot_loss[loss=0.1467, simple_loss=0.2273, pruned_loss=0.03304, over 891377.41 frames. ], batch size: 144, lr: 3.73e-03, grad_scale: 32.0
+2023-03-29 15:19:07,915 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.581e+02 3.605e+02 4.013e+02 4.746e+02 9.116e+02, threshold=8.026e+02, percent-clipped=1.0
+2023-03-29 15:20:06,008 INFO [train.py:892] (3/4) Epoch 42, batch 100, loss[loss=0.1512, simple_loss=0.2345, pruned_loss=0.03397, over 19535.00 frames. ], tot_loss[loss=0.1504, simple_loss=0.2314, pruned_loss=0.03474, over 1568287.13 frames. ], batch size: 46, lr: 3.73e-03, grad_scale: 32.0
+2023-03-29 15:21:58,787 INFO [train.py:892] (3/4) Epoch 42, batch 150, loss[loss=0.1423, simple_loss=0.2355, pruned_loss=0.02453, over 19626.00 frames. ], tot_loss[loss=0.1498, simple_loss=0.2316, pruned_loss=0.03405, over 2094705.69 frames. ], batch size: 52, lr: 3.73e-03, grad_scale: 32.0
+2023-03-29 15:22:57,765 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.293e+02 3.477e+02 4.297e+02 5.129e+02 1.202e+03, threshold=8.594e+02, percent-clipped=2.0
+2023-03-29 15:23:17,411 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2014, 3.5105, 3.0333, 2.6331, 3.1713, 3.3908, 3.3512, 3.3975],
+       device='cuda:3'), covar=tensor([0.0363, 0.0261, 0.0320, 0.0604, 0.0351, 0.0292, 0.0288, 0.0233],
+       device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0108, 0.0108, 0.0109, 0.0112, 0.0097, 0.0099, 0.0097],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 15:23:55,629 INFO [train.py:892] (3/4) Epoch 42, batch 200, loss[loss=0.1456, simple_loss=0.2254, pruned_loss=0.03283, over 19768.00 frames. ], tot_loss[loss=0.1495, simple_loss=0.2313, pruned_loss=0.03386, over 2507168.91 frames. ], batch size: 247, lr: 3.73e-03, grad_scale: 32.0
+2023-03-29 15:24:27,155 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76269.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 15:25:30,409 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76294.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 15:25:57,515 INFO [train.py:892] (3/4) Epoch 42, batch 250, loss[loss=0.1778, simple_loss=0.2456, pruned_loss=0.055, over 19819.00 frames. ], tot_loss[loss=0.1504, simple_loss=0.2322, pruned_loss=0.03433, over 2823559.38 frames. ], batch size: 229, lr: 3.73e-03, grad_scale: 32.0
+2023-03-29 15:26:38,896 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76323.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 15:26:56,855 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76330.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 15:26:57,681 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.384e+02 3.381e+02 3.992e+02 4.681e+02 8.709e+02, threshold=7.984e+02, percent-clipped=1.0
+2023-03-29 15:26:58,922 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8613, 3.9410, 2.4415, 4.0972, 4.2421, 1.9225, 3.5292, 3.3395],
+       device='cuda:3'), covar=tensor([0.0777, 0.0852, 0.2735, 0.0869, 0.0634, 0.2837, 0.1148, 0.0888],
+       device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0271, 0.0240, 0.0292, 0.0270, 0.0210, 0.0248, 0.0211],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 15:27:53,400 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0
+2023-03-29 15:27:54,695 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76355.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 15:27:55,584 INFO [train.py:892] (3/4) Epoch 42, batch 300, loss[loss=0.1576, simple_loss=0.246, pruned_loss=0.03459, over 19729.00 frames. ], tot_loss[loss=0.1507, simple_loss=0.2325, pruned_loss=0.03441, over 3073038.96 frames. ], batch size: 269, lr: 3.73e-03, grad_scale: 32.0
+2023-03-29 15:28:02,922 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3975, 4.4915, 2.7095, 4.6936, 4.9193, 2.1341, 4.2153, 3.6604],
+       device='cuda:3'), covar=tensor([0.0664, 0.0703, 0.2651, 0.0711, 0.0556, 0.2848, 0.0887, 0.0854],
+       device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0271, 0.0240, 0.0291, 0.0269, 0.0210, 0.0247, 0.0211],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 15:28:18,364 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8750, 2.7286, 2.8585, 3.0134, 2.8246, 2.7717, 2.8564, 3.0772],
+       device='cuda:3'), covar=tensor([0.0363, 0.0433, 0.0473, 0.0304, 0.0449, 0.0438, 0.0403, 0.0384],
+       device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0092, 0.0094, 0.0088, 0.0102, 0.0094, 0.0110, 0.0082],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 15:29:02,867 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76384.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 15:29:53,495 INFO [train.py:892] (3/4) Epoch 42, batch 350, loss[loss=0.1665, simple_loss=0.2523, pruned_loss=0.04039, over 19762.00 frames. ], tot_loss[loss=0.151, simple_loss=0.2332, pruned_loss=0.03443, over 3266108.25 frames. ], batch size: 198, lr: 3.73e-03, grad_scale: 32.0
+2023-03-29 15:30:29,379 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0784, 5.3016, 5.3749, 5.2441, 5.0908, 5.3141, 4.8378, 4.7815],
+       device='cuda:3'), covar=tensor([0.0467, 0.0475, 0.0479, 0.0437, 0.0604, 0.0527, 0.0658, 0.1006],
+       device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0309, 0.0319, 0.0278, 0.0288, 0.0268, 0.0282, 0.0331],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 15:30:58,173 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.636e+02 3.511e+02 4.196e+02 4.977e+02 1.197e+03, threshold=8.393e+02, percent-clipped=1.0
+2023-03-29 15:31:51,678 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5083, 2.6760, 3.9240, 3.1132, 3.2139, 3.0638, 2.3067, 2.4415],
+       device='cuda:3'), covar=tensor([0.1234, 0.3213, 0.0584, 0.1170, 0.1953, 0.1667, 0.2872, 0.2881],
+       device='cuda:3'), in_proj_covar=tensor([0.0360, 0.0402, 0.0357, 0.0298, 0.0382, 0.0397, 0.0389, 0.0363],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 15:32:05,163 INFO [train.py:892] (3/4) Epoch 42, batch 400, loss[loss=0.1464, simple_loss=0.2243, pruned_loss=0.03421, over 19793.00 frames. ], tot_loss[loss=0.1514, simple_loss=0.2332, pruned_loss=0.0348, over 3416684.07 frames. ], batch size: 162, lr: 3.73e-03, grad_scale: 32.0
+2023-03-29 15:32:14,262 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7396, 4.4154, 4.4713, 4.6592, 4.4501, 4.7792, 4.8000, 4.9914],
+       device='cuda:3'), covar=tensor([0.0636, 0.0401, 0.0504, 0.0359, 0.0624, 0.0446, 0.0404, 0.0286],
+       device='cuda:3'), in_proj_covar=tensor([0.0160, 0.0185, 0.0207, 0.0183, 0.0183, 0.0166, 0.0158, 0.0206],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 15:32:43,282 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0
+2023-03-29 15:34:12,002 INFO [train.py:892] (3/4) Epoch 42, batch 450, loss[loss=0.1512, simple_loss=0.2156, pruned_loss=0.04345, over 19866.00 frames. ], tot_loss[loss=0.1517, simple_loss=0.2332, pruned_loss=0.03507, over 3537128.65 frames. ], batch size: 129, lr: 3.72e-03, grad_scale: 32.0
+2023-03-29 15:34:46,389 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0
+2023-03-29 15:35:17,324 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.526e+02 3.433e+02 4.142e+02 5.148e+02 8.158e+02, threshold=8.284e+02, percent-clipped=0.0
+2023-03-29 15:35:19,186 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0
+2023-03-29 15:35:23,218 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0
+2023-03-29 15:36:19,147 INFO [train.py:892] (3/4) Epoch 42, batch 500, loss[loss=0.1613, simple_loss=0.2422, pruned_loss=0.0402, over 19756.00 frames. ], tot_loss[loss=0.1514, simple_loss=0.2327, pruned_loss=0.03506, over 3629039.31 frames. ], batch size: 256, lr: 3.72e-03, grad_scale: 32.0
+2023-03-29 15:36:38,371 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0
+2023-03-29 15:37:32,134 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9787, 5.0644, 5.3993, 5.1669, 5.2519, 4.9648, 5.1454, 4.9377],
+       device='cuda:3'), covar=tensor([0.1375, 0.1437, 0.0790, 0.1229, 0.0694, 0.0836, 0.1536, 0.1840],
+       device='cuda:3'), in_proj_covar=tensor([0.0306, 0.0349, 0.0383, 0.0316, 0.0291, 0.0295, 0.0371, 0.0403],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 15:38:18,891 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2905, 2.6432, 4.5583, 3.9136, 4.3361, 4.5134, 4.3572, 4.2434],
+       device='cuda:3'), covar=tensor([0.0634, 0.1063, 0.0110, 0.0791, 0.0164, 0.0241, 0.0166, 0.0172],
+       device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0107, 0.0092, 0.0154, 0.0091, 0.0103, 0.0094, 0.0091],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 15:38:21,675 INFO [train.py:892] (3/4) Epoch 42, batch 550, loss[loss=0.144, simple_loss=0.2341, pruned_loss=0.02696, over 19854.00 frames. ], tot_loss[loss=0.1507, simple_loss=0.2323, pruned_loss=0.03453, over 3699170.76 frames. ], batch size: 78, lr: 3.72e-03, grad_scale: 32.0
+2023-03-29 15:38:44,707 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5857, 2.9005, 3.0889, 3.5257, 2.3673, 3.0777, 2.2769, 2.2929],
+       device='cuda:3'), covar=tensor([0.0581, 0.1548, 0.1027, 0.0525, 0.2201, 0.0849, 0.1486, 0.1662],
+       device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0331, 0.0256, 0.0214, 0.0253, 0.0217, 0.0226, 0.0222],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 15:39:05,813 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76625.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 15:39:18,290 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.079e+02 3.230e+02 3.815e+02 4.831e+02 8.297e+02, threshold=7.631e+02, percent-clipped=1.0
+2023-03-29 15:39:33,861 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1175, 2.9166, 3.2360, 2.8071, 3.3734, 3.2670, 3.9204, 4.2948],
+       device='cuda:3'), covar=tensor([0.0599, 0.1752, 0.1543, 0.2204, 0.1612, 0.1549, 0.0669, 0.0586],
+       device='cuda:3'), in_proj_covar=tensor([0.0259, 0.0244, 0.0272, 0.0260, 0.0305, 0.0263, 0.0238, 0.0266],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 15:40:08,071 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 15:40:21,750 INFO [train.py:892] (3/4) Epoch 42, batch 600, loss[loss=0.1581, simple_loss=0.2438, pruned_loss=0.03622, over 19750.00 frames. ], tot_loss[loss=0.1505, simple_loss=0.2325, pruned_loss=0.03427, over 3754000.37 frames. ], batch size: 250, lr: 3.72e-03, grad_scale: 32.0
+2023-03-29 15:41:19,016 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76679.0, num_to_drop=1, layers_to_drop={2}
+2023-03-29 15:41:55,800 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2410, 4.0179, 4.0783, 4.2745, 3.9758, 4.3001, 4.2752, 4.5226],
+       device='cuda:3'), covar=tensor([0.0684, 0.0431, 0.0507, 0.0410, 0.0788, 0.0523, 0.0488, 0.0325],
+       device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0186, 0.0209, 0.0184, 0.0184, 0.0167, 0.0159, 0.0207],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 15:42:25,331 INFO [train.py:892] (3/4) Epoch 42, batch 650, loss[loss=0.1372, simple_loss=0.216, pruned_loss=0.02917, over 19872.00 frames. ], tot_loss[loss=0.151, simple_loss=0.2329, pruned_loss=0.0345, over 3793113.99 frames. ], batch size: 108, lr: 3.72e-03, grad_scale: 32.0
+2023-03-29 15:43:30,260 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.500e+02 3.354e+02 4.128e+02 5.229e+02 1.136e+03, threshold=8.256e+02, percent-clipped=6.0
+2023-03-29 15:44:09,710 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2763, 3.4380, 1.9876, 4.0059, 3.5680, 3.9583, 3.9854, 3.0572],
+       device='cuda:3'), covar=tensor([0.0684, 0.0642, 0.1789, 0.0601, 0.0628, 0.0460, 0.0635, 0.0870],
+       device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0151, 0.0146, 0.0160, 0.0139, 0.0144, 0.0155, 0.0153],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 15:44:32,833 INFO [train.py:892] (3/4) Epoch 42, batch 700, loss[loss=0.1455, simple_loss=0.2291, pruned_loss=0.03099, over 19681.00 frames. ], tot_loss[loss=0.1517, simple_loss=0.2335, pruned_loss=0.03499, over 3828315.85 frames. ], batch size: 52, lr: 3.72e-03, grad_scale: 32.0
+2023-03-29 15:46:12,049 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0
+2023-03-29 15:46:33,837 INFO [train.py:892] (3/4) Epoch 42, batch 750, loss[loss=0.1462, simple_loss=0.2264, pruned_loss=0.03296, over 19853.00 frames. ], tot_loss[loss=0.1508, simple_loss=0.2324, pruned_loss=0.03459, over 3856542.57 frames. ], batch size: 112, lr: 3.72e-03, grad_scale: 32.0
+2023-03-29 15:47:36,724 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.174e+02 3.338e+02 3.927e+02 4.685e+02 7.677e+02, threshold=7.855e+02, percent-clipped=0.0
+2023-03-29 15:47:59,400 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 15:48:42,232 INFO [train.py:892] (3/4) Epoch 42, batch 800, loss[loss=0.1885, simple_loss=0.2714, pruned_loss=0.05278, over 19740.00 frames. ], tot_loss[loss=0.1522, simple_loss=0.2337, pruned_loss=0.03531, over 3876952.08 frames. ], batch size: 291, lr: 3.72e-03, grad_scale: 32.0
+2023-03-29 15:50:35,792 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 15:50:50,211 INFO [train.py:892] (3/4) Epoch 42, batch 850, loss[loss=0.1742, simple_loss=0.2667, pruned_loss=0.04084, over 19662.00 frames. ], tot_loss[loss=0.1527, simple_loss=0.2343, pruned_loss=0.03559, over 3893153.27 frames. ], batch size: 55, lr: 3.71e-03, grad_scale: 32.0
+2023-03-29 15:51:39,933 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76925.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 15:51:54,868 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.468e+02 3.585e+02 3.941e+02 4.611e+02 9.107e+02, threshold=7.882e+02, percent-clipped=2.0
+2023-03-29 15:52:44,094 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 15:52:58,350 INFO [train.py:892] (3/4) Epoch 42, batch 900, loss[loss=0.1659, simple_loss=0.2406, pruned_loss=0.04563, over 19790.00 frames. ], tot_loss[loss=0.1531, simple_loss=0.2347, pruned_loss=0.03576, over 3904543.25 frames. ], batch size: 174, lr: 3.71e-03, grad_scale: 32.0
+2023-03-29 15:53:41,473 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=76973.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 15:53:57,628 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76979.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 15:54:37,716 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0
+2023-03-29 15:54:43,529 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=76998.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 15:55:01,522 INFO [train.py:892] (3/4) Epoch 42, batch 950, loss[loss=0.1728, simple_loss=0.2486, pruned_loss=0.04856, over 19677.00 frames. ], tot_loss[loss=0.1524, simple_loss=0.234, pruned_loss=0.03541, over 3913592.84 frames. ], batch size: 64, lr: 3.71e-03, grad_scale: 32.0
+2023-03-29 15:55:19,915 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0
+2023-03-29 15:55:38,592 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2839, 4.2906, 4.6067, 4.4333, 4.5676, 4.1631, 4.3746, 4.1760],
+       device='cuda:3'), covar=tensor([0.1491, 0.1674, 0.0946, 0.1362, 0.0913, 0.1016, 0.1703, 0.2054],
+       device='cuda:3'), in_proj_covar=tensor([0.0310, 0.0356, 0.0388, 0.0320, 0.0295, 0.0299, 0.0377, 0.0410],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 15:55:45,020 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77027.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 15:55:53,562 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.093e+02 3.626e+02 4.310e+02 4.989e+02 8.639e+02, threshold=8.620e+02, percent-clipped=2.0
+2023-03-29 15:56:11,012 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5509, 3.4657, 3.4294, 3.2108, 3.5394, 2.6338, 2.8450, 1.6713],
+       device='cuda:3'), covar=tensor([0.0249, 0.0272, 0.0179, 0.0228, 0.0185, 0.1464, 0.0657, 0.1939],
+       device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0153, 0.0117, 0.0139, 0.0123, 0.0139, 0.0144, 0.0131],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 15:56:59,670 INFO [train.py:892] (3/4) Epoch 42, batch 1000, loss[loss=0.1479, simple_loss=0.2259, pruned_loss=0.035, over 19801.00 frames. ], tot_loss[loss=0.1521, simple_loss=0.2337, pruned_loss=0.03527, over 3922258.41 frames. ], batch size: 224, lr: 3.71e-03, grad_scale: 32.0
+2023-03-29 15:58:38,011 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77095.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 15:58:46,923 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0
+2023-03-29 15:59:03,088 INFO [train.py:892] (3/4) Epoch 42, batch 1050, loss[loss=0.166, simple_loss=0.2597, pruned_loss=0.03619, over 19865.00 frames. ], tot_loss[loss=0.1518, simple_loss=0.2334, pruned_loss=0.03512, over 3929531.79 frames. ], batch size: 51, lr: 3.71e-03, grad_scale: 32.0
+2023-03-29 15:59:43,612 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1353, 4.0351, 4.4214, 4.0203, 3.7459, 4.2719, 4.0809, 4.4662],
+       device='cuda:3'), covar=tensor([0.0716, 0.0364, 0.0326, 0.0393, 0.1066, 0.0578, 0.0487, 0.0345],
+       device='cuda:3'), in_proj_covar=tensor([0.0286, 0.0229, 0.0229, 0.0241, 0.0211, 0.0253, 0.0242, 0.0228],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 16:00:02,964 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.508e+02 3.618e+02 4.248e+02 5.006e+02 7.833e+02, threshold=8.496e+02, percent-clipped=0.0
+2023-03-29 16:01:07,350 INFO [train.py:892] (3/4) Epoch 42, batch 1100, loss[loss=0.1442, simple_loss=0.2297, pruned_loss=0.02931, over 19595.00 frames. ], tot_loss[loss=0.152, simple_loss=0.2338, pruned_loss=0.03513, over 3933485.05 frames. ], batch size: 45, lr: 3.71e-03, grad_scale: 32.0
+2023-03-29 16:01:08,489 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77156.0, num_to_drop=1, layers_to_drop={3}
+2023-03-29 16:02:45,665 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77195.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:03:12,212 INFO [train.py:892] (3/4) Epoch 42, batch 1150, loss[loss=0.1611, simple_loss=0.233, pruned_loss=0.04463, over 19771.00 frames. ], tot_loss[loss=0.1521, simple_loss=0.234, pruned_loss=0.0351, over 3937935.20 frames. ], batch size: 198, lr: 3.71e-03, grad_scale: 32.0
+2023-03-29 16:03:27,518 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77211.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:04:11,558 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4951, 2.2368, 3.7904, 3.3997, 3.7542, 3.8074, 3.6037, 3.6024],
+       device='cuda:3'), covar=tensor([0.0934, 0.1257, 0.0150, 0.0525, 0.0221, 0.0295, 0.0259, 0.0234],
+       device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0107, 0.0092, 0.0155, 0.0091, 0.0103, 0.0094, 0.0091],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 16:04:16,659 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.545e+02 3.808e+02 4.258e+02 4.963e+02 8.566e+02, threshold=8.515e+02, percent-clipped=1.0
+2023-03-29 16:04:59,997 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0
+2023-03-29 16:05:19,677 INFO [train.py:892] (3/4) Epoch 42, batch 1200, loss[loss=0.1388, simple_loss=0.2206, pruned_loss=0.02851, over 19879.00 frames. ], tot_loss[loss=0.1516, simple_loss=0.2334, pruned_loss=0.0349, over 3941173.03 frames. ], batch size: 88, lr: 3.71e-03, grad_scale: 32.0
+2023-03-29 16:05:59,919 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7087, 3.0238, 2.6348, 2.2232, 2.7596, 2.9205, 2.8845, 2.9480],
+       device='cuda:3'), covar=tensor([0.0386, 0.0327, 0.0358, 0.0629, 0.0387, 0.0338, 0.0342, 0.0280],
+       device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0108, 0.0109, 0.0109, 0.0112, 0.0098, 0.0100, 0.0098],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 16:06:01,904 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:06:04,109 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:06:16,051 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77277.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:06:29,815 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0
+2023-03-29 16:07:30,266 INFO [train.py:892] (3/4) Epoch 42, batch 1250, loss[loss=0.16, simple_loss=0.2337, pruned_loss=0.04314, over 19745.00 frames. ], tot_loss[loss=0.1521, simple_loss=0.2337, pruned_loss=0.03525, over 3942502.18 frames. ], batch size: 134, lr: 3.70e-03, grad_scale: 32.0
+2023-03-29 16:08:30,863 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.435e+02 3.557e+02 4.199e+02 4.989e+02 7.948e+02, threshold=8.398e+02, percent-clipped=0.0
+2023-03-29 16:08:40,262 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77334.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:08:52,542 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77338.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:09:01,671 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9816, 4.7355, 4.7426, 5.0603, 4.7536, 5.2434, 5.1309, 5.2881],
+       device='cuda:3'), covar=tensor([0.0665, 0.0425, 0.0479, 0.0360, 0.0718, 0.0398, 0.0416, 0.0330],
+       device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0188, 0.0210, 0.0186, 0.0185, 0.0168, 0.0160, 0.0209],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 16:09:35,072 INFO [train.py:892] (3/4) Epoch 42, batch 1300, loss[loss=0.1405, simple_loss=0.2224, pruned_loss=0.02923, over 19821.00 frames. ], tot_loss[loss=0.1527, simple_loss=0.2346, pruned_loss=0.03543, over 3943397.42 frames. ], batch size: 75, lr: 3.70e-03, grad_scale: 16.0
+2023-03-29 16:11:05,307 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9951, 4.1296, 4.1749, 4.0209, 3.9731, 4.1172, 3.7237, 3.7130],
+       device='cuda:3'), covar=tensor([0.0571, 0.0569, 0.0545, 0.0531, 0.0692, 0.0542, 0.0673, 0.1101],
+       device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0311, 0.0321, 0.0280, 0.0290, 0.0270, 0.0285, 0.0334],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 16:11:31,852 INFO [train.py:892] (3/4) Epoch 42, batch 1350, loss[loss=0.1495, simple_loss=0.2289, pruned_loss=0.03508, over 19807.00 frames. ], tot_loss[loss=0.1522, simple_loss=0.2339, pruned_loss=0.03527, over 3946072.96 frames. ], batch size: 47, lr: 3.70e-03, grad_scale: 16.0
+2023-03-29 16:12:12,532 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-29 16:12:23,058 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.49 vs. limit=2.0
+2023-03-29 16:12:30,582 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.502e+02 3.333e+02 3.969e+02 4.891e+02 7.952e+02, threshold=7.937e+02, percent-clipped=0.0
+2023-03-29 16:13:20,710 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77451.0, num_to_drop=1, layers_to_drop={3}
+2023-03-29 16:13:34,062 INFO [train.py:892] (3/4) Epoch 42, batch 1400, loss[loss=0.166, simple_loss=0.2575, pruned_loss=0.03723, over 19762.00 frames. ], tot_loss[loss=0.1516, simple_loss=0.2335, pruned_loss=0.03488, over 3944925.61 frames. ], batch size: 321, lr: 3.70e-03, grad_scale: 16.0
+2023-03-29 16:14:41,949 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.56 vs. limit=5.0
+2023-03-29 16:15:10,590 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77495.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:15:36,682 INFO [train.py:892] (3/4) Epoch 42, batch 1450, loss[loss=0.1512, simple_loss=0.2338, pruned_loss=0.03432, over 19685.00 frames. ], tot_loss[loss=0.1511, simple_loss=0.233, pruned_loss=0.03456, over 3947132.48 frames. ], batch size: 82, lr: 3.70e-03, grad_scale: 16.0
+2023-03-29 16:16:40,363 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.038e+02 3.421e+02 4.179e+02 5.316e+02 8.524e+02, threshold=8.358e+02, percent-clipped=3.0
+2023-03-29 16:17:10,158 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77543.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:17:44,835 INFO [train.py:892] (3/4) Epoch 42, batch 1500, loss[loss=0.1676, simple_loss=0.2417, pruned_loss=0.04675, over 19763.00 frames. ], tot_loss[loss=0.1501, simple_loss=0.2317, pruned_loss=0.03424, over 3948963.32 frames. ], batch size: 44, lr: 3.70e-03, grad_scale: 16.0
+2023-03-29 16:18:16,107 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77567.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:18:33,395 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6545, 2.7409, 2.7382, 2.8179, 2.6938, 2.7642, 2.6602, 2.8235],
+       device='cuda:3'), covar=tensor([0.0358, 0.0346, 0.0426, 0.0305, 0.0479, 0.0362, 0.0421, 0.0433],
+       device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0091, 0.0093, 0.0088, 0.0100, 0.0094, 0.0110, 0.0082],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 16:19:13,689 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8142, 3.6072, 3.9285, 2.9841, 3.9635, 3.2796, 3.5421, 3.9256],
+       device='cuda:3'), covar=tensor([0.0599, 0.0430, 0.0456, 0.0740, 0.0527, 0.0446, 0.0488, 0.0343],
+       device='cuda:3'), in_proj_covar=tensor([0.0084, 0.0094, 0.0090, 0.0114, 0.0085, 0.0089, 0.0087, 0.0082],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 16:19:49,136 INFO [train.py:892] (3/4) Epoch 42, batch 1550, loss[loss=0.135, simple_loss=0.2113, pruned_loss=0.02937, over 19789.00 frames. ], tot_loss[loss=0.1497, simple_loss=0.2316, pruned_loss=0.03394, over 3947584.48 frames. ], batch size: 191, lr: 3.70e-03, grad_scale: 16.0
+2023-03-29 16:20:15,280 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77616.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:20:50,234 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77629.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:20:56,270 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.329e+02 3.286e+02 3.992e+02 4.919e+02 1.011e+03, threshold=7.983e+02, percent-clipped=1.0
+2023-03-29 16:20:59,840 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:21:54,229 INFO [train.py:892] (3/4) Epoch 42, batch 1600, loss[loss=0.1679, simple_loss=0.2488, pruned_loss=0.04347, over 19764.00 frames. ], tot_loss[loss=0.1491, simple_loss=0.2309, pruned_loss=0.03364, over 3946854.19 frames. ], batch size: 244, lr: 3.70e-03, grad_scale: 16.0
+2023-03-29 16:22:46,782 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:22:46,827 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77677.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:23:40,086 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9091, 3.0447, 4.4079, 3.3393, 3.5433, 3.4093, 2.4752, 2.7933],
+       device='cuda:3'), covar=tensor([0.1085, 0.3276, 0.0535, 0.1199, 0.1884, 0.1599, 0.2821, 0.2511],
+       device='cuda:3'), in_proj_covar=tensor([0.0357, 0.0401, 0.0357, 0.0298, 0.0380, 0.0396, 0.0388, 0.0363],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 16:23:59,734 INFO [train.py:892] (3/4) Epoch 42, batch 1650, loss[loss=0.1929, simple_loss=0.3172, pruned_loss=0.03433, over 17960.00 frames. ], tot_loss[loss=0.1498, simple_loss=0.2315, pruned_loss=0.03403, over 3945674.82 frames. ], batch size: 633, lr: 3.70e-03, grad_scale: 16.0
+2023-03-29 16:25:04,056 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.346e+02 3.655e+02 4.276e+02 5.251e+02 1.145e+03, threshold=8.553e+02, percent-clipped=2.0
+2023-03-29 16:25:21,200 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77738.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:25:52,478 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77751.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 16:26:01,320 INFO [train.py:892] (3/4) Epoch 42, batch 1700, loss[loss=0.1515, simple_loss=0.2294, pruned_loss=0.0368, over 19761.00 frames. ], tot_loss[loss=0.1499, simple_loss=0.2315, pruned_loss=0.03414, over 3947528.50 frames. ], batch size: 100, lr: 3.69e-03, grad_scale: 16.0
+2023-03-29 16:26:21,987 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.51 vs. limit=2.0
+2023-03-29 16:27:11,487 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2890, 2.6082, 4.4061, 3.8390, 4.2769, 4.3997, 4.2064, 4.1226],
+       device='cuda:3'), covar=tensor([0.0589, 0.1029, 0.0120, 0.0679, 0.0161, 0.0191, 0.0178, 0.0184],
+       device='cuda:3'), in_proj_covar=tensor([0.0104, 0.0107, 0.0092, 0.0154, 0.0091, 0.0103, 0.0094, 0.0091],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 16:27:35,829 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77799.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:27:51,338 INFO [train.py:892] (3/4) Epoch 42, batch 1750, loss[loss=0.1679, simple_loss=0.2561, pruned_loss=0.03986, over 19632.00 frames. ], tot_loss[loss=0.1497, simple_loss=0.2311, pruned_loss=0.03412, over 3948020.12 frames. ], batch size: 359, lr: 3.69e-03, grad_scale: 16.0
+2023-03-29 16:28:04,662 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.7170, 5.9684, 6.0071, 5.8794, 5.7188, 6.0040, 5.3180, 5.4108],
+       device='cuda:3'), covar=tensor([0.0451, 0.0506, 0.0536, 0.0435, 0.0537, 0.0515, 0.0732, 0.1058],
+       device='cuda:3'), in_proj_covar=tensor([0.0294, 0.0312, 0.0322, 0.0281, 0.0291, 0.0272, 0.0286, 0.0336],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 16:28:49,556 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.255e+02 3.364e+02 4.068e+02 4.963e+02 7.822e+02, threshold=8.136e+02, percent-clipped=0.0
+2023-03-29 16:29:41,598 INFO [train.py:892] (3/4) Epoch 42, batch 1800, loss[loss=0.157, simple_loss=0.2427, pruned_loss=0.03561, over 19878.00 frames. ], tot_loss[loss=0.1502, simple_loss=0.2315, pruned_loss=0.03449, over 3948950.30 frames. ], batch size: 61, lr: 3.69e-03, grad_scale: 16.0
+2023-03-29 16:29:42,703 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4891, 3.6027, 2.3681, 4.2050, 3.7541, 4.1812, 4.1799, 3.3374],
+       device='cuda:3'), covar=tensor([0.0603, 0.0639, 0.1499, 0.0578, 0.0737, 0.0408, 0.0658, 0.0806],
+       device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0153, 0.0148, 0.0163, 0.0143, 0.0147, 0.0157, 0.0155],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0003],
+       device='cuda:3')
+2023-03-29 16:30:04,905 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77867.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:31:23,157 INFO [train.py:892] (3/4) Epoch 42, batch 1850, loss[loss=0.1739, simple_loss=0.2602, pruned_loss=0.04381, over 19686.00 frames. ], tot_loss[loss=0.1515, simple_loss=0.2342, pruned_loss=0.03438, over 3947032.17 frames. ], batch size: 56, lr: 3.69e-03, grad_scale: 16.0
+2023-03-29 16:32:27,865 INFO [train.py:892] (3/4) Epoch 43, batch 0, loss[loss=0.1368, simple_loss=0.2128, pruned_loss=0.03039, over 19780.00 frames. ], tot_loss[loss=0.1368, simple_loss=0.2128, pruned_loss=0.03039, over 19780.00 frames. ], batch size: 120, lr: 3.65e-03, grad_scale: 16.0
+2023-03-29 16:32:27,865 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-29 16:33:04,433 INFO [train.py:926] (3/4) Epoch 43, validation: loss=0.1873, simple_loss=0.2496, pruned_loss=0.06254, over 2883724.00 frames. 
+2023-03-29 16:33:04,435 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-29 16:33:16,460 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77915.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:33:39,071 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8938, 2.4041, 3.9430, 3.5299, 3.9622, 3.9887, 3.7665, 3.7583],
+       device='cuda:3'), covar=tensor([0.0665, 0.1039, 0.0122, 0.0517, 0.0170, 0.0258, 0.0196, 0.0199],
+       device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0108, 0.0093, 0.0155, 0.0092, 0.0104, 0.0095, 0.0092],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 16:33:52,653 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77929.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:33:58,730 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.066e+02 3.128e+02 3.651e+02 4.371e+02 8.409e+02, threshold=7.303e+02, percent-clipped=1.0
+2023-03-29 16:34:02,226 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77933.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:35:13,147 INFO [train.py:892] (3/4) Epoch 43, batch 50, loss[loss=0.1579, simple_loss=0.2472, pruned_loss=0.03434, over 19568.00 frames. ], tot_loss[loss=0.1437, simple_loss=0.2237, pruned_loss=0.03181, over 891513.47 frames. ], batch size: 41, lr: 3.65e-03, grad_scale: 16.0
+2023-03-29 16:35:44,858 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77972.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:35:55,993 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77977.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:36:05,061 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77981.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:37:24,670 INFO [train.py:892] (3/4) Epoch 43, batch 100, loss[loss=0.1419, simple_loss=0.2229, pruned_loss=0.03047, over 19396.00 frames. ], tot_loss[loss=0.148, simple_loss=0.2297, pruned_loss=0.03315, over 1568889.53 frames. ], batch size: 40, lr: 3.64e-03, grad_scale: 16.0
+2023-03-29 16:37:32,383 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0
+2023-03-29 16:38:14,010 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.210e+02 3.457e+02 4.175e+02 5.002e+02 8.622e+02, threshold=8.350e+02, percent-clipped=3.0
+2023-03-29 16:38:20,397 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78033.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:39:26,757 INFO [train.py:892] (3/4) Epoch 43, batch 150, loss[loss=0.1713, simple_loss=0.2529, pruned_loss=0.04484, over 19897.00 frames. ], tot_loss[loss=0.1478, simple_loss=0.2297, pruned_loss=0.03297, over 2097886.00 frames. ], batch size: 62, lr: 3.64e-03, grad_scale: 16.0
+2023-03-29 16:41:09,135 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0
+2023-03-29 16:41:35,061 INFO [train.py:892] (3/4) Epoch 43, batch 200, loss[loss=0.146, simple_loss=0.2291, pruned_loss=0.03146, over 19814.00 frames. ], tot_loss[loss=0.1494, simple_loss=0.2314, pruned_loss=0.03376, over 2507472.70 frames. ], batch size: 181, lr: 3.64e-03, grad_scale: 16.0
+2023-03-29 16:42:26,498 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.115e+02 3.506e+02 4.148e+02 4.888e+02 9.722e+02, threshold=8.295e+02, percent-clipped=1.0
+2023-03-29 16:42:50,085 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.85 vs. limit=5.0
+2023-03-29 16:42:51,673 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2405, 3.0883, 3.2663, 2.6720, 3.2747, 2.8491, 3.2499, 3.2649],
+       device='cuda:3'), covar=tensor([0.0579, 0.0488, 0.0530, 0.0731, 0.0449, 0.0461, 0.0440, 0.0348],
+       device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0094, 0.0090, 0.0114, 0.0085, 0.0089, 0.0087, 0.0082],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 16:43:34,487 INFO [train.py:892] (3/4) Epoch 43, batch 250, loss[loss=0.1541, simple_loss=0.2291, pruned_loss=0.03962, over 19806.00 frames. ], tot_loss[loss=0.1494, simple_loss=0.2309, pruned_loss=0.03395, over 2826720.69 frames. ], batch size: 195, lr: 3.64e-03, grad_scale: 16.0
+2023-03-29 16:45:33,967 INFO [train.py:892] (3/4) Epoch 43, batch 300, loss[loss=0.1622, simple_loss=0.2486, pruned_loss=0.03788, over 19764.00 frames. ], tot_loss[loss=0.15, simple_loss=0.2318, pruned_loss=0.0341, over 3073091.40 frames. ], batch size: 244, lr: 3.64e-03, grad_scale: 16.0
+2023-03-29 16:46:28,808 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.098e+02 3.467e+02 4.146e+02 4.960e+02 1.292e+03, threshold=8.293e+02, percent-clipped=2.0
+2023-03-29 16:46:31,145 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78232.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:47:06,764 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2904, 3.6087, 3.8431, 4.3540, 2.8050, 3.3669, 2.7307, 2.7016],
+       device='cuda:3'), covar=tensor([0.0527, 0.1807, 0.0852, 0.0405, 0.2052, 0.0960, 0.1289, 0.1603],
+       device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0326, 0.0253, 0.0212, 0.0250, 0.0215, 0.0223, 0.0220],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 16:47:40,909 INFO [train.py:892] (3/4) Epoch 43, batch 350, loss[loss=0.1591, simple_loss=0.2529, pruned_loss=0.03265, over 19527.00 frames. ], tot_loss[loss=0.1499, simple_loss=0.2313, pruned_loss=0.03422, over 3267693.76 frames. ], batch size: 54, lr: 3.64e-03, grad_scale: 16.0
+2023-03-29 16:48:09,190 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78272.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:48:14,236 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.1011, 1.5742, 1.7129, 2.3501, 2.5259, 2.6628, 2.4921, 2.5976],
+       device='cuda:3'), covar=tensor([0.1296, 0.2009, 0.1886, 0.0898, 0.0699, 0.0499, 0.0608, 0.0595],
+       device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0171, 0.0184, 0.0157, 0.0143, 0.0140, 0.0132, 0.0123],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 16:48:59,404 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78293.0, num_to_drop=1, layers_to_drop={3}
+2023-03-29 16:49:45,890 INFO [train.py:892] (3/4) Epoch 43, batch 400, loss[loss=0.1618, simple_loss=0.2453, pruned_loss=0.03916, over 19666.00 frames. ], tot_loss[loss=0.1507, simple_loss=0.2324, pruned_loss=0.03453, over 3417937.46 frames. ], batch size: 64, lr: 3.64e-03, grad_scale: 16.0
+2023-03-29 16:50:09,461 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78320.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:50:37,822 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.238e+02 3.252e+02 3.876e+02 4.451e+02 9.041e+02, threshold=7.752e+02, percent-clipped=1.0
+2023-03-29 16:50:42,523 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:51:50,607 INFO [train.py:892] (3/4) Epoch 43, batch 450, loss[loss=0.1373, simple_loss=0.217, pruned_loss=0.02876, over 19751.00 frames. ], tot_loss[loss=0.1508, simple_loss=0.2328, pruned_loss=0.03437, over 3535453.08 frames. ], batch size: 44, lr: 3.64e-03, grad_scale: 16.0
+2023-03-29 16:52:43,042 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78381.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:53:55,364 INFO [train.py:892] (3/4) Epoch 43, batch 500, loss[loss=0.1509, simple_loss=0.2296, pruned_loss=0.03608, over 19846.00 frames. ], tot_loss[loss=0.1521, simple_loss=0.2339, pruned_loss=0.03514, over 3625458.36 frames. ], batch size: 112, lr: 3.63e-03, grad_scale: 16.0
+2023-03-29 16:54:44,932 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.422e+02 3.338e+02 4.015e+02 4.887e+02 8.174e+02, threshold=8.030e+02, percent-clipped=1.0
+2023-03-29 16:55:57,309 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78460.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:55:58,409 INFO [train.py:892] (3/4) Epoch 43, batch 550, loss[loss=0.1432, simple_loss=0.2188, pruned_loss=0.03377, over 19840.00 frames. ], tot_loss[loss=0.1517, simple_loss=0.2332, pruned_loss=0.03507, over 3698122.35 frames. ], batch size: 160, lr: 3.63e-03, grad_scale: 16.0
+2023-03-29 16:57:58,314 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0
+2023-03-29 16:58:01,148 INFO [train.py:892] (3/4) Epoch 43, batch 600, loss[loss=0.1371, simple_loss=0.2136, pruned_loss=0.03031, over 19792.00 frames. ], tot_loss[loss=0.1513, simple_loss=0.2326, pruned_loss=0.035, over 3755471.11 frames. ], batch size: 120, lr: 3.63e-03, grad_scale: 16.0
+2023-03-29 16:58:04,746 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0543, 4.1465, 2.3646, 4.3984, 4.5527, 1.9748, 3.8322, 3.3341],
+       device='cuda:3'), covar=tensor([0.0720, 0.0783, 0.2802, 0.0640, 0.0475, 0.2791, 0.0909, 0.0937],
+       device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0269, 0.0240, 0.0290, 0.0270, 0.0209, 0.0246, 0.0211],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 16:58:09,894 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5312, 3.6481, 2.2076, 4.3416, 3.7821, 4.2585, 4.2790, 3.3164],
+       device='cuda:3'), covar=tensor([0.0619, 0.0585, 0.1626, 0.0583, 0.0741, 0.0489, 0.0710, 0.0825],
+       device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0150, 0.0147, 0.0160, 0.0140, 0.0145, 0.0155, 0.0153],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 16:58:12,132 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.1994, 2.1670, 2.3138, 2.2656, 2.2805, 2.3173, 2.2344, 2.3743],
+       device='cuda:3'), covar=tensor([0.0450, 0.0419, 0.0390, 0.0386, 0.0497, 0.0389, 0.0571, 0.0366],
+       device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0091, 0.0094, 0.0088, 0.0100, 0.0094, 0.0109, 0.0082],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 16:58:31,028 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78521.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 16:58:40,723 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9004, 3.7944, 3.7303, 3.5468, 3.8872, 2.8027, 3.2162, 1.9352],
+       device='cuda:3'), covar=tensor([0.0204, 0.0239, 0.0168, 0.0203, 0.0162, 0.1176, 0.0585, 0.1574],
+       device='cuda:3'), in_proj_covar=tensor([0.0109, 0.0152, 0.0117, 0.0139, 0.0123, 0.0138, 0.0145, 0.0131],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 16:58:57,278 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.184e+02 3.257e+02 4.249e+02 5.407e+02 1.207e+03, threshold=8.497e+02, percent-clipped=3.0
+2023-03-29 16:59:53,665 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78556.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:00:06,556 INFO [train.py:892] (3/4) Epoch 43, batch 650, loss[loss=0.1298, simple_loss=0.215, pruned_loss=0.02228, over 19894.00 frames. ], tot_loss[loss=0.1515, simple_loss=0.2326, pruned_loss=0.03517, over 3798135.89 frames. ], batch size: 92, lr: 3.63e-03, grad_scale: 16.0
+2023-03-29 17:01:08,217 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78588.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 17:02:02,236 INFO [train.py:892] (3/4) Epoch 43, batch 700, loss[loss=0.1817, simple_loss=0.2973, pruned_loss=0.03305, over 18758.00 frames. ], tot_loss[loss=0.1516, simple_loss=0.2335, pruned_loss=0.03486, over 3829397.78 frames. ], batch size: 564, lr: 3.63e-03, grad_scale: 16.0
+2023-03-29 17:02:16,609 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78617.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:02:27,920 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2545, 3.5579, 3.0778, 2.6753, 3.1512, 3.4908, 3.3874, 3.4896],
+       device='cuda:3'), covar=tensor([0.0297, 0.0347, 0.0314, 0.0531, 0.0351, 0.0252, 0.0238, 0.0231],
+       device='cuda:3'), in_proj_covar=tensor([0.0114, 0.0107, 0.0109, 0.0108, 0.0111, 0.0097, 0.0099, 0.0097],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 17:02:54,405 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.290e+02 3.555e+02 4.216e+02 4.848e+02 9.327e+02, threshold=8.432e+02, percent-clipped=1.0
+2023-03-29 17:03:25,494 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0535, 3.0525, 4.5366, 3.4965, 3.6127, 3.4702, 2.4413, 2.7967],
+       device='cuda:3'), covar=tensor([0.0942, 0.3077, 0.0508, 0.1080, 0.1933, 0.1642, 0.2810, 0.2626],
+       device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0403, 0.0357, 0.0297, 0.0381, 0.0398, 0.0390, 0.0364],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 17:03:59,434 INFO [train.py:892] (3/4) Epoch 43, batch 750, loss[loss=0.1381, simple_loss=0.2159, pruned_loss=0.03013, over 19595.00 frames. ], tot_loss[loss=0.1512, simple_loss=0.233, pruned_loss=0.03464, over 3856385.84 frames. ], batch size: 45, lr: 3.63e-03, grad_scale: 16.0
+2023-03-29 17:04:46,013 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6931, 2.7585, 2.8408, 2.8933, 2.7211, 2.8445, 2.7974, 2.9476],
+       device='cuda:3'), covar=tensor([0.0387, 0.0383, 0.0377, 0.0283, 0.0422, 0.0388, 0.0391, 0.0376],
+       device='cuda:3'), in_proj_covar=tensor([0.0098, 0.0092, 0.0095, 0.0089, 0.0101, 0.0094, 0.0110, 0.0082],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 17:05:55,902 INFO [train.py:892] (3/4) Epoch 43, batch 800, loss[loss=0.14, simple_loss=0.2176, pruned_loss=0.03114, over 19805.00 frames. ], tot_loss[loss=0.1512, simple_loss=0.2332, pruned_loss=0.03457, over 3876536.24 frames. ], batch size: 167, lr: 3.63e-03, grad_scale: 16.0
+2023-03-29 17:06:42,329 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.297e+02 3.365e+02 3.945e+02 4.793e+02 1.520e+03, threshold=7.889e+02, percent-clipped=1.0
+2023-03-29 17:07:52,458 INFO [train.py:892] (3/4) Epoch 43, batch 850, loss[loss=0.1451, simple_loss=0.2209, pruned_loss=0.03469, over 19755.00 frames. ], tot_loss[loss=0.1508, simple_loss=0.2331, pruned_loss=0.03425, over 3891651.88 frames. ], batch size: 188, lr: 3.63e-03, grad_scale: 16.0
+2023-03-29 17:09:51,551 INFO [train.py:892] (3/4) Epoch 43, batch 900, loss[loss=0.1481, simple_loss=0.2279, pruned_loss=0.03411, over 19810.00 frames. ], tot_loss[loss=0.1505, simple_loss=0.2328, pruned_loss=0.03414, over 3904495.60 frames. ], batch size: 231, lr: 3.63e-03, grad_scale: 16.0
+2023-03-29 17:10:06,750 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78816.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:10:39,610 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.520e+02 3.462e+02 4.060e+02 5.051e+02 1.168e+03, threshold=8.120e+02, percent-clipped=4.0
+2023-03-29 17:11:49,241 INFO [train.py:892] (3/4) Epoch 43, batch 950, loss[loss=0.1458, simple_loss=0.2335, pruned_loss=0.0291, over 19887.00 frames. ], tot_loss[loss=0.1496, simple_loss=0.2319, pruned_loss=0.03365, over 3914946.85 frames. ], batch size: 97, lr: 3.62e-03, grad_scale: 16.0
+2023-03-29 17:12:56,177 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78888.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:13:45,656 INFO [train.py:892] (3/4) Epoch 43, batch 1000, loss[loss=0.1471, simple_loss=0.225, pruned_loss=0.03457, over 19759.00 frames. ], tot_loss[loss=0.1501, simple_loss=0.2324, pruned_loss=0.03389, over 3921058.71 frames. ], batch size: 188, lr: 3.62e-03, grad_scale: 16.0
+2023-03-29 17:13:50,603 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78912.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:14:06,953 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2999, 3.3718, 2.1464, 3.4571, 3.5545, 1.7806, 3.0015, 2.8318],
+       device='cuda:3'), covar=tensor([0.0901, 0.0921, 0.2800, 0.0974, 0.0768, 0.2672, 0.1166, 0.1023],
+       device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0269, 0.0239, 0.0289, 0.0270, 0.0208, 0.0245, 0.0211],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 17:14:35,073 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.508e+02 3.436e+02 3.973e+02 4.908e+02 9.421e+02, threshold=7.947e+02, percent-clipped=2.0
+2023-03-29 17:14:44,414 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:15:39,038 INFO [train.py:892] (3/4) Epoch 43, batch 1050, loss[loss=0.1532, simple_loss=0.2391, pruned_loss=0.03371, over 19623.00 frames. ], tot_loss[loss=0.15, simple_loss=0.2321, pruned_loss=0.03391, over 3928438.47 frames. ], batch size: 65, lr: 3.62e-03, grad_scale: 16.0
+2023-03-29 17:15:48,961 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78965.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:17:33,144 INFO [train.py:892] (3/4) Epoch 43, batch 1100, loss[loss=0.118, simple_loss=0.1973, pruned_loss=0.01935, over 19859.00 frames. ], tot_loss[loss=0.1504, simple_loss=0.2326, pruned_loss=0.03414, over 3932487.29 frames. ], batch size: 46, lr: 3.62e-03, grad_scale: 16.0
+2023-03-29 17:17:42,462 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.7895, 2.7922, 1.8961, 3.1157, 2.9280, 3.0450, 3.1560, 2.5972],
+       device='cuda:3'), covar=tensor([0.0734, 0.0847, 0.1513, 0.0735, 0.0688, 0.0625, 0.0650, 0.0950],
+       device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0152, 0.0149, 0.0163, 0.0142, 0.0147, 0.0158, 0.0155],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0003],
+       device='cuda:3')
+2023-03-29 17:17:58,013 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5549, 3.7982, 4.0728, 4.5842, 3.0015, 3.5343, 2.7890, 2.8646],
+       device='cuda:3'), covar=tensor([0.0447, 0.1687, 0.0745, 0.0358, 0.1973, 0.1006, 0.1303, 0.1606],
+       device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0326, 0.0254, 0.0213, 0.0249, 0.0216, 0.0224, 0.0220],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 17:18:10,091 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79026.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:18:23,422 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.685e+02 3.705e+02 4.386e+02 5.336e+02 1.015e+03, threshold=8.773e+02, percent-clipped=4.0
+2023-03-29 17:19:32,191 INFO [train.py:892] (3/4) Epoch 43, batch 1150, loss[loss=0.1661, simple_loss=0.2519, pruned_loss=0.04012, over 19688.00 frames. ], tot_loss[loss=0.1512, simple_loss=0.2329, pruned_loss=0.03477, over 3935927.41 frames. ], batch size: 64, lr: 3.62e-03, grad_scale: 16.0
+2023-03-29 17:21:27,679 INFO [train.py:892] (3/4) Epoch 43, batch 1200, loss[loss=0.164, simple_loss=0.2431, pruned_loss=0.04251, over 19788.00 frames. ], tot_loss[loss=0.1507, simple_loss=0.2323, pruned_loss=0.03452, over 3940810.59 frames. ], batch size: 247, lr: 3.62e-03, grad_scale: 16.0
+2023-03-29 17:21:40,644 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:21:40,933 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0444, 3.3807, 3.8124, 3.2817, 4.1417, 4.0981, 4.6889, 5.3062],
+       device='cuda:3'), covar=tensor([0.0354, 0.1408, 0.1218, 0.2025, 0.1271, 0.1182, 0.0531, 0.0289],
+       device='cuda:3'), in_proj_covar=tensor([0.0263, 0.0247, 0.0275, 0.0263, 0.0307, 0.0267, 0.0241, 0.0272],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 17:22:16,933 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.729e+02 3.557e+02 4.069e+02 4.739e+02 8.583e+02, threshold=8.137e+02, percent-clipped=0.0
+2023-03-29 17:22:27,468 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.08 vs. limit=2.0
+2023-03-29 17:23:25,185 INFO [train.py:892] (3/4) Epoch 43, batch 1250, loss[loss=0.1552, simple_loss=0.2419, pruned_loss=0.03428, over 19573.00 frames. ], tot_loss[loss=0.1509, simple_loss=0.2325, pruned_loss=0.03465, over 3942013.31 frames. ], batch size: 53, lr: 3.62e-03, grad_scale: 16.0
+2023-03-29 17:23:32,537 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79164.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:24:11,809 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7255, 3.5235, 3.8288, 2.8780, 3.9332, 3.2611, 3.5967, 3.9942],
+       device='cuda:3'), covar=tensor([0.0727, 0.0437, 0.0542, 0.0817, 0.0477, 0.0472, 0.0487, 0.0264],
+       device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0095, 0.0092, 0.0115, 0.0086, 0.0090, 0.0087, 0.0082],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 17:25:19,896 INFO [train.py:892] (3/4) Epoch 43, batch 1300, loss[loss=0.1321, simple_loss=0.2157, pruned_loss=0.02426, over 19706.00 frames. ], tot_loss[loss=0.1498, simple_loss=0.2313, pruned_loss=0.03419, over 3943928.97 frames. ], batch size: 78, lr: 3.62e-03, grad_scale: 16.0
+2023-03-29 17:25:23,205 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79212.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:26:08,979 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.005e+02 3.409e+02 4.094e+02 5.234e+02 1.153e+03, threshold=8.189e+02, percent-clipped=5.0
+2023-03-29 17:27:14,216 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:27:15,467 INFO [train.py:892] (3/4) Epoch 43, batch 1350, loss[loss=0.168, simple_loss=0.2499, pruned_loss=0.043, over 19788.00 frames. ], tot_loss[loss=0.1498, simple_loss=0.2315, pruned_loss=0.03407, over 3944976.46 frames. ], batch size: 211, lr: 3.62e-03, grad_scale: 16.0
+2023-03-29 17:28:09,918 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79285.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 17:29:03,776 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.9991, 1.8648, 2.0334, 2.0489, 2.0025, 2.0869, 1.9600, 2.0895],
+       device='cuda:3'), covar=tensor([0.0456, 0.0412, 0.0387, 0.0379, 0.0535, 0.0345, 0.0508, 0.0387],
+       device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0091, 0.0094, 0.0088, 0.0100, 0.0094, 0.0110, 0.0082],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 17:29:06,868 INFO [train.py:892] (3/4) Epoch 43, batch 1400, loss[loss=0.1284, simple_loss=0.2094, pruned_loss=0.0237, over 19872.00 frames. ], tot_loss[loss=0.1499, simple_loss=0.2314, pruned_loss=0.03417, over 3946852.55 frames. ], batch size: 108, lr: 3.61e-03, grad_scale: 16.0
+2023-03-29 17:29:31,622 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79321.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:29:56,378 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.584e+02 3.305e+02 4.038e+02 5.143e+02 6.996e+02, threshold=8.075e+02, percent-clipped=0.0
+2023-03-29 17:29:59,799 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79333.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:30:19,840 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0
+2023-03-29 17:30:29,655 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79346.0, num_to_drop=1, layers_to_drop={2}
+2023-03-29 17:30:49,367 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4592, 3.7023, 3.2544, 2.8003, 3.2690, 3.6413, 3.5679, 3.6542],
+       device='cuda:3'), covar=tensor([0.0253, 0.0293, 0.0288, 0.0473, 0.0315, 0.0272, 0.0225, 0.0193],
+       device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0109, 0.0109, 0.0109, 0.0112, 0.0098, 0.0100, 0.0098],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 17:31:04,342 INFO [train.py:892] (3/4) Epoch 43, batch 1450, loss[loss=0.1367, simple_loss=0.2107, pruned_loss=0.03131, over 19870.00 frames. ], tot_loss[loss=0.1492, simple_loss=0.231, pruned_loss=0.03372, over 3949281.20 frames. ], batch size: 158, lr: 3.61e-03, grad_scale: 32.0
+2023-03-29 17:31:13,046 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79364.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:31:37,449 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79375.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 17:31:55,649 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0
+2023-03-29 17:32:03,270 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9287, 2.4567, 4.0357, 3.5954, 3.9616, 4.0636, 3.8362, 3.7914],
+       device='cuda:3'), covar=tensor([0.0646, 0.1053, 0.0126, 0.0564, 0.0176, 0.0266, 0.0196, 0.0199],
+       device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0107, 0.0093, 0.0155, 0.0091, 0.0105, 0.0095, 0.0091],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 17:32:20,048 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79394.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:32:59,406 INFO [train.py:892] (3/4) Epoch 43, batch 1500, loss[loss=0.1401, simple_loss=0.2159, pruned_loss=0.03213, over 19767.00 frames. ], tot_loss[loss=0.1504, simple_loss=0.2322, pruned_loss=0.03429, over 3947713.09 frames. ], batch size: 130, lr: 3.61e-03, grad_scale: 32.0
+2023-03-29 17:33:29,257 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:33:41,754 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.542e+02 3.517e+02 4.248e+02 5.342e+02 8.488e+02, threshold=8.496e+02, percent-clipped=1.0
+2023-03-29 17:33:52,861 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79436.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 17:34:48,458 INFO [train.py:892] (3/4) Epoch 43, batch 1550, loss[loss=0.1466, simple_loss=0.2239, pruned_loss=0.03467, over 19879.00 frames. ], tot_loss[loss=0.15, simple_loss=0.2316, pruned_loss=0.03419, over 3948803.18 frames. ], batch size: 134, lr: 3.61e-03, grad_scale: 32.0
+2023-03-29 17:35:33,632 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3511, 3.1324, 3.3740, 2.4771, 3.3969, 2.9120, 3.2643, 3.4386],
+       device='cuda:3'), covar=tensor([0.0532, 0.0492, 0.0612, 0.0967, 0.0431, 0.0511, 0.0411, 0.0299],
+       device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0094, 0.0091, 0.0114, 0.0085, 0.0089, 0.0086, 0.0081],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 17:35:49,861 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6690, 4.6963, 5.0317, 4.8200, 4.9106, 4.4301, 4.7595, 4.6032],
+       device='cuda:3'), covar=tensor([0.1382, 0.1751, 0.0805, 0.1272, 0.0806, 0.1052, 0.1787, 0.1961],
+       device='cuda:3'), in_proj_covar=tensor([0.0309, 0.0360, 0.0386, 0.0322, 0.0296, 0.0302, 0.0380, 0.0410],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 17:36:45,452 INFO [train.py:892] (3/4) Epoch 43, batch 1600, loss[loss=0.1376, simple_loss=0.2132, pruned_loss=0.03098, over 19549.00 frames. ], tot_loss[loss=0.15, simple_loss=0.2313, pruned_loss=0.03436, over 3949334.05 frames. ], batch size: 47, lr: 3.61e-03, grad_scale: 32.0
+2023-03-29 17:37:34,254 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.484e+02 3.446e+02 3.965e+02 4.827e+02 6.779e+02, threshold=7.930e+02, percent-clipped=0.0
+2023-03-29 17:38:42,121 INFO [train.py:892] (3/4) Epoch 43, batch 1650, loss[loss=0.1566, simple_loss=0.2321, pruned_loss=0.04054, over 19840.00 frames. ], tot_loss[loss=0.1494, simple_loss=0.2308, pruned_loss=0.03402, over 3950394.29 frames. ], batch size: 190, lr: 3.61e-03, grad_scale: 32.0
+2023-03-29 17:39:40,533 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79585.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:39:57,749 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9885, 2.4426, 3.1031, 3.2190, 3.7243, 4.0784, 3.8553, 3.9521],
+       device='cuda:3'), covar=tensor([0.0990, 0.1784, 0.1253, 0.0727, 0.0428, 0.0237, 0.0384, 0.0531],
+       device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0172, 0.0184, 0.0158, 0.0143, 0.0139, 0.0133, 0.0123],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 17:40:30,187 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5672, 4.7683, 4.8175, 4.7170, 4.5384, 4.7766, 4.3477, 4.3485],
+       device='cuda:3'), covar=tensor([0.0502, 0.0519, 0.0524, 0.0441, 0.0613, 0.0510, 0.0655, 0.0978],
+       device='cuda:3'), in_proj_covar=tensor([0.0294, 0.0312, 0.0322, 0.0280, 0.0290, 0.0274, 0.0285, 0.0335],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 17:40:37,524 INFO [train.py:892] (3/4) Epoch 43, batch 1700, loss[loss=0.1482, simple_loss=0.2221, pruned_loss=0.03716, over 19838.00 frames. ], tot_loss[loss=0.1504, simple_loss=0.2318, pruned_loss=0.03446, over 3950253.95 frames. ], batch size: 137, lr: 3.61e-03, grad_scale: 32.0
+2023-03-29 17:41:03,074 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79621.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:41:26,804 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.071e+02 3.435e+02 4.157e+02 5.071e+02 8.490e+02, threshold=8.314e+02, percent-clipped=3.0
+2023-03-29 17:41:49,585 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79641.0, num_to_drop=1, layers_to_drop={3}
+2023-03-29 17:41:58,829 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0
+2023-03-29 17:42:00,383 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79646.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:42:21,746 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:42:28,985 INFO [train.py:892] (3/4) Epoch 43, batch 1750, loss[loss=0.1327, simple_loss=0.216, pruned_loss=0.02467, over 19655.00 frames. ], tot_loss[loss=0.1504, simple_loss=0.2321, pruned_loss=0.0344, over 3950166.85 frames. ], batch size: 43, lr: 3.61e-03, grad_scale: 32.0
+2023-03-29 17:42:46,829 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79669.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:43:30,081 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:44:13,216 INFO [train.py:892] (3/4) Epoch 43, batch 1800, loss[loss=0.1505, simple_loss=0.2361, pruned_loss=0.03244, over 19764.00 frames. ], tot_loss[loss=0.1504, simple_loss=0.2323, pruned_loss=0.03429, over 3949093.76 frames. ], batch size: 244, lr: 3.61e-03, grad_scale: 32.0
+2023-03-29 17:44:27,565 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:44:31,253 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79720.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:44:52,873 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79731.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 17:44:53,948 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.513e+02 3.637e+02 4.285e+02 5.199e+02 9.167e+02, threshold=8.570e+02, percent-clipped=2.0
+2023-03-29 17:45:29,577 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6373, 3.5046, 3.8091, 2.9518, 3.9947, 3.3332, 3.6102, 3.7514],
+       device='cuda:3'), covar=tensor([0.0683, 0.0413, 0.0701, 0.0760, 0.0369, 0.0404, 0.0394, 0.0403],
+       device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0095, 0.0091, 0.0115, 0.0086, 0.0090, 0.0087, 0.0082],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 17:45:48,623 INFO [train.py:892] (3/4) Epoch 43, batch 1850, loss[loss=0.1552, simple_loss=0.2499, pruned_loss=0.03028, over 19673.00 frames. ], tot_loss[loss=0.152, simple_loss=0.2343, pruned_loss=0.0349, over 3949012.04 frames. ], batch size: 55, lr: 3.60e-03, grad_scale: 32.0
+2023-03-29 17:45:50,986 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79762.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:46:59,246 INFO [train.py:892] (3/4) Epoch 44, batch 0, loss[loss=0.1336, simple_loss=0.212, pruned_loss=0.02759, over 19566.00 frames. ], tot_loss[loss=0.1336, simple_loss=0.212, pruned_loss=0.02759, over 19566.00 frames. ], batch size: 53, lr: 3.56e-03, grad_scale: 32.0
+2023-03-29 17:46:59,247 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-29 17:47:18,876 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0705, 4.0971, 2.4173, 4.2577, 4.4327, 2.0323, 3.7228, 3.3638],
+       device='cuda:3'), covar=tensor([0.0677, 0.0782, 0.3010, 0.0764, 0.0657, 0.2819, 0.1004, 0.0931],
+       device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0271, 0.0241, 0.0293, 0.0272, 0.0210, 0.0247, 0.0212],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 17:47:25,136 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8442, 2.7770, 5.0052, 4.1864, 4.7195, 4.9246, 4.6382, 4.6181],
+       device='cuda:3'), covar=tensor([0.0553, 0.1117, 0.0096, 0.0753, 0.0146, 0.0185, 0.0165, 0.0160],
+       device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0108, 0.0093, 0.0156, 0.0091, 0.0104, 0.0095, 0.0091],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 17:47:38,449 INFO [train.py:926] (3/4) Epoch 44, validation: loss=0.1877, simple_loss=0.2498, pruned_loss=0.06277, over 2883724.00 frames. 
+2023-03-29 17:47:38,450 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-29 17:49:41,928 INFO [train.py:892] (3/4) Epoch 44, batch 50, loss[loss=0.1607, simple_loss=0.2395, pruned_loss=0.0409, over 19671.00 frames. ], tot_loss[loss=0.1475, simple_loss=0.2286, pruned_loss=0.0332, over 891357.08 frames. ], batch size: 64, lr: 3.56e-03, grad_scale: 32.0
+2023-03-29 17:49:58,356 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79823.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:50:15,367 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.212e+02 3.427e+02 4.163e+02 5.028e+02 9.332e+02, threshold=8.326e+02, percent-clipped=1.0
+2023-03-29 17:51:32,084 INFO [train.py:892] (3/4) Epoch 44, batch 100, loss[loss=0.1746, simple_loss=0.2902, pruned_loss=0.02955, over 18715.00 frames. ], tot_loss[loss=0.1444, simple_loss=0.2258, pruned_loss=0.03151, over 1568932.65 frames. ], batch size: 564, lr: 3.56e-03, grad_scale: 32.0
+2023-03-29 17:51:58,278 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0
+2023-03-29 17:52:10,487 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.11 vs. limit=2.0
+2023-03-29 17:52:12,068 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7110, 3.8140, 4.0592, 3.7476, 4.0062, 3.7458, 3.7800, 3.5516],
+       device='cuda:3'), covar=tensor([0.1450, 0.1706, 0.0922, 0.1516, 0.1211, 0.0990, 0.1910, 0.2270],
+       device='cuda:3'), in_proj_covar=tensor([0.0306, 0.0354, 0.0382, 0.0317, 0.0293, 0.0297, 0.0375, 0.0406],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 17:52:14,526 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.7597, 1.6256, 1.7968, 1.8575, 1.7886, 1.8385, 1.6708, 1.8843],
+       device='cuda:3'), covar=tensor([0.0442, 0.0444, 0.0428, 0.0349, 0.0479, 0.0364, 0.0542, 0.0336],
+       device='cuda:3'), in_proj_covar=tensor([0.0097, 0.0091, 0.0094, 0.0088, 0.0100, 0.0094, 0.0109, 0.0082],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 17:53:03,752 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7026, 3.4676, 3.5777, 3.7234, 3.5170, 3.7804, 3.7091, 3.9330],
+       device='cuda:3'), covar=tensor([0.0858, 0.0679, 0.0657, 0.0561, 0.0856, 0.0706, 0.0765, 0.0506],
+       device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0187, 0.0208, 0.0186, 0.0184, 0.0167, 0.0161, 0.0207],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 17:53:33,280 INFO [train.py:892] (3/4) Epoch 44, batch 150, loss[loss=0.136, simple_loss=0.2186, pruned_loss=0.02671, over 19851.00 frames. ], tot_loss[loss=0.1476, simple_loss=0.2289, pruned_loss=0.03311, over 2097182.93 frames. ], batch size: 104, lr: 3.56e-03, grad_scale: 32.0
+2023-03-29 17:54:11,532 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.056e+02 3.280e+02 3.825e+02 4.551e+02 8.712e+02, threshold=7.650e+02, percent-clipped=2.0
+2023-03-29 17:54:35,915 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79941.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:54:35,957 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79941.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 17:54:37,983 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2575, 2.8578, 3.2862, 3.4101, 3.9031, 4.3792, 4.1230, 4.3748],
+       device='cuda:3'), covar=tensor([0.0911, 0.1544, 0.1263, 0.0667, 0.0429, 0.0224, 0.0379, 0.0356],
+       device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0171, 0.0183, 0.0156, 0.0142, 0.0138, 0.0133, 0.0122],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 17:55:19,639 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5877, 2.9445, 3.0647, 3.5297, 2.4317, 3.0388, 2.2122, 2.2365],
+       device='cuda:3'), covar=tensor([0.0607, 0.1654, 0.1109, 0.0524, 0.2159, 0.0871, 0.1533, 0.1679],
+       device='cuda:3'), in_proj_covar=tensor([0.0254, 0.0329, 0.0256, 0.0214, 0.0252, 0.0217, 0.0225, 0.0221],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 17:55:37,714 INFO [train.py:892] (3/4) Epoch 44, batch 200, loss[loss=0.1578, simple_loss=0.2368, pruned_loss=0.03939, over 19824.00 frames. ], tot_loss[loss=0.1479, simple_loss=0.2296, pruned_loss=0.03309, over 2508466.44 frames. ], batch size: 76, lr: 3.56e-03, grad_scale: 32.0
+2023-03-29 17:56:34,764 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79989.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 17:56:34,809 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:57:33,037 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80013.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:57:40,470 INFO [train.py:892] (3/4) Epoch 44, batch 250, loss[loss=0.1309, simple_loss=0.2143, pruned_loss=0.02373, over 19915.00 frames. ], tot_loss[loss=0.1484, simple_loss=0.2304, pruned_loss=0.03319, over 2825622.71 frames. ], batch size: 45, lr: 3.56e-03, grad_scale: 16.0
+2023-03-29 17:57:53,847 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80020.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:58:17,279 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80031.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 17:58:20,300 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.104e+02 3.486e+02 4.096e+02 4.877e+02 1.090e+03, threshold=8.193e+02, percent-clipped=2.0
+2023-03-29 17:58:30,516 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80037.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 17:59:38,760 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1262, 3.3049, 2.1397, 3.7838, 3.4831, 3.7581, 3.7845, 2.9769],
+       device='cuda:3'), covar=tensor([0.0724, 0.0680, 0.1569, 0.0710, 0.0597, 0.0560, 0.0719, 0.0888],
+       device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0151, 0.0147, 0.0161, 0.0141, 0.0147, 0.0156, 0.0155],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 17:59:39,688 INFO [train.py:892] (3/4) Epoch 44, batch 300, loss[loss=0.1499, simple_loss=0.24, pruned_loss=0.02995, over 19820.00 frames. ], tot_loss[loss=0.1503, simple_loss=0.2326, pruned_loss=0.034, over 3072579.23 frames. ], batch size: 98, lr: 3.56e-03, grad_scale: 16.0
+2023-03-29 17:59:44,494 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80068.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:00:10,029 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80079.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 18:01:38,218 INFO [train.py:892] (3/4) Epoch 44, batch 350, loss[loss=0.1419, simple_loss=0.2281, pruned_loss=0.02787, over 19773.00 frames. ], tot_loss[loss=0.1504, simple_loss=0.2327, pruned_loss=0.03411, over 3267541.81 frames. ], batch size: 70, lr: 3.55e-03, grad_scale: 16.0
+2023-03-29 18:01:44,088 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80118.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:02:20,969 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.499e+02 3.590e+02 4.276e+02 5.127e+02 9.462e+02, threshold=8.552e+02, percent-clipped=1.0
+2023-03-29 18:03:35,982 INFO [train.py:892] (3/4) Epoch 44, batch 400, loss[loss=0.1469, simple_loss=0.2234, pruned_loss=0.03516, over 19836.00 frames. ], tot_loss[loss=0.1504, simple_loss=0.2326, pruned_loss=0.03404, over 3417554.72 frames. ], batch size: 171, lr: 3.55e-03, grad_scale: 16.0
+2023-03-29 18:04:17,512 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9623, 2.6153, 4.1152, 3.7609, 4.1406, 4.2009, 3.9926, 3.9335],
+       device='cuda:3'), covar=tensor([0.0657, 0.1010, 0.0117, 0.0516, 0.0148, 0.0214, 0.0192, 0.0188],
+       device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0109, 0.0093, 0.0157, 0.0092, 0.0105, 0.0096, 0.0093],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 18:04:43,380 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0
+2023-03-29 18:04:47,436 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7960, 4.0630, 4.3949, 4.9425, 3.1924, 3.5199, 2.8852, 3.1567],
+       device='cuda:3'), covar=tensor([0.0430, 0.1677, 0.0684, 0.0310, 0.1948, 0.1190, 0.1329, 0.1482],
+       device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0327, 0.0254, 0.0213, 0.0251, 0.0217, 0.0225, 0.0220],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 18:05:38,912 INFO [train.py:892] (3/4) Epoch 44, batch 450, loss[loss=0.1576, simple_loss=0.2336, pruned_loss=0.04078, over 19828.00 frames. ], tot_loss[loss=0.1506, simple_loss=0.2328, pruned_loss=0.03423, over 3536759.59 frames. ], batch size: 128, lr: 3.55e-03, grad_scale: 16.0
+2023-03-29 18:06:18,572 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.753e+02 3.555e+02 3.976e+02 4.443e+02 7.591e+02, threshold=7.952e+02, percent-clipped=1.0
+2023-03-29 18:06:38,991 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80241.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:07:34,116 INFO [train.py:892] (3/4) Epoch 44, batch 500, loss[loss=0.1593, simple_loss=0.2308, pruned_loss=0.0439, over 19769.00 frames. ], tot_loss[loss=0.15, simple_loss=0.2319, pruned_loss=0.03401, over 3629768.03 frames. ], batch size: 100, lr: 3.55e-03, grad_scale: 16.0
+2023-03-29 18:07:57,490 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0274, 2.8624, 3.2247, 2.7796, 3.2538, 3.2335, 3.8283, 4.1994],
+       device='cuda:3'), covar=tensor([0.0554, 0.1710, 0.1474, 0.2182, 0.1701, 0.1461, 0.0632, 0.0586],
+       device='cuda:3'), in_proj_covar=tensor([0.0264, 0.0248, 0.0276, 0.0263, 0.0308, 0.0267, 0.0242, 0.0273],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 18:08:14,375 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80283.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:08:28,366 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:08:51,900 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0
+2023-03-29 18:09:27,917 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:09:33,218 INFO [train.py:892] (3/4) Epoch 44, batch 550, loss[loss=0.1651, simple_loss=0.2425, pruned_loss=0.04384, over 19764.00 frames. ], tot_loss[loss=0.1509, simple_loss=0.2329, pruned_loss=0.03452, over 3700309.96 frames. ], batch size: 244, lr: 3.55e-03, grad_scale: 16.0
+2023-03-29 18:09:38,888 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0
+2023-03-29 18:10:13,726 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.312e+02 3.379e+02 4.084e+02 4.890e+02 8.912e+02, threshold=8.167e+02, percent-clipped=1.0
+2023-03-29 18:10:41,201 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80344.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:11:16,082 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.80 vs. limit=5.0
+2023-03-29 18:11:21,907 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80361.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:11:31,195 INFO [train.py:892] (3/4) Epoch 44, batch 600, loss[loss=0.1598, simple_loss=0.239, pruned_loss=0.04028, over 19745.00 frames. ], tot_loss[loss=0.1505, simple_loss=0.2322, pruned_loss=0.03439, over 3755895.24 frames. ], batch size: 77, lr: 3.55e-03, grad_scale: 16.0
+2023-03-29 18:12:34,825 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80393.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:12:56,259 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2907, 3.4775, 3.0629, 2.6508, 3.0677, 3.4737, 3.4903, 3.4717],
+       device='cuda:3'), covar=tensor([0.0321, 0.0335, 0.0305, 0.0514, 0.0357, 0.0322, 0.0219, 0.0202],
+       device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0108, 0.0108, 0.0109, 0.0112, 0.0098, 0.0099, 0.0098],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 18:13:27,033 INFO [train.py:892] (3/4) Epoch 44, batch 650, loss[loss=0.2032, simple_loss=0.2728, pruned_loss=0.06676, over 19763.00 frames. ], tot_loss[loss=0.1497, simple_loss=0.2313, pruned_loss=0.03403, over 3799772.51 frames. ], batch size: 253, lr: 3.55e-03, grad_scale: 16.0
+2023-03-29 18:13:32,844 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80418.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:14:07,142 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.620e+02 3.666e+02 4.287e+02 5.079e+02 9.209e+02, threshold=8.573e+02, percent-clipped=4.0
+2023-03-29 18:14:55,609 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80454.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:15:14,933 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80463.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:15:20,978 INFO [train.py:892] (3/4) Epoch 44, batch 700, loss[loss=0.1545, simple_loss=0.2345, pruned_loss=0.03721, over 19800.00 frames. ], tot_loss[loss=0.1502, simple_loss=0.2316, pruned_loss=0.03437, over 3833767.21 frames. ], batch size: 72, lr: 3.55e-03, grad_scale: 16.0
+2023-03-29 18:15:23,434 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80466.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:16:26,769 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80494.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:16:54,133 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4560, 3.3791, 3.5927, 2.7238, 3.6526, 3.0269, 3.4208, 3.6345],
+       device='cuda:3'), covar=tensor([0.0667, 0.0394, 0.0503, 0.0841, 0.0466, 0.0489, 0.0500, 0.0368],
+       device='cuda:3'), in_proj_covar=tensor([0.0085, 0.0095, 0.0091, 0.0114, 0.0086, 0.0090, 0.0086, 0.0082],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 18:17:18,258 INFO [train.py:892] (3/4) Epoch 44, batch 750, loss[loss=0.163, simple_loss=0.2447, pruned_loss=0.04065, over 19612.00 frames. ], tot_loss[loss=0.1504, simple_loss=0.2321, pruned_loss=0.03432, over 3859690.09 frames. ], batch size: 65, lr: 3.55e-03, grad_scale: 16.0
+2023-03-29 18:17:38,342 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80524.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:17:57,773 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.299e+02 3.592e+02 4.231e+02 5.310e+02 9.600e+02, threshold=8.463e+02, percent-clipped=2.0
+2023-03-29 18:18:50,561 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:19:14,321 INFO [train.py:892] (3/4) Epoch 44, batch 800, loss[loss=0.1668, simple_loss=0.2517, pruned_loss=0.041, over 19707.00 frames. ], tot_loss[loss=0.1501, simple_loss=0.2319, pruned_loss=0.0341, over 3879583.26 frames. ], batch size: 81, lr: 3.54e-03, grad_scale: 16.0
+2023-03-29 18:20:46,910 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.13 vs. limit=5.0
+2023-03-29 18:21:15,111 INFO [train.py:892] (3/4) Epoch 44, batch 850, loss[loss=0.1324, simple_loss=0.2216, pruned_loss=0.02157, over 19609.00 frames. ], tot_loss[loss=0.1498, simple_loss=0.2319, pruned_loss=0.0338, over 3894955.36 frames. ], batch size: 51, lr: 3.54e-03, grad_scale: 16.0
+2023-03-29 18:21:54,770 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.578e+02 3.349e+02 3.795e+02 4.586e+02 1.230e+03, threshold=7.591e+02, percent-clipped=1.0
+2023-03-29 18:22:09,873 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80639.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:23:11,187 INFO [train.py:892] (3/4) Epoch 44, batch 900, loss[loss=0.1463, simple_loss=0.2316, pruned_loss=0.03051, over 19750.00 frames. ], tot_loss[loss=0.1486, simple_loss=0.2307, pruned_loss=0.03325, over 3907838.99 frames. ], batch size: 250, lr: 3.54e-03, grad_scale: 16.0
+2023-03-29 18:24:56,080 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4448, 5.7402, 5.7995, 5.6493, 5.4202, 5.7358, 5.1896, 5.1895],
+       device='cuda:3'), covar=tensor([0.0449, 0.0500, 0.0433, 0.0437, 0.0606, 0.0498, 0.0630, 0.1005],
+       device='cuda:3'), in_proj_covar=tensor([0.0298, 0.0316, 0.0325, 0.0282, 0.0295, 0.0277, 0.0287, 0.0338],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 18:25:06,280 INFO [train.py:892] (3/4) Epoch 44, batch 950, loss[loss=0.151, simple_loss=0.236, pruned_loss=0.03294, over 19708.00 frames. ], tot_loss[loss=0.1489, simple_loss=0.2309, pruned_loss=0.03342, over 3917588.59 frames. ], batch size: 283, lr: 3.54e-03, grad_scale: 16.0
+2023-03-29 18:25:43,572 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7007, 3.8404, 2.3244, 3.9699, 4.1056, 1.8506, 3.4476, 3.1599],
+       device='cuda:3'), covar=tensor([0.0806, 0.0833, 0.2847, 0.0788, 0.0800, 0.2920, 0.1050, 0.0970],
+       device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0270, 0.0241, 0.0292, 0.0273, 0.0210, 0.0247, 0.0213],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 18:25:49,402 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.433e+02 3.452e+02 4.047e+02 4.823e+02 9.247e+02, threshold=8.094e+02, percent-clipped=4.0
+2023-03-29 18:25:56,874 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80736.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:26:26,605 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80749.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:27:06,666 INFO [train.py:892] (3/4) Epoch 44, batch 1000, loss[loss=0.1519, simple_loss=0.2371, pruned_loss=0.03337, over 19564.00 frames. ], tot_loss[loss=0.1483, simple_loss=0.2302, pruned_loss=0.0332, over 3925662.57 frames. ], batch size: 53, lr: 3.54e-03, grad_scale: 16.0
+2023-03-29 18:27:24,856 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:28:19,072 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80797.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:29:01,253 INFO [train.py:892] (3/4) Epoch 44, batch 1050, loss[loss=0.1822, simple_loss=0.2512, pruned_loss=0.05656, over 19723.00 frames. ], tot_loss[loss=0.1495, simple_loss=0.2313, pruned_loss=0.03379, over 3930892.11 frames. ], batch size: 219, lr: 3.54e-03, grad_scale: 16.0
+2023-03-29 18:29:09,884 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:29:41,370 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.092e+02 3.498e+02 3.993e+02 4.688e+02 1.348e+03, threshold=7.986e+02, percent-clipped=2.0
+2023-03-29 18:29:44,710 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80834.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:30:22,007 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80850.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:30:47,241 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0169, 5.2234, 5.4658, 5.2448, 5.2882, 5.0959, 5.1823, 4.9952],
+       device='cuda:3'), covar=tensor([0.1589, 0.1467, 0.0864, 0.1304, 0.0777, 0.0813, 0.1796, 0.2096],
+       device='cuda:3'), in_proj_covar=tensor([0.0308, 0.0358, 0.0385, 0.0320, 0.0293, 0.0301, 0.0379, 0.0409],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 18:30:57,642 INFO [train.py:892] (3/4) Epoch 44, batch 1100, loss[loss=0.1378, simple_loss=0.2185, pruned_loss=0.02855, over 19726.00 frames. ], tot_loss[loss=0.1491, simple_loss=0.2309, pruned_loss=0.03367, over 3934843.38 frames. ], batch size: 76, lr: 3.54e-03, grad_scale: 16.0
+2023-03-29 18:32:57,629 INFO [train.py:892] (3/4) Epoch 44, batch 1150, loss[loss=0.1444, simple_loss=0.2287, pruned_loss=0.03, over 19797.00 frames. ], tot_loss[loss=0.149, simple_loss=0.2303, pruned_loss=0.03383, over 3938409.08 frames. ], batch size: 51, lr: 3.54e-03, grad_scale: 16.0
+2023-03-29 18:33:35,847 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.488e+02 3.446e+02 4.222e+02 4.835e+02 9.572e+02, threshold=8.444e+02, percent-clipped=1.0
+2023-03-29 18:33:50,172 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:34:56,413 INFO [train.py:892] (3/4) Epoch 44, batch 1200, loss[loss=0.139, simple_loss=0.2294, pruned_loss=0.02426, over 19785.00 frames. ], tot_loss[loss=0.1487, simple_loss=0.2299, pruned_loss=0.03372, over 3942688.08 frames. ], batch size: 48, lr: 3.54e-03, grad_scale: 16.0
+2023-03-29 18:35:47,304 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:36:53,881 INFO [train.py:892] (3/4) Epoch 44, batch 1250, loss[loss=0.1462, simple_loss=0.222, pruned_loss=0.03513, over 19851.00 frames. ], tot_loss[loss=0.1479, simple_loss=0.2292, pruned_loss=0.03326, over 3944978.30 frames. ], batch size: 43, lr: 3.53e-03, grad_scale: 16.0
+2023-03-29 18:37:34,781 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.062e+02 3.416e+02 3.866e+02 4.633e+02 7.617e+02, threshold=7.733e+02, percent-clipped=0.0
+2023-03-29 18:38:13,533 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81049.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:38:23,915 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2770, 3.5969, 3.0230, 2.6910, 3.1609, 3.4581, 3.3953, 3.5161],
+       device='cuda:3'), covar=tensor([0.0298, 0.0244, 0.0308, 0.0499, 0.0337, 0.0303, 0.0290, 0.0214],
+       device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0108, 0.0108, 0.0109, 0.0112, 0.0099, 0.0100, 0.0098],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 18:38:51,629 INFO [train.py:892] (3/4) Epoch 44, batch 1300, loss[loss=0.1453, simple_loss=0.2193, pruned_loss=0.03563, over 19866.00 frames. ], tot_loss[loss=0.1475, simple_loss=0.2287, pruned_loss=0.03312, over 3946335.66 frames. ], batch size: 129, lr: 3.53e-03, grad_scale: 16.0
+2023-03-29 18:39:52,207 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81092.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:40:04,809 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:40:48,678 INFO [train.py:892] (3/4) Epoch 44, batch 1350, loss[loss=0.2022, simple_loss=0.3214, pruned_loss=0.04145, over 18959.00 frames. ], tot_loss[loss=0.1484, simple_loss=0.2299, pruned_loss=0.03344, over 3945401.06 frames. ], batch size: 514, lr: 3.53e-03, grad_scale: 16.0
+2023-03-29 18:40:49,847 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:40:56,043 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81119.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:41:19,220 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81129.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:41:27,849 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.454e+02 3.446e+02 4.149e+02 4.801e+02 1.146e+03, threshold=8.297e+02, percent-clipped=2.0
+2023-03-29 18:42:04,475 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81150.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:42:39,479 INFO [train.py:892] (3/4) Epoch 44, batch 1400, loss[loss=0.1221, simple_loss=0.2055, pruned_loss=0.01937, over 19901.00 frames. ], tot_loss[loss=0.1482, simple_loss=0.2295, pruned_loss=0.03341, over 3946989.05 frames. ], batch size: 113, lr: 3.53e-03, grad_scale: 16.0
+2023-03-29 18:42:42,798 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81167.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:43:08,394 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81177.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:43:15,216 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5308, 2.8645, 2.5177, 2.0391, 2.5812, 2.7991, 2.7018, 2.8257],
+       device='cuda:3'), covar=tensor([0.0446, 0.0341, 0.0390, 0.0633, 0.0415, 0.0313, 0.0362, 0.0301],
+       device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0108, 0.0108, 0.0109, 0.0112, 0.0099, 0.0100, 0.0098],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 18:43:58,141 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81198.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:44:00,675 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5146, 2.6080, 4.6232, 3.9418, 4.4215, 4.5862, 4.3842, 4.3122],
+       device='cuda:3'), covar=tensor([0.0617, 0.1147, 0.0117, 0.0864, 0.0170, 0.0212, 0.0180, 0.0180],
+       device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0108, 0.0093, 0.0155, 0.0091, 0.0105, 0.0095, 0.0092],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 18:44:41,825 INFO [train.py:892] (3/4) Epoch 44, batch 1450, loss[loss=0.1448, simple_loss=0.228, pruned_loss=0.03077, over 19821.00 frames. ], tot_loss[loss=0.149, simple_loss=0.2308, pruned_loss=0.03354, over 3945868.07 frames. ], batch size: 72, lr: 3.53e-03, grad_scale: 16.0
+2023-03-29 18:45:20,234 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.287e+02 3.615e+02 3.975e+02 5.064e+02 7.888e+02, threshold=7.950e+02, percent-clipped=0.0
+2023-03-29 18:45:49,626 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8132, 3.1656, 3.2448, 3.6826, 2.6168, 3.1664, 2.3351, 2.4511],
+       device='cuda:3'), covar=tensor([0.0535, 0.1585, 0.1071, 0.0520, 0.2080, 0.0945, 0.1507, 0.1723],
+       device='cuda:3'), in_proj_covar=tensor([0.0252, 0.0325, 0.0253, 0.0212, 0.0250, 0.0215, 0.0222, 0.0220],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 18:46:29,920 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3003, 2.5777, 3.6596, 3.0072, 3.0928, 2.8975, 2.2575, 2.3679],
+       device='cuda:3'), covar=tensor([0.1199, 0.2926, 0.0612, 0.1127, 0.1852, 0.1682, 0.2698, 0.2651],
+       device='cuda:3'), in_proj_covar=tensor([0.0358, 0.0403, 0.0358, 0.0298, 0.0379, 0.0401, 0.0389, 0.0366],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 18:46:39,803 INFO [train.py:892] (3/4) Epoch 44, batch 1500, loss[loss=0.1647, simple_loss=0.2453, pruned_loss=0.04204, over 19795.00 frames. ], tot_loss[loss=0.1493, simple_loss=0.2313, pruned_loss=0.03361, over 3946685.14 frames. ], batch size: 195, lr: 3.53e-03, grad_scale: 16.0
+2023-03-29 18:47:22,360 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0309, 3.0127, 2.0146, 3.4520, 3.2821, 3.4500, 3.5062, 2.8660],
+       device='cuda:3'), covar=tensor([0.0730, 0.0774, 0.1695, 0.0762, 0.0674, 0.0527, 0.0715, 0.0838],
+       device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0150, 0.0146, 0.0160, 0.0140, 0.0147, 0.0156, 0.0154],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 18:47:32,979 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2794, 3.9922, 4.0954, 4.2645, 4.0336, 4.3023, 4.3676, 4.5428],
+       device='cuda:3'), covar=tensor([0.0663, 0.0518, 0.0584, 0.0445, 0.0874, 0.0622, 0.0482, 0.0355],
+       device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0188, 0.0209, 0.0187, 0.0185, 0.0168, 0.0161, 0.0208],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 18:48:01,985 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81299.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:48:06,212 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81301.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:48:39,313 INFO [train.py:892] (3/4) Epoch 44, batch 1550, loss[loss=0.1642, simple_loss=0.2533, pruned_loss=0.03754, over 19691.00 frames. ], tot_loss[loss=0.1488, simple_loss=0.2308, pruned_loss=0.03343, over 3947199.80 frames. ], batch size: 325, lr: 3.53e-03, grad_scale: 16.0
+2023-03-29 18:49:04,255 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3070, 4.0655, 4.1234, 3.8794, 4.2820, 2.9636, 3.6443, 2.1970],
+       device='cuda:3'), covar=tensor([0.0176, 0.0236, 0.0151, 0.0196, 0.0157, 0.1112, 0.0610, 0.1418],
+       device='cuda:3'), in_proj_covar=tensor([0.0111, 0.0154, 0.0119, 0.0141, 0.0125, 0.0140, 0.0146, 0.0132],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003],
+       device='cuda:3')
+2023-03-29 18:49:09,166 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8864, 2.8867, 1.8958, 3.3190, 3.1186, 3.2713, 3.3405, 2.7371],
+       device='cuda:3'), covar=tensor([0.0767, 0.0796, 0.1631, 0.0694, 0.0637, 0.0537, 0.0630, 0.0876],
+       device='cuda:3'), in_proj_covar=tensor([0.0150, 0.0150, 0.0146, 0.0160, 0.0140, 0.0146, 0.0155, 0.0153],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 18:49:19,866 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.242e+02 3.436e+02 3.963e+02 4.737e+02 1.006e+03, threshold=7.927e+02, percent-clipped=2.0
+2023-03-29 18:49:32,884 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6252, 3.7481, 3.9694, 3.6864, 3.9791, 3.6632, 3.6754, 3.4844],
+       device='cuda:3'), covar=tensor([0.1590, 0.1933, 0.1141, 0.1643, 0.1207, 0.1210, 0.2147, 0.2549],
+       device='cuda:3'), in_proj_covar=tensor([0.0309, 0.0359, 0.0386, 0.0321, 0.0295, 0.0303, 0.0381, 0.0410],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 18:49:47,786 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0940, 3.3937, 2.9088, 2.6020, 2.9598, 3.3576, 3.2559, 3.3332],
+       device='cuda:3'), covar=tensor([0.0331, 0.0319, 0.0360, 0.0517, 0.0368, 0.0278, 0.0245, 0.0231],
+       device='cuda:3'), in_proj_covar=tensor([0.0115, 0.0108, 0.0108, 0.0108, 0.0111, 0.0098, 0.0099, 0.0097],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 18:50:24,326 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81360.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:50:28,558 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:50:36,277 INFO [train.py:892] (3/4) Epoch 44, batch 1600, loss[loss=0.1209, simple_loss=0.1979, pruned_loss=0.02192, over 19728.00 frames. ], tot_loss[loss=0.1496, simple_loss=0.2316, pruned_loss=0.03382, over 3947380.77 frames. ], batch size: 47, lr: 3.53e-03, grad_scale: 16.0
+2023-03-29 18:51:40,701 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81392.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:52:28,041 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8581, 5.1034, 5.1674, 5.0521, 4.8127, 5.1501, 4.6595, 4.6911],
+       device='cuda:3'), covar=tensor([0.0504, 0.0483, 0.0458, 0.0425, 0.0626, 0.0457, 0.0648, 0.0922],
+       device='cuda:3'), in_proj_covar=tensor([0.0295, 0.0312, 0.0323, 0.0281, 0.0292, 0.0273, 0.0285, 0.0336],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 18:52:37,062 INFO [train.py:892] (3/4) Epoch 44, batch 1650, loss[loss=0.1323, simple_loss=0.2083, pruned_loss=0.0281, over 19817.00 frames. ], tot_loss[loss=0.1492, simple_loss=0.2311, pruned_loss=0.03366, over 3947475.60 frames. ], batch size: 133, lr: 3.53e-03, grad_scale: 16.0
+2023-03-29 18:53:06,914 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81429.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:53:16,900 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.426e+02 3.380e+02 3.975e+02 4.611e+02 1.350e+03, threshold=7.949e+02, percent-clipped=2.0
+2023-03-29 18:53:33,677 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:54:32,117 INFO [train.py:892] (3/4) Epoch 44, batch 1700, loss[loss=0.1406, simple_loss=0.214, pruned_loss=0.03359, over 19781.00 frames. ], tot_loss[loss=0.1483, simple_loss=0.23, pruned_loss=0.03335, over 3949976.74 frames. ], batch size: 131, lr: 3.52e-03, grad_scale: 16.0
+2023-03-29 18:54:39,436 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3650, 4.5849, 2.6148, 4.7999, 4.9761, 2.2346, 4.2467, 3.5528],
+       device='cuda:3'), covar=tensor([0.0699, 0.0624, 0.2800, 0.0660, 0.0423, 0.2699, 0.0904, 0.0954],
+       device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0269, 0.0241, 0.0291, 0.0271, 0.0209, 0.0247, 0.0212],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 18:54:49,225 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81472.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:55:00,355 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81477.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 18:56:28,458 INFO [train.py:892] (3/4) Epoch 44, batch 1750, loss[loss=0.1301, simple_loss=0.2103, pruned_loss=0.02499, over 19619.00 frames. ], tot_loss[loss=0.1474, simple_loss=0.2291, pruned_loss=0.03291, over 3950737.50 frames. ], batch size: 52, lr: 3.52e-03, grad_scale: 16.0
+2023-03-29 18:57:03,565 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.489e+02 3.354e+02 3.892e+02 4.607e+02 1.019e+03, threshold=7.783e+02, percent-clipped=2.0
+2023-03-29 18:58:09,233 INFO [train.py:892] (3/4) Epoch 44, batch 1800, loss[loss=0.1631, simple_loss=0.255, pruned_loss=0.03556, over 19669.00 frames. ], tot_loss[loss=0.1489, simple_loss=0.2307, pruned_loss=0.03355, over 3949581.37 frames. ], batch size: 55, lr: 3.52e-03, grad_scale: 16.0
+2023-03-29 18:58:43,892 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4989, 2.8391, 2.6258, 2.1054, 2.7009, 2.7853, 2.8261, 2.8199],
+       device='cuda:3'), covar=tensor([0.0458, 0.0364, 0.0346, 0.0597, 0.0393, 0.0327, 0.0304, 0.0325],
+       device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0108, 0.0108, 0.0109, 0.0112, 0.0099, 0.0100, 0.0098],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 18:59:35,025 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.62 vs. limit=2.0
+2023-03-29 18:59:40,692 INFO [train.py:892] (3/4) Epoch 44, batch 1850, loss[loss=0.1737, simple_loss=0.2659, pruned_loss=0.04077, over 19834.00 frames. ], tot_loss[loss=0.1504, simple_loss=0.2331, pruned_loss=0.03379, over 3949165.01 frames. ], batch size: 57, lr: 3.52e-03, grad_scale: 16.0
+2023-03-29 19:00:42,759 INFO [train.py:892] (3/4) Epoch 45, batch 0, loss[loss=0.1267, simple_loss=0.2028, pruned_loss=0.02532, over 19845.00 frames. ], tot_loss[loss=0.1267, simple_loss=0.2028, pruned_loss=0.02532, over 19845.00 frames. ], batch size: 197, lr: 3.48e-03, grad_scale: 16.0
+2023-03-29 19:00:42,759 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-29 19:01:20,024 INFO [train.py:926] (3/4) Epoch 45, validation: loss=0.1889, simple_loss=0.2504, pruned_loss=0.0637, over 2883724.00 frames. 
+2023-03-29 19:01:20,026 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-29 19:01:47,747 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.448e+02 3.358e+02 4.011e+02 4.800e+02 7.460e+02, threshold=8.023e+02, percent-clipped=0.0
+2023-03-29 19:01:56,426 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0
+2023-03-29 19:02:13,801 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0
+2023-03-29 19:02:42,895 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:02:48,833 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=81657.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:03:18,125 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81670.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:03:18,987 INFO [train.py:892] (3/4) Epoch 45, batch 50, loss[loss=0.1422, simple_loss=0.2253, pruned_loss=0.02952, over 19754.00 frames. ], tot_loss[loss=0.1451, simple_loss=0.2264, pruned_loss=0.03187, over 889436.06 frames. ], batch size: 213, lr: 3.48e-03, grad_scale: 16.0
+2023-03-29 19:04:26,925 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2948, 4.2473, 4.6299, 4.3686, 4.5489, 4.1690, 4.3833, 4.1378],
+       device='cuda:3'), covar=tensor([0.1411, 0.1627, 0.0961, 0.1396, 0.0920, 0.1008, 0.1895, 0.2145],
+       device='cuda:3'), in_proj_covar=tensor([0.0307, 0.0358, 0.0387, 0.0320, 0.0294, 0.0301, 0.0379, 0.0408],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 19:05:21,204 INFO [train.py:892] (3/4) Epoch 45, batch 100, loss[loss=0.1285, simple_loss=0.2042, pruned_loss=0.02639, over 19766.00 frames. ], tot_loss[loss=0.1477, simple_loss=0.2292, pruned_loss=0.03303, over 1567253.93 frames. ], batch size: 155, lr: 3.48e-03, grad_scale: 16.0
+2023-03-29 19:05:44,858 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=81731.0, num_to_drop=1, layers_to_drop={3}
+2023-03-29 19:05:49,403 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.145e+02 3.372e+02 3.916e+02 5.007e+02 1.092e+03, threshold=7.832e+02, percent-clipped=2.0
+2023-03-29 19:06:03,340 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1258, 2.7358, 3.2500, 3.4184, 3.8838, 4.4676, 4.2144, 4.2927],
+       device='cuda:3'), covar=tensor([0.0974, 0.1623, 0.1257, 0.0689, 0.0401, 0.0198, 0.0341, 0.0406],
+       device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0169, 0.0183, 0.0157, 0.0144, 0.0138, 0.0133, 0.0122],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 19:06:14,119 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1663, 3.0956, 4.8773, 3.5661, 3.8025, 3.4976, 2.5643, 2.7987],
+       device='cuda:3'), covar=tensor([0.1028, 0.3070, 0.0370, 0.1121, 0.1835, 0.1620, 0.2785, 0.2739],
+       device='cuda:3'), in_proj_covar=tensor([0.0361, 0.0406, 0.0360, 0.0300, 0.0382, 0.0404, 0.0392, 0.0369],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 19:07:18,290 INFO [train.py:892] (3/4) Epoch 45, batch 150, loss[loss=0.1365, simple_loss=0.225, pruned_loss=0.02401, over 19808.00 frames. ], tot_loss[loss=0.1464, simple_loss=0.228, pruned_loss=0.03237, over 2095262.11 frames. ], batch size: 82, lr: 3.48e-03, grad_scale: 16.0
+2023-03-29 19:07:21,634 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81772.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:09:11,564 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=81820.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:09:12,760 INFO [train.py:892] (3/4) Epoch 45, batch 200, loss[loss=0.1442, simple_loss=0.2294, pruned_loss=0.02946, over 19783.00 frames. ], tot_loss[loss=0.1476, simple_loss=0.2293, pruned_loss=0.03296, over 2507515.66 frames. ], batch size: 263, lr: 3.48e-03, grad_scale: 16.0
+2023-03-29 19:09:39,855 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.431e+02 3.452e+02 4.222e+02 4.821e+02 9.570e+02, threshold=8.444e+02, percent-clipped=2.0
+2023-03-29 19:10:54,950 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0
+2023-03-29 19:11:08,769 INFO [train.py:892] (3/4) Epoch 45, batch 250, loss[loss=0.2872, simple_loss=0.3585, pruned_loss=0.108, over 19133.00 frames. ], tot_loss[loss=0.1504, simple_loss=0.2331, pruned_loss=0.0339, over 2821941.01 frames. ], batch size: 452, lr: 3.48e-03, grad_scale: 16.0
+2023-03-29 19:12:49,841 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.57 vs. limit=5.0
+2023-03-29 19:13:09,881 INFO [train.py:892] (3/4) Epoch 45, batch 300, loss[loss=0.1638, simple_loss=0.2431, pruned_loss=0.04228, over 19859.00 frames. ], tot_loss[loss=0.1502, simple_loss=0.2328, pruned_loss=0.03385, over 3072262.25 frames. ], batch size: 51, lr: 3.48e-03, grad_scale: 16.0
+2023-03-29 19:13:37,304 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.189e+02 3.336e+02 4.013e+02 4.745e+02 8.684e+02, threshold=8.026e+02, percent-clipped=2.0
+2023-03-29 19:14:30,562 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81955.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:14:34,965 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=81957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:15:04,633 INFO [train.py:892] (3/4) Epoch 45, batch 350, loss[loss=0.1553, simple_loss=0.2349, pruned_loss=0.03782, over 19795.00 frames. ], tot_loss[loss=0.1493, simple_loss=0.2319, pruned_loss=0.03338, over 3265435.66 frames. ], batch size: 185, lr: 3.47e-03, grad_scale: 16.0
+2023-03-29 19:15:45,343 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=81987.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:16:22,889 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82003.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:16:28,423 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82005.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:17:03,785 INFO [train.py:892] (3/4) Epoch 45, batch 400, loss[loss=0.1348, simple_loss=0.221, pruned_loss=0.0243, over 19486.00 frames. ], tot_loss[loss=0.1486, simple_loss=0.231, pruned_loss=0.03306, over 3417341.50 frames. ], batch size: 43, lr: 3.47e-03, grad_scale: 32.0
+2023-03-29 19:17:15,235 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82026.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 19:17:33,550 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.143e+02 3.235e+02 3.968e+02 4.770e+02 8.639e+02, threshold=7.936e+02, percent-clipped=1.0
+2023-03-29 19:18:09,193 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82048.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:18:26,184 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82055.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:18:40,167 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2101, 3.0015, 3.0829, 3.2544, 3.1611, 3.1010, 3.2623, 3.4547],
+       device='cuda:3'), covar=tensor([0.0777, 0.0569, 0.0634, 0.0481, 0.0715, 0.0890, 0.0562, 0.0401],
+       device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0186, 0.0207, 0.0185, 0.0183, 0.0167, 0.0161, 0.0207],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 19:19:03,151 INFO [train.py:892] (3/4) Epoch 45, batch 450, loss[loss=0.1587, simple_loss=0.2419, pruned_loss=0.03779, over 19691.00 frames. ], tot_loss[loss=0.1489, simple_loss=0.2311, pruned_loss=0.03335, over 3535627.43 frames. ], batch size: 315, lr: 3.47e-03, grad_scale: 32.0
+2023-03-29 19:19:07,107 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.90 vs. limit=5.0
+2023-03-29 19:20:10,898 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6028, 2.0751, 2.3608, 2.8318, 3.1188, 3.2968, 3.1312, 3.1786],
+       device='cuda:3'), covar=tensor([0.1092, 0.1790, 0.1459, 0.0790, 0.0636, 0.0394, 0.0577, 0.0506],
+       device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0169, 0.0183, 0.0157, 0.0144, 0.0138, 0.0133, 0.0123],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 19:20:51,707 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:21:01,463 INFO [train.py:892] (3/4) Epoch 45, batch 500, loss[loss=0.1763, simple_loss=0.2651, pruned_loss=0.04369, over 19849.00 frames. ], tot_loss[loss=0.1504, simple_loss=0.2326, pruned_loss=0.03413, over 3625645.14 frames. ], batch size: 56, lr: 3.47e-03, grad_scale: 32.0
+2023-03-29 19:21:29,430 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.527e+02 3.621e+02 4.307e+02 5.011e+02 9.871e+02, threshold=8.613e+02, percent-clipped=3.0
+2023-03-29 19:21:30,959 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6782, 3.7564, 2.3271, 3.9074, 4.0136, 1.9057, 3.3547, 3.1190],
+       device='cuda:3'), covar=tensor([0.0801, 0.0864, 0.2655, 0.0864, 0.0680, 0.2655, 0.1149, 0.0944],
+       device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0268, 0.0239, 0.0290, 0.0270, 0.0209, 0.0246, 0.0211],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 19:22:58,982 INFO [train.py:892] (3/4) Epoch 45, batch 550, loss[loss=0.1328, simple_loss=0.218, pruned_loss=0.02379, over 19744.00 frames. ], tot_loss[loss=0.1512, simple_loss=0.2331, pruned_loss=0.03463, over 3696518.67 frames. ], batch size: 219, lr: 3.47e-03, grad_scale: 32.0
+2023-03-29 19:24:02,782 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8118, 2.3527, 2.6357, 3.0547, 3.3741, 3.6250, 3.4926, 3.5336],
+       device='cuda:3'), covar=tensor([0.1055, 0.1757, 0.1465, 0.0749, 0.0576, 0.0375, 0.0486, 0.0525],
+       device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0172, 0.0185, 0.0158, 0.0146, 0.0140, 0.0135, 0.0124],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 19:24:56,171 INFO [train.py:892] (3/4) Epoch 45, batch 600, loss[loss=0.143, simple_loss=0.2221, pruned_loss=0.03191, over 19646.00 frames. ], tot_loss[loss=0.1509, simple_loss=0.2328, pruned_loss=0.03456, over 3752634.81 frames. ], batch size: 47, lr: 3.47e-03, grad_scale: 32.0
+2023-03-29 19:25:22,519 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.595e+02 3.638e+02 4.311e+02 5.322e+02 1.783e+03, threshold=8.623e+02, percent-clipped=2.0
+2023-03-29 19:25:33,473 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0065, 2.6118, 3.0998, 3.2892, 3.7303, 4.2431, 4.0020, 4.0389],
+       device='cuda:3'), covar=tensor([0.1049, 0.1763, 0.1377, 0.0740, 0.0537, 0.0237, 0.0408, 0.0536],
+       device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0172, 0.0185, 0.0158, 0.0146, 0.0140, 0.0134, 0.0124],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 19:26:49,063 INFO [train.py:892] (3/4) Epoch 45, batch 650, loss[loss=0.1533, simple_loss=0.2296, pruned_loss=0.03853, over 19757.00 frames. ], tot_loss[loss=0.1496, simple_loss=0.2314, pruned_loss=0.03388, over 3797017.29 frames. ], batch size: 125, lr: 3.47e-03, grad_scale: 32.0
+2023-03-29 19:28:46,100 INFO [train.py:892] (3/4) Epoch 45, batch 700, loss[loss=0.1463, simple_loss=0.2272, pruned_loss=0.03268, over 19795.00 frames. ], tot_loss[loss=0.1496, simple_loss=0.2316, pruned_loss=0.03376, over 3831080.01 frames. ], batch size: 185, lr: 3.47e-03, grad_scale: 32.0
+2023-03-29 19:28:59,358 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1058, 4.1439, 2.4850, 4.3807, 4.5537, 1.9912, 3.7651, 3.4990],
+       device='cuda:3'), covar=tensor([0.0704, 0.0801, 0.2748, 0.0752, 0.0590, 0.2788, 0.0982, 0.0925],
+       device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0267, 0.0239, 0.0289, 0.0269, 0.0208, 0.0246, 0.0210],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 19:29:01,309 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82326.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:29:15,655 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.176e+02 3.457e+02 4.012e+02 4.787e+02 1.008e+03, threshold=8.024e+02, percent-clipped=2.0
+2023-03-29 19:29:40,122 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82343.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:29:46,476 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82346.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:29:46,510 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2029, 3.4243, 2.9320, 2.6069, 3.0350, 3.4192, 3.3771, 3.3743],
+       device='cuda:3'), covar=tensor([0.0377, 0.0321, 0.0343, 0.0554, 0.0368, 0.0251, 0.0232, 0.0238],
+       device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0109, 0.0109, 0.0110, 0.0113, 0.0099, 0.0101, 0.0099],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 19:30:08,256 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0288, 3.9289, 4.4107, 3.9659, 3.8107, 4.3520, 4.0692, 4.5071],
+       device='cuda:3'), covar=tensor([0.1042, 0.0485, 0.0498, 0.0495, 0.1211, 0.0657, 0.0667, 0.0485],
+       device='cuda:3'), in_proj_covar=tensor([0.0287, 0.0231, 0.0232, 0.0244, 0.0213, 0.0257, 0.0245, 0.0231],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 19:30:42,586 INFO [train.py:892] (3/4) Epoch 45, batch 750, loss[loss=0.1375, simple_loss=0.2236, pruned_loss=0.02569, over 19612.00 frames. ], tot_loss[loss=0.1489, simple_loss=0.2309, pruned_loss=0.03343, over 3856692.79 frames. ], batch size: 46, lr: 3.47e-03, grad_scale: 32.0
+2023-03-29 19:30:49,719 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82374.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:32:10,302 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82407.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:32:14,001 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.9068, 3.6332, 3.7918, 3.9231, 3.7837, 4.0229, 3.9809, 4.1723],
+       device='cuda:3'), covar=tensor([0.0742, 0.0530, 0.0554, 0.0448, 0.0745, 0.0569, 0.0510, 0.0348],
+       device='cuda:3'), in_proj_covar=tensor([0.0162, 0.0187, 0.0207, 0.0185, 0.0184, 0.0167, 0.0161, 0.0208],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 19:32:19,251 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82411.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:32:38,394 INFO [train.py:892] (3/4) Epoch 45, batch 800, loss[loss=0.1402, simple_loss=0.2218, pruned_loss=0.0293, over 19853.00 frames. ], tot_loss[loss=0.1485, simple_loss=0.2308, pruned_loss=0.03311, over 3877563.95 frames. ], batch size: 112, lr: 3.46e-03, grad_scale: 32.0
+2023-03-29 19:32:48,990 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3111, 4.4308, 2.6179, 4.6389, 4.8442, 2.1820, 4.1619, 3.5619],
+       device='cuda:3'), covar=tensor([0.0665, 0.0769, 0.2730, 0.0703, 0.0472, 0.2668, 0.0898, 0.0911],
+       device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0267, 0.0239, 0.0289, 0.0269, 0.0208, 0.0246, 0.0210],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 19:33:03,770 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.510e+02 3.458e+02 4.305e+02 5.045e+02 7.264e+02, threshold=8.610e+02, percent-clipped=0.0
+2023-03-29 19:33:29,995 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.31 vs. limit=5.0
+2023-03-29 19:34:07,042 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0
+2023-03-29 19:34:36,614 INFO [train.py:892] (3/4) Epoch 45, batch 850, loss[loss=0.1857, simple_loss=0.3028, pruned_loss=0.03428, over 18788.00 frames. ], tot_loss[loss=0.1495, simple_loss=0.2316, pruned_loss=0.03367, over 3894082.30 frames. ], batch size: 564, lr: 3.46e-03, grad_scale: 32.0
+2023-03-29 19:34:55,779 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4774, 4.1215, 4.2299, 4.4010, 4.2260, 4.4941, 4.5029, 4.7072],
+       device='cuda:3'), covar=tensor([0.0613, 0.0512, 0.0580, 0.0415, 0.0710, 0.0551, 0.0500, 0.0337],
+       device='cuda:3'), in_proj_covar=tensor([0.0161, 0.0186, 0.0206, 0.0184, 0.0182, 0.0166, 0.0160, 0.0206],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 19:35:25,153 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82493.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 19:36:29,809 INFO [train.py:892] (3/4) Epoch 45, batch 900, loss[loss=0.1492, simple_loss=0.2322, pruned_loss=0.03312, over 19843.00 frames. ], tot_loss[loss=0.1497, simple_loss=0.2318, pruned_loss=0.03381, over 3906934.82 frames. ], batch size: 59, lr: 3.46e-03, grad_scale: 32.0
+2023-03-29 19:36:57,926 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.335e+02 3.589e+02 4.303e+02 4.859e+02 1.125e+03, threshold=8.606e+02, percent-clipped=1.0
+2023-03-29 19:37:49,584 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82554.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 19:38:17,425 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3887, 2.9903, 3.3816, 2.9187, 3.5603, 3.4970, 4.2304, 4.6234],
+       device='cuda:3'), covar=tensor([0.0532, 0.1700, 0.1456, 0.2284, 0.1517, 0.1469, 0.0588, 0.0518],
+       device='cuda:3'), in_proj_covar=tensor([0.0264, 0.0248, 0.0276, 0.0264, 0.0310, 0.0267, 0.0242, 0.0274],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 19:38:23,974 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82569.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:38:27,762 INFO [train.py:892] (3/4) Epoch 45, batch 950, loss[loss=0.1581, simple_loss=0.245, pruned_loss=0.03561, over 19706.00 frames. ], tot_loss[loss=0.1488, simple_loss=0.2309, pruned_loss=0.03333, over 3916802.54 frames. ], batch size: 283, lr: 3.46e-03, grad_scale: 16.0
+2023-03-29 19:38:38,172 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1956, 4.2688, 2.5215, 4.4972, 4.6847, 2.0947, 3.9066, 3.4632],
+       device='cuda:3'), covar=tensor([0.0676, 0.0793, 0.2798, 0.0706, 0.0498, 0.2877, 0.1022, 0.0922],
+       device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0269, 0.0240, 0.0291, 0.0270, 0.0209, 0.0247, 0.0211],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 19:39:17,883 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4817, 2.6546, 3.8877, 3.0852, 3.2150, 3.0267, 2.3119, 2.4337],
+       device='cuda:3'), covar=tensor([0.1218, 0.3459, 0.0633, 0.1246, 0.1910, 0.1826, 0.2777, 0.2857],
+       device='cuda:3'), in_proj_covar=tensor([0.0359, 0.0405, 0.0358, 0.0299, 0.0381, 0.0403, 0.0391, 0.0368],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 19:40:26,374 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3729, 3.4633, 2.1310, 3.5527, 3.6272, 1.7980, 3.0577, 2.8217],
+       device='cuda:3'), covar=tensor([0.0855, 0.0881, 0.2785, 0.0879, 0.0673, 0.2627, 0.1161, 0.1013],
+       device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0270, 0.0242, 0.0292, 0.0272, 0.0210, 0.0249, 0.0212],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 19:40:27,184 INFO [train.py:892] (3/4) Epoch 45, batch 1000, loss[loss=0.1535, simple_loss=0.2394, pruned_loss=0.03379, over 19638.00 frames. ], tot_loss[loss=0.1485, simple_loss=0.2308, pruned_loss=0.03313, over 3922557.20 frames. ], batch size: 299, lr: 3.46e-03, grad_scale: 16.0
+2023-03-29 19:40:51,818 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=82630.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:40:57,024 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.59 vs. limit=2.0
+2023-03-29 19:41:01,299 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.945e+02 3.409e+02 3.840e+02 4.535e+02 7.731e+02, threshold=7.679e+02, percent-clipped=0.0
+2023-03-29 19:41:22,866 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82643.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:42:26,383 INFO [train.py:892] (3/4) Epoch 45, batch 1050, loss[loss=0.1503, simple_loss=0.2377, pruned_loss=0.03146, over 19754.00 frames. ], tot_loss[loss=0.1487, simple_loss=0.2312, pruned_loss=0.03304, over 3929110.42 frames. ], batch size: 100, lr: 3.46e-03, grad_scale: 16.0
+2023-03-29 19:42:45,171 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3149, 3.2245, 5.1401, 3.7147, 3.8648, 3.7286, 2.7742, 2.9643],
+       device='cuda:3'), covar=tensor([0.0978, 0.3400, 0.0388, 0.1172, 0.2022, 0.1648, 0.2760, 0.2585],
+       device='cuda:3'), in_proj_covar=tensor([0.0360, 0.0405, 0.0358, 0.0300, 0.0382, 0.0403, 0.0392, 0.0368],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 19:43:12,599 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82691.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:43:39,493 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82702.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:43:59,378 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=82711.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:44:19,006 INFO [train.py:892] (3/4) Epoch 45, batch 1100, loss[loss=0.1367, simple_loss=0.2127, pruned_loss=0.03028, over 19875.00 frames. ], tot_loss[loss=0.1486, simple_loss=0.2308, pruned_loss=0.03316, over 3933753.98 frames. ], batch size: 157, lr: 3.46e-03, grad_scale: 16.0
+2023-03-29 19:44:29,219 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0
+2023-03-29 19:44:49,682 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.363e+02 3.616e+02 4.255e+02 4.865e+02 1.038e+03, threshold=8.510e+02, percent-clipped=5.0
+2023-03-29 19:45:48,461 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=82759.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:46:16,999 INFO [train.py:892] (3/4) Epoch 45, batch 1150, loss[loss=0.1407, simple_loss=0.2251, pruned_loss=0.02819, over 19766.00 frames. ], tot_loss[loss=0.1486, simple_loss=0.2308, pruned_loss=0.03321, over 3936157.53 frames. ], batch size: 233, lr: 3.46e-03, grad_scale: 16.0
+2023-03-29 19:48:14,831 INFO [train.py:892] (3/4) Epoch 45, batch 1200, loss[loss=0.1293, simple_loss=0.2014, pruned_loss=0.02857, over 19757.00 frames. ], tot_loss[loss=0.1486, simple_loss=0.2305, pruned_loss=0.03338, over 3940483.59 frames. ], batch size: 129, lr: 3.46e-03, grad_scale: 16.0
+2023-03-29 19:48:45,775 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.475e+02 3.607e+02 4.196e+02 5.114e+02 1.465e+03, threshold=8.391e+02, percent-clipped=1.0
+2023-03-29 19:49:22,178 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82849.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 19:50:06,281 INFO [train.py:892] (3/4) Epoch 45, batch 1250, loss[loss=0.122, simple_loss=0.1987, pruned_loss=0.02264, over 19728.00 frames. ], tot_loss[loss=0.1485, simple_loss=0.2306, pruned_loss=0.03322, over 3939718.17 frames. ], batch size: 95, lr: 3.46e-03, grad_scale: 16.0
+2023-03-29 19:52:02,832 INFO [train.py:892] (3/4) Epoch 45, batch 1300, loss[loss=0.1699, simple_loss=0.259, pruned_loss=0.04045, over 19655.00 frames. ], tot_loss[loss=0.1495, simple_loss=0.2315, pruned_loss=0.03377, over 3942249.75 frames. ], batch size: 67, lr: 3.45e-03, grad_scale: 16.0
+2023-03-29 19:52:15,216 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=82925.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:52:34,755 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.586e+02 3.425e+02 4.088e+02 4.844e+02 1.609e+03, threshold=8.176e+02, percent-clipped=1.0
+2023-03-29 19:52:38,636 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6321, 4.4851, 4.9706, 4.5294, 4.0717, 4.7276, 4.5759, 5.0723],
+       device='cuda:3'), covar=tensor([0.0750, 0.0347, 0.0312, 0.0369, 0.0893, 0.0520, 0.0506, 0.0301],
+       device='cuda:3'), in_proj_covar=tensor([0.0289, 0.0232, 0.0234, 0.0245, 0.0215, 0.0258, 0.0246, 0.0232],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 19:53:03,189 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0
+2023-03-29 19:53:43,661 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=82963.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:54:02,334 INFO [train.py:892] (3/4) Epoch 45, batch 1350, loss[loss=0.1611, simple_loss=0.2528, pruned_loss=0.03472, over 19847.00 frames. ], tot_loss[loss=0.1496, simple_loss=0.2317, pruned_loss=0.03375, over 3942792.82 frames. ], batch size: 56, lr: 3.45e-03, grad_scale: 16.0
+2023-03-29 19:54:55,268 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.5936, 6.0622, 6.1378, 5.8406, 5.7819, 5.8625, 5.8064, 5.6096],
+       device='cuda:3'), covar=tensor([0.1303, 0.1398, 0.0801, 0.1391, 0.0679, 0.0698, 0.1885, 0.2149],
+       device='cuda:3'), in_proj_covar=tensor([0.0307, 0.0356, 0.0386, 0.0321, 0.0296, 0.0300, 0.0380, 0.0409],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 19:55:17,289 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:55:59,059 INFO [train.py:892] (3/4) Epoch 45, batch 1400, loss[loss=0.1577, simple_loss=0.2354, pruned_loss=0.04, over 19714.00 frames. ], tot_loss[loss=0.1482, simple_loss=0.2302, pruned_loss=0.03306, over 3945408.95 frames. ], batch size: 85, lr: 3.45e-03, grad_scale: 16.0
+2023-03-29 19:56:09,271 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83024.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:56:29,948 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.035e+02 3.427e+02 4.004e+02 4.659e+02 9.090e+02, threshold=8.008e+02, percent-clipped=1.0
+2023-03-29 19:57:07,647 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83050.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 19:57:14,739 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5547, 2.6415, 2.7680, 2.6829, 2.6667, 2.7037, 2.6406, 2.7741],
+       device='cuda:3'), covar=tensor([0.0445, 0.0423, 0.0367, 0.0386, 0.0502, 0.0378, 0.0481, 0.0367],
+       device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0093, 0.0095, 0.0090, 0.0103, 0.0095, 0.0111, 0.0083],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 19:57:56,937 INFO [train.py:892] (3/4) Epoch 45, batch 1450, loss[loss=0.1503, simple_loss=0.2268, pruned_loss=0.03692, over 19793.00 frames. ], tot_loss[loss=0.1473, simple_loss=0.2292, pruned_loss=0.0327, over 3948069.29 frames. ], batch size: 120, lr: 3.45e-03, grad_scale: 16.0
+2023-03-29 19:58:11,836 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-29 19:58:23,760 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.75 vs. limit=2.0
+2023-03-29 19:59:49,387 INFO [train.py:892] (3/4) Epoch 45, batch 1500, loss[loss=0.1445, simple_loss=0.2177, pruned_loss=0.03567, over 19794.00 frames. ], tot_loss[loss=0.1482, simple_loss=0.2302, pruned_loss=0.03306, over 3947602.86 frames. ], batch size: 126, lr: 3.45e-03, grad_scale: 16.0
+2023-03-29 20:00:17,873 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.217e+02 3.289e+02 3.880e+02 4.653e+02 8.208e+02, threshold=7.760e+02, percent-clipped=2.0
+2023-03-29 20:00:54,323 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0
+2023-03-29 20:00:55,916 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83149.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 20:01:45,177 INFO [train.py:892] (3/4) Epoch 45, batch 1550, loss[loss=0.16, simple_loss=0.2448, pruned_loss=0.03758, over 19727.00 frames. ], tot_loss[loss=0.1493, simple_loss=0.2317, pruned_loss=0.03348, over 3945856.79 frames. ], batch size: 259, lr: 3.45e-03, grad_scale: 16.0
+2023-03-29 20:02:46,130 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83196.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:02:48,068 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83197.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 20:03:45,838 INFO [train.py:892] (3/4) Epoch 45, batch 1600, loss[loss=0.1378, simple_loss=0.2149, pruned_loss=0.03038, over 19757.00 frames. ], tot_loss[loss=0.1488, simple_loss=0.2313, pruned_loss=0.0332, over 3945356.51 frames. ], batch size: 129, lr: 3.45e-03, grad_scale: 16.0
+2023-03-29 20:03:49,326 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3707, 3.7025, 3.8394, 4.4445, 3.0208, 3.3835, 2.7340, 2.7667],
+       device='cuda:3'), covar=tensor([0.0485, 0.1816, 0.0904, 0.0364, 0.1833, 0.1033, 0.1335, 0.1584],
+       device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0330, 0.0256, 0.0214, 0.0251, 0.0218, 0.0225, 0.0222],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 20:03:57,341 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83225.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:04:10,576 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0
+2023-03-29 20:04:15,359 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.230e+02 3.453e+02 4.061e+02 4.762e+02 8.632e+02, threshold=8.122e+02, percent-clipped=1.0
+2023-03-29 20:05:10,922 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83257.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:05:33,404 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0
+2023-03-29 20:05:41,846 INFO [train.py:892] (3/4) Epoch 45, batch 1650, loss[loss=0.1251, simple_loss=0.2043, pruned_loss=0.02297, over 19921.00 frames. ], tot_loss[loss=0.1475, simple_loss=0.2297, pruned_loss=0.03266, over 3946463.47 frames. ], batch size: 45, lr: 3.45e-03, grad_scale: 16.0
+2023-03-29 20:05:47,358 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83273.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:05:52,890 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0
+2023-03-29 20:06:11,750 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:07:29,139 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83319.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:07:33,426 INFO [train.py:892] (3/4) Epoch 45, batch 1700, loss[loss=0.1565, simple_loss=0.2383, pruned_loss=0.0374, over 19764.00 frames. ], tot_loss[loss=0.1484, simple_loss=0.2305, pruned_loss=0.03313, over 3947302.77 frames. ], batch size: 244, lr: 3.45e-03, grad_scale: 16.0
+2023-03-29 20:08:02,058 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.889e+02 3.340e+02 3.879e+02 4.594e+02 1.209e+03, threshold=7.757e+02, percent-clipped=3.0
+2023-03-29 20:08:28,649 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83345.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:09:04,008 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0
+2023-03-29 20:09:20,244 INFO [train.py:892] (3/4) Epoch 45, batch 1750, loss[loss=0.1309, simple_loss=0.2171, pruned_loss=0.02232, over 19709.00 frames. ], tot_loss[loss=0.1486, simple_loss=0.2306, pruned_loss=0.03331, over 3947563.60 frames. ], batch size: 101, lr: 3.44e-03, grad_scale: 16.0
+2023-03-29 20:10:31,543 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4230, 4.8998, 5.0978, 4.7346, 5.3797, 3.2920, 4.2841, 2.6246],
+       device='cuda:3'), covar=tensor([0.0163, 0.0208, 0.0137, 0.0211, 0.0123, 0.0999, 0.0851, 0.1508],
+       device='cuda:3'), in_proj_covar=tensor([0.0111, 0.0155, 0.0120, 0.0142, 0.0126, 0.0140, 0.0147, 0.0133],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003],
+       device='cuda:3')
+2023-03-29 20:10:59,579 INFO [train.py:892] (3/4) Epoch 45, batch 1800, loss[loss=0.1667, simple_loss=0.2436, pruned_loss=0.04487, over 19762.00 frames. ], tot_loss[loss=0.1484, simple_loss=0.2301, pruned_loss=0.03333, over 3948753.68 frames. ], batch size: 276, lr: 3.44e-03, grad_scale: 16.0
+2023-03-29 20:11:14,404 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.3718, 2.5743, 2.2808, 1.8555, 2.3846, 2.5337, 2.5184, 2.5879],
+       device='cuda:3'), covar=tensor([0.0457, 0.0409, 0.0450, 0.0690, 0.0485, 0.0380, 0.0374, 0.0317],
+       device='cuda:3'), in_proj_covar=tensor([0.0116, 0.0109, 0.0110, 0.0109, 0.0113, 0.0099, 0.0101, 0.0099],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 20:11:23,992 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.557e+02 3.459e+02 4.040e+02 5.185e+02 7.885e+02, threshold=8.081e+02, percent-clipped=1.0
+2023-03-29 20:12:17,130 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-29 20:12:35,865 INFO [train.py:892] (3/4) Epoch 45, batch 1850, loss[loss=0.1646, simple_loss=0.2441, pruned_loss=0.04257, over 19824.00 frames. ], tot_loss[loss=0.1487, simple_loss=0.2313, pruned_loss=0.03307, over 3946733.12 frames. ], batch size: 57, lr: 3.44e-03, grad_scale: 16.0
+2023-03-29 20:13:42,298 INFO [train.py:892] (3/4) Epoch 46, batch 0, loss[loss=0.1411, simple_loss=0.2125, pruned_loss=0.03489, over 19857.00 frames. ], tot_loss[loss=0.1411, simple_loss=0.2125, pruned_loss=0.03489, over 19857.00 frames. ], batch size: 118, lr: 3.40e-03, grad_scale: 16.0
+2023-03-29 20:13:42,299 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-29 20:14:19,282 INFO [train.py:926] (3/4) Epoch 46, validation: loss=0.1879, simple_loss=0.2498, pruned_loss=0.06295, over 2883724.00 frames. 
+2023-03-29 20:14:19,284 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-29 20:16:21,270 INFO [train.py:892] (3/4) Epoch 46, batch 50, loss[loss=0.1551, simple_loss=0.2362, pruned_loss=0.03697, over 19791.00 frames. ], tot_loss[loss=0.1474, simple_loss=0.2283, pruned_loss=0.03324, over 891584.97 frames. ], batch size: 83, lr: 3.40e-03, grad_scale: 16.0
+2023-03-29 20:16:38,139 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.206e+02 3.234e+02 3.698e+02 4.952e+02 1.024e+03, threshold=7.395e+02, percent-clipped=5.0
+2023-03-29 20:16:51,497 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0
+2023-03-29 20:17:21,536 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83552.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:17:25,663 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3649, 3.0628, 3.3826, 2.9538, 3.5742, 3.5575, 4.1356, 4.5874],
+       device='cuda:3'), covar=tensor([0.0581, 0.1697, 0.1537, 0.2230, 0.1696, 0.1492, 0.0698, 0.0586],
+       device='cuda:3'), in_proj_covar=tensor([0.0265, 0.0249, 0.0278, 0.0265, 0.0312, 0.0268, 0.0243, 0.0275],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 20:18:14,882 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2947, 2.5439, 4.5558, 3.8718, 4.1606, 4.4121, 4.2084, 4.1079],
+       device='cuda:3'), covar=tensor([0.0665, 0.1135, 0.0118, 0.0711, 0.0214, 0.0233, 0.0204, 0.0204],
+       device='cuda:3'), in_proj_covar=tensor([0.0106, 0.0108, 0.0093, 0.0155, 0.0091, 0.0105, 0.0095, 0.0091],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 20:18:16,040 INFO [train.py:892] (3/4) Epoch 46, batch 100, loss[loss=0.1438, simple_loss=0.23, pruned_loss=0.02879, over 19786.00 frames. ], tot_loss[loss=0.1444, simple_loss=0.2258, pruned_loss=0.03145, over 1570988.68 frames. ], batch size: 91, lr: 3.40e-03, grad_scale: 16.0
+2023-03-29 20:19:28,159 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2356, 2.6997, 3.2828, 3.3355, 3.8586, 4.3298, 4.0933, 4.2346],
+       device='cuda:3'), covar=tensor([0.0934, 0.1696, 0.1240, 0.0740, 0.0417, 0.0228, 0.0411, 0.0382],
+       device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0171, 0.0183, 0.0158, 0.0145, 0.0139, 0.0134, 0.0123],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 20:19:59,578 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83619.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:20:13,506 INFO [train.py:892] (3/4) Epoch 46, batch 150, loss[loss=0.1747, simple_loss=0.2669, pruned_loss=0.04126, over 19604.00 frames. ], tot_loss[loss=0.1457, simple_loss=0.2273, pruned_loss=0.03202, over 2097224.28 frames. ], batch size: 376, lr: 3.40e-03, grad_scale: 16.0
+2023-03-29 20:20:31,959 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.341e+02 3.214e+02 3.787e+02 4.556e+02 6.937e+02, threshold=7.575e+02, percent-clipped=0.0
+2023-03-29 20:20:51,720 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=83640.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:21:18,050 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.4510, 3.6648, 2.3041, 4.2284, 3.7646, 4.1943, 4.2946, 3.2614],
+       device='cuda:3'), covar=tensor([0.0621, 0.0537, 0.1453, 0.0520, 0.0609, 0.0442, 0.0506, 0.0837],
+       device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0152, 0.0149, 0.0163, 0.0141, 0.0148, 0.0157, 0.0155],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0003],
+       device='cuda:3')
+2023-03-29 20:21:25,340 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5197, 3.6015, 2.2660, 3.7277, 3.8548, 1.8668, 3.2540, 2.9694],
+       device='cuda:3'), covar=tensor([0.0897, 0.0990, 0.2929, 0.0911, 0.0699, 0.2689, 0.1176, 0.1072],
+       device='cuda:3'), in_proj_covar=tensor([0.0243, 0.0270, 0.0242, 0.0292, 0.0271, 0.0209, 0.0249, 0.0212],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 20:21:54,680 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83667.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:22:16,395 INFO [train.py:892] (3/4) Epoch 46, batch 200, loss[loss=0.1277, simple_loss=0.2096, pruned_loss=0.02289, over 19757.00 frames. ], tot_loss[loss=0.1473, simple_loss=0.2296, pruned_loss=0.03252, over 2506953.17 frames. ], batch size: 102, lr: 3.40e-03, grad_scale: 16.0
+2023-03-29 20:24:05,081 INFO [train.py:892] (3/4) Epoch 46, batch 250, loss[loss=0.1461, simple_loss=0.2267, pruned_loss=0.03277, over 19803.00 frames. ], tot_loss[loss=0.1473, simple_loss=0.2292, pruned_loss=0.03273, over 2826604.57 frames. ], batch size: 86, lr: 3.40e-03, grad_scale: 16.0
+2023-03-29 20:24:21,764 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.227e+02 3.614e+02 4.201e+02 5.267e+02 1.212e+03, threshold=8.403e+02, percent-clipped=3.0
+2023-03-29 20:25:26,481 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83760.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:26:06,556 INFO [train.py:892] (3/4) Epoch 46, batch 300, loss[loss=0.1888, simple_loss=0.2801, pruned_loss=0.04876, over 19619.00 frames. ], tot_loss[loss=0.1481, simple_loss=0.23, pruned_loss=0.03311, over 3075587.29 frames. ], batch size: 52, lr: 3.40e-03, grad_scale: 16.0
+2023-03-29 20:26:43,064 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0594, 2.7401, 3.2750, 3.2814, 3.7608, 4.1793, 3.9335, 4.1071],
+       device='cuda:3'), covar=tensor([0.0990, 0.1547, 0.1193, 0.0742, 0.0464, 0.0302, 0.0449, 0.0468],
+       device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0171, 0.0183, 0.0159, 0.0145, 0.0139, 0.0135, 0.0124],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 20:26:53,368 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.92 vs. limit=5.0
+2023-03-29 20:28:00,309 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=83821.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:28:11,289 INFO [train.py:892] (3/4) Epoch 46, batch 350, loss[loss=0.1925, simple_loss=0.2757, pruned_loss=0.05458, over 19631.00 frames. ], tot_loss[loss=0.1482, simple_loss=0.2302, pruned_loss=0.03306, over 3268961.46 frames. ], batch size: 387, lr: 3.40e-03, grad_scale: 16.0
+2023-03-29 20:28:31,193 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.272e+02 3.501e+02 4.159e+02 5.209e+02 9.522e+02, threshold=8.317e+02, percent-clipped=2.0
+2023-03-29 20:29:18,709 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83852.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:29:42,927 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0
+2023-03-29 20:30:09,191 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6237, 4.6973, 4.9758, 4.7853, 4.8798, 4.6460, 4.7656, 4.5265],
+       device='cuda:3'), covar=tensor([0.1478, 0.1467, 0.0895, 0.1329, 0.0860, 0.0817, 0.1772, 0.1993],
+       device='cuda:3'), in_proj_covar=tensor([0.0306, 0.0356, 0.0387, 0.0321, 0.0295, 0.0301, 0.0379, 0.0408],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 20:30:17,358 INFO [train.py:892] (3/4) Epoch 46, batch 400, loss[loss=0.1409, simple_loss=0.2139, pruned_loss=0.03395, over 19727.00 frames. ], tot_loss[loss=0.1479, simple_loss=0.2298, pruned_loss=0.03295, over 3420305.40 frames. ], batch size: 134, lr: 3.40e-03, grad_scale: 16.0
+2023-03-29 20:31:22,493 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83900.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:31:59,067 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2012, 2.6225, 3.1302, 3.4399, 3.8849, 4.4674, 4.2881, 4.3429],
+       device='cuda:3'), covar=tensor([0.0945, 0.1952, 0.1495, 0.0685, 0.0435, 0.0256, 0.0386, 0.0385],
+       device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0172, 0.0184, 0.0159, 0.0145, 0.0140, 0.0135, 0.0124],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 20:32:25,447 INFO [train.py:892] (3/4) Epoch 46, batch 450, loss[loss=0.1715, simple_loss=0.2412, pruned_loss=0.05087, over 19844.00 frames. ], tot_loss[loss=0.1489, simple_loss=0.2309, pruned_loss=0.03341, over 3536687.04 frames. ], batch size: 161, lr: 3.40e-03, grad_scale: 8.0
+2023-03-29 20:32:48,231 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.948e+02 3.420e+02 3.775e+02 4.552e+02 8.094e+02, threshold=7.550e+02, percent-clipped=0.0
+2023-03-29 20:33:01,832 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=83940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:33:23,143 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83949.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:34:29,741 INFO [train.py:892] (3/4) Epoch 46, batch 500, loss[loss=0.1311, simple_loss=0.2162, pruned_loss=0.02304, over 19765.00 frames. ], tot_loss[loss=0.1491, simple_loss=0.231, pruned_loss=0.03361, over 3628166.44 frames. ], batch size: 100, lr: 3.39e-03, grad_scale: 8.0
+2023-03-29 20:34:51,645 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83985.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:34:59,176 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=83988.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:34:59,879 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.43 vs. limit=5.0
+2023-03-29 20:35:55,163 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.0796, 3.2917, 3.3832, 3.3242, 3.1815, 3.3331, 3.2108, 3.2894],
+       device='cuda:3'), covar=tensor([0.0438, 0.0300, 0.0340, 0.0278, 0.0459, 0.0331, 0.0388, 0.0457],
+       device='cuda:3'), in_proj_covar=tensor([0.0099, 0.0092, 0.0095, 0.0089, 0.0102, 0.0095, 0.0111, 0.0083],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 20:35:59,571 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84010.0, num_to_drop=1, layers_to_drop={2}
+2023-03-29 20:36:35,440 INFO [train.py:892] (3/4) Epoch 46, batch 550, loss[loss=0.1622, simple_loss=0.2532, pruned_loss=0.03559, over 19665.00 frames. ], tot_loss[loss=0.1501, simple_loss=0.2317, pruned_loss=0.03425, over 3699579.57 frames. ], batch size: 343, lr: 3.39e-03, grad_scale: 8.0
+2023-03-29 20:36:58,613 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.358e+02 3.554e+02 4.105e+02 5.072e+02 7.939e+02, threshold=8.210e+02, percent-clipped=2.0
+2023-03-29 20:37:27,549 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84046.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:38:42,405 INFO [train.py:892] (3/4) Epoch 46, batch 600, loss[loss=0.1789, simple_loss=0.2949, pruned_loss=0.03142, over 18960.00 frames. ], tot_loss[loss=0.1501, simple_loss=0.2319, pruned_loss=0.03417, over 3753459.17 frames. ], batch size: 514, lr: 3.39e-03, grad_scale: 8.0
+2023-03-29 20:38:56,797 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.09 vs. limit=2.0
+2023-03-29 20:39:34,333 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84097.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:40:17,171 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84116.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:40:40,128 INFO [train.py:892] (3/4) Epoch 46, batch 650, loss[loss=0.1128, simple_loss=0.1893, pruned_loss=0.01814, over 19774.00 frames. ], tot_loss[loss=0.1492, simple_loss=0.2314, pruned_loss=0.0335, over 3797170.11 frames. ], batch size: 116, lr: 3.39e-03, grad_scale: 8.0
+2023-03-29 20:40:55,785 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8642, 5.1738, 5.2543, 5.1612, 4.8842, 5.2392, 4.7064, 4.7390],
+       device='cuda:3'), covar=tensor([0.0514, 0.0497, 0.0459, 0.0418, 0.0613, 0.0476, 0.0729, 0.0965],
+       device='cuda:3'), in_proj_covar=tensor([0.0296, 0.0314, 0.0323, 0.0283, 0.0294, 0.0275, 0.0288, 0.0336],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 20:41:04,421 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.462e+02 3.455e+02 4.092e+02 4.989e+02 7.207e+02, threshold=8.185e+02, percent-clipped=0.0
+2023-03-29 20:41:39,396 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7159, 3.5282, 3.7780, 2.9685, 3.9852, 3.3368, 3.4563, 3.8713],
+       device='cuda:3'), covar=tensor([0.0732, 0.0465, 0.0746, 0.0780, 0.0408, 0.0495, 0.0549, 0.0366],
+       device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0096, 0.0093, 0.0116, 0.0088, 0.0092, 0.0088, 0.0083],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 20:42:03,732 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84158.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:42:22,350 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1829, 3.2316, 2.0745, 3.7619, 3.4842, 3.7266, 3.7929, 3.0299],
+       device='cuda:3'), covar=tensor([0.0701, 0.0700, 0.1762, 0.0651, 0.0604, 0.0448, 0.0571, 0.0830],
+       device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0151, 0.0148, 0.0162, 0.0141, 0.0148, 0.0157, 0.0155],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0003],
+       device='cuda:3')
+2023-03-29 20:42:46,710 INFO [train.py:892] (3/4) Epoch 46, batch 700, loss[loss=0.1363, simple_loss=0.2212, pruned_loss=0.02566, over 19747.00 frames. ], tot_loss[loss=0.1493, simple_loss=0.2315, pruned_loss=0.03358, over 3831112.53 frames. ], batch size: 89, lr: 3.39e-03, grad_scale: 8.0
+2023-03-29 20:43:45,645 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84199.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:44:54,100 INFO [train.py:892] (3/4) Epoch 46, batch 750, loss[loss=0.1262, simple_loss=0.2067, pruned_loss=0.0228, over 19896.00 frames. ], tot_loss[loss=0.1498, simple_loss=0.2316, pruned_loss=0.034, over 3857387.84 frames. ], batch size: 91, lr: 3.39e-03, grad_scale: 8.0
+2023-03-29 20:45:03,484 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4319, 4.2626, 4.7555, 4.3251, 4.0075, 4.5436, 4.4144, 4.8144],
+       device='cuda:3'), covar=tensor([0.0723, 0.0383, 0.0358, 0.0393, 0.0924, 0.0525, 0.0477, 0.0317],
+       device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0233, 0.0235, 0.0246, 0.0216, 0.0260, 0.0248, 0.0232],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 20:45:18,696 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.966e+02 3.322e+02 4.189e+02 5.032e+02 8.486e+02, threshold=8.378e+02, percent-clipped=1.0
+2023-03-29 20:45:22,620 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:46:21,979 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84260.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:46:38,596 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6774, 4.7102, 2.7366, 4.9693, 5.1985, 2.3089, 4.4181, 3.7312],
+       device='cuda:3'), covar=tensor([0.0567, 0.0712, 0.2792, 0.0780, 0.0466, 0.2884, 0.0923, 0.0902],
+       device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0268, 0.0240, 0.0289, 0.0269, 0.0208, 0.0247, 0.0211],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 20:46:56,842 INFO [train.py:892] (3/4) Epoch 46, batch 800, loss[loss=0.1438, simple_loss=0.2258, pruned_loss=0.0309, over 19748.00 frames. ], tot_loss[loss=0.1488, simple_loss=0.2306, pruned_loss=0.03352, over 3877759.04 frames. ], batch size: 259, lr: 3.39e-03, grad_scale: 8.0
+2023-03-29 20:47:33,163 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4249, 4.4358, 2.7395, 4.7293, 4.9370, 2.2622, 4.3209, 3.6883],
+       device='cuda:3'), covar=tensor([0.0614, 0.0871, 0.2491, 0.0711, 0.0510, 0.2552, 0.0822, 0.0851],
+       device='cuda:3'), in_proj_covar=tensor([0.0242, 0.0269, 0.0241, 0.0290, 0.0270, 0.0208, 0.0247, 0.0211],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 20:47:48,414 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84297.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:48:06,952 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84305.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 20:48:46,263 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.11 vs. limit=2.0
+2023-03-29 20:48:59,469 INFO [train.py:892] (3/4) Epoch 46, batch 850, loss[loss=0.14, simple_loss=0.2263, pruned_loss=0.02682, over 19672.00 frames. ], tot_loss[loss=0.1484, simple_loss=0.2305, pruned_loss=0.03314, over 3894430.88 frames. ], batch size: 73, lr: 3.39e-03, grad_scale: 8.0
+2023-03-29 20:49:22,805 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.218e+02 3.307e+02 4.048e+02 4.791e+02 1.064e+03, threshold=8.096e+02, percent-clipped=1.0
+2023-03-29 20:49:27,537 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84336.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:49:41,443 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84341.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:49:46,806 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.08 vs. limit=2.0
+2023-03-29 20:51:08,461 INFO [train.py:892] (3/4) Epoch 46, batch 900, loss[loss=0.1441, simple_loss=0.2284, pruned_loss=0.02988, over 19763.00 frames. ], tot_loss[loss=0.1485, simple_loss=0.2308, pruned_loss=0.03309, over 3905665.37 frames. ], batch size: 88, lr: 3.39e-03, grad_scale: 8.0
+2023-03-29 20:51:40,992 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4329, 1.9184, 2.1675, 2.6778, 2.9596, 3.0592, 2.9169, 2.9548],
+       device='cuda:3'), covar=tensor([0.1180, 0.1866, 0.1625, 0.0812, 0.0595, 0.0476, 0.0550, 0.0586],
+       device='cuda:3'), in_proj_covar=tensor([0.0168, 0.0172, 0.0184, 0.0160, 0.0145, 0.0140, 0.0136, 0.0125],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 20:52:00,060 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84397.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:52:47,321 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84416.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:53:09,207 INFO [train.py:892] (3/4) Epoch 46, batch 950, loss[loss=0.1438, simple_loss=0.221, pruned_loss=0.03331, over 19815.00 frames. ], tot_loss[loss=0.149, simple_loss=0.2313, pruned_loss=0.03339, over 3916297.13 frames. ], batch size: 181, lr: 3.39e-03, grad_scale: 8.0
+2023-03-29 20:53:32,152 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.409e+02 3.247e+02 3.846e+02 4.740e+02 1.002e+03, threshold=7.692e+02, percent-clipped=2.0
+2023-03-29 20:54:17,733 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84453.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:54:45,384 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84464.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:55:13,053 INFO [train.py:892] (3/4) Epoch 46, batch 1000, loss[loss=0.1325, simple_loss=0.2137, pruned_loss=0.02563, over 19817.00 frames. ], tot_loss[loss=0.1493, simple_loss=0.2313, pruned_loss=0.0336, over 3922487.04 frames. ], batch size: 133, lr: 3.38e-03, grad_scale: 8.0
+2023-03-29 20:56:23,790 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6368, 4.4918, 4.9912, 4.5222, 4.1347, 4.8037, 4.6819, 5.1200],
+       device='cuda:3'), covar=tensor([0.0849, 0.0394, 0.0338, 0.0388, 0.0846, 0.0496, 0.0456, 0.0326],
+       device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0234, 0.0236, 0.0247, 0.0217, 0.0262, 0.0248, 0.0233],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 20:57:09,432 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84525.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:57:10,330 INFO [train.py:892] (3/4) Epoch 46, batch 1050, loss[loss=0.1401, simple_loss=0.2185, pruned_loss=0.03084, over 19901.00 frames. ], tot_loss[loss=0.1496, simple_loss=0.2321, pruned_loss=0.03354, over 3928164.31 frames. ], batch size: 113, lr: 3.38e-03, grad_scale: 8.0
+2023-03-29 20:57:31,620 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.128e+02 3.422e+02 4.030e+02 5.116e+02 8.679e+02, threshold=8.059e+02, percent-clipped=4.0
+2023-03-29 20:58:20,668 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84555.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:59:10,800 INFO [train.py:892] (3/4) Epoch 46, batch 1100, loss[loss=0.1635, simple_loss=0.2527, pruned_loss=0.03718, over 19884.00 frames. ], tot_loss[loss=0.1488, simple_loss=0.231, pruned_loss=0.03329, over 3932230.24 frames. ], batch size: 84, lr: 3.38e-03, grad_scale: 8.0
+2023-03-29 20:59:34,895 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 20:59:49,723 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84592.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:00:23,221 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:00:52,813 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0
+2023-03-29 21:01:12,928 INFO [train.py:892] (3/4) Epoch 46, batch 1150, loss[loss=0.1429, simple_loss=0.2274, pruned_loss=0.02919, over 19580.00 frames. ], tot_loss[loss=0.1479, simple_loss=0.2301, pruned_loss=0.03289, over 3936742.44 frames. ], batch size: 53, lr: 3.38e-03, grad_scale: 8.0
+2023-03-29 21:01:31,628 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84633.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:01:34,947 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.372e+02 3.414e+02 3.912e+02 4.793e+02 1.001e+03, threshold=7.825e+02, percent-clipped=1.0
+2023-03-29 21:01:53,032 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:02:22,584 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84653.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:03:17,294 INFO [train.py:892] (3/4) Epoch 46, batch 1200, loss[loss=0.1541, simple_loss=0.2377, pruned_loss=0.03524, over 19802.00 frames. ], tot_loss[loss=0.1486, simple_loss=0.2307, pruned_loss=0.03324, over 3938340.68 frames. ], batch size: 74, lr: 3.38e-03, grad_scale: 8.0
+2023-03-29 21:03:50,241 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84689.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:03:56,738 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84692.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:04:02,548 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84694.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:05:19,009 INFO [train.py:892] (3/4) Epoch 46, batch 1250, loss[loss=0.1312, simple_loss=0.2096, pruned_loss=0.02637, over 19759.00 frames. ], tot_loss[loss=0.1487, simple_loss=0.2312, pruned_loss=0.0331, over 3938764.14 frames. ], batch size: 102, lr: 3.38e-03, grad_scale: 8.0
+2023-03-29 21:05:20,083 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7685, 4.4497, 4.5361, 4.2370, 4.7360, 3.2524, 3.9695, 2.3555],
+       device='cuda:3'), covar=tensor([0.0155, 0.0229, 0.0142, 0.0191, 0.0137, 0.0956, 0.0731, 0.1476],
+       device='cuda:3'), in_proj_covar=tensor([0.0110, 0.0156, 0.0119, 0.0142, 0.0125, 0.0141, 0.0147, 0.0134],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003],
+       device='cuda:3')
+2023-03-29 21:05:39,514 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.527e+02 3.466e+02 3.877e+02 4.842e+02 1.121e+03, threshold=7.755e+02, percent-clipped=3.0
+2023-03-29 21:06:28,620 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84753.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:06:39,816 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.1404, 2.4238, 2.2241, 1.6417, 2.2459, 2.4061, 2.2821, 2.3404],
+       device='cuda:3'), covar=tensor([0.0504, 0.0340, 0.0400, 0.0632, 0.0474, 0.0373, 0.0383, 0.0377],
+       device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0109, 0.0110, 0.0110, 0.0113, 0.0100, 0.0101, 0.0100],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 21:07:13,230 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8988, 3.2155, 3.3512, 3.8162, 2.6175, 3.1357, 2.5268, 2.4969],
+       device='cuda:3'), covar=tensor([0.0601, 0.1910, 0.1068, 0.0540, 0.2105, 0.1080, 0.1497, 0.1696],
+       device='cuda:3'), in_proj_covar=tensor([0.0257, 0.0332, 0.0258, 0.0216, 0.0252, 0.0220, 0.0227, 0.0223],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 21:07:25,197 INFO [train.py:892] (3/4) Epoch 46, batch 1300, loss[loss=0.1475, simple_loss=0.231, pruned_loss=0.03202, over 19678.00 frames. ], tot_loss[loss=0.1481, simple_loss=0.2306, pruned_loss=0.03284, over 3940416.98 frames. ], batch size: 49, lr: 3.38e-03, grad_scale: 8.0
+2023-03-29 21:07:38,787 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([1.9847, 2.1760, 1.9776, 1.4191, 1.9978, 2.1433, 2.0393, 2.1251],
+       device='cuda:3'), covar=tensor([0.0529, 0.0411, 0.0457, 0.0719, 0.0498, 0.0399, 0.0406, 0.0379],
+       device='cuda:3'), in_proj_covar=tensor([0.0117, 0.0109, 0.0110, 0.0110, 0.0114, 0.0100, 0.0101, 0.0100],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002],
+       device='cuda:3')
+2023-03-29 21:08:25,875 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84800.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:08:27,885 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84801.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:09:27,329 INFO [train.py:892] (3/4) Epoch 46, batch 1350, loss[loss=0.1499, simple_loss=0.226, pruned_loss=0.03685, over 19782.00 frames. ], tot_loss[loss=0.1478, simple_loss=0.2304, pruned_loss=0.0326, over 3943277.94 frames. ], batch size: 163, lr: 3.38e-03, grad_scale: 8.0
+2023-03-29 21:09:40,676 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0217, 4.8243, 5.4067, 4.9151, 4.3311, 5.1538, 5.0868, 5.5591],
+       device='cuda:3'), covar=tensor([0.0820, 0.0392, 0.0311, 0.0368, 0.0796, 0.0433, 0.0423, 0.0295],
+       device='cuda:3'), in_proj_covar=tensor([0.0290, 0.0233, 0.0235, 0.0246, 0.0216, 0.0261, 0.0248, 0.0232],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 21:09:43,213 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8077, 3.9263, 2.2937, 4.0771, 4.1988, 1.9603, 3.5395, 3.1466],
+       device='cuda:3'), covar=tensor([0.0793, 0.0796, 0.2875, 0.0821, 0.0590, 0.2733, 0.1028, 0.0948],
+       device='cuda:3'), in_proj_covar=tensor([0.0244, 0.0271, 0.0241, 0.0291, 0.0272, 0.0210, 0.0248, 0.0212],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 21:09:50,710 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.135e+02 3.546e+02 4.239e+02 5.224e+02 8.019e+02, threshold=8.477e+02, percent-clipped=4.0
+2023-03-29 21:10:05,567 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0
+2023-03-29 21:10:45,745 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84855.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:10:48,289 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2173, 2.9521, 3.3546, 2.8385, 3.4158, 3.3010, 4.0038, 4.4042],
+       device='cuda:3'), covar=tensor([0.0547, 0.1727, 0.1443, 0.2263, 0.1708, 0.1529, 0.0600, 0.0576],
+       device='cuda:3'), in_proj_covar=tensor([0.0265, 0.0248, 0.0276, 0.0265, 0.0311, 0.0268, 0.0242, 0.0275],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 21:11:00,568 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=84861.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:11:35,523 INFO [train.py:892] (3/4) Epoch 46, batch 1400, loss[loss=0.1245, simple_loss=0.2086, pruned_loss=0.02017, over 19749.00 frames. ], tot_loss[loss=0.1488, simple_loss=0.2313, pruned_loss=0.03319, over 3944472.32 frames. ], batch size: 100, lr: 3.38e-03, grad_scale: 8.0
+2023-03-29 21:11:46,574 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84881.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:11:46,823 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5896, 2.8491, 3.0456, 3.4138, 2.3930, 2.8619, 2.3611, 2.3301],
+       device='cuda:3'), covar=tensor([0.0585, 0.1691, 0.1080, 0.0575, 0.2060, 0.1063, 0.1478, 0.1657],
+       device='cuda:3'), in_proj_covar=tensor([0.0257, 0.0333, 0.0259, 0.0216, 0.0252, 0.0220, 0.0227, 0.0223],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 21:11:55,620 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4989, 1.9505, 2.1540, 2.7009, 3.0223, 3.1159, 2.9884, 3.0779],
+       device='cuda:3'), covar=tensor([0.1151, 0.1899, 0.1676, 0.0823, 0.0557, 0.0442, 0.0553, 0.0550],
+       device='cuda:3'), in_proj_covar=tensor([0.0167, 0.0171, 0.0183, 0.0159, 0.0145, 0.0139, 0.0135, 0.0124],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 21:12:09,161 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84892.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:12:20,507 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.3426, 4.8901, 4.9383, 4.6444, 5.2897, 3.1191, 4.2061, 2.4855],
+       device='cuda:3'), covar=tensor([0.0146, 0.0201, 0.0144, 0.0206, 0.0121, 0.1141, 0.0847, 0.1622],
+       device='cuda:3'), in_proj_covar=tensor([0.0111, 0.0156, 0.0119, 0.0143, 0.0125, 0.0140, 0.0147, 0.0134],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003],
+       device='cuda:3')
+2023-03-29 21:12:38,649 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84903.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:12:43,610 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5868, 2.5929, 5.0282, 4.2046, 4.6992, 4.8992, 4.7847, 4.5365],
+       device='cuda:3'), covar=tensor([0.0621, 0.1104, 0.0091, 0.0819, 0.0139, 0.0229, 0.0156, 0.0179],
+       device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0107, 0.0092, 0.0153, 0.0091, 0.0105, 0.0094, 0.0091],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 21:13:32,223 INFO [train.py:892] (3/4) Epoch 46, batch 1450, loss[loss=0.1405, simple_loss=0.2325, pruned_loss=0.02426, over 19747.00 frames. ], tot_loss[loss=0.148, simple_loss=0.2302, pruned_loss=0.03291, over 3945865.59 frames. ], batch size: 97, lr: 3.38e-03, grad_scale: 8.0
+2023-03-29 21:13:55,749 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.367e+02 3.403e+02 4.016e+02 4.651e+02 7.911e+02, threshold=8.032e+02, percent-clipped=0.0
+2023-03-29 21:14:08,225 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=84940.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:14:36,326 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=84951.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:15:44,114 INFO [train.py:892] (3/4) Epoch 46, batch 1500, loss[loss=0.1695, simple_loss=0.2852, pruned_loss=0.02695, over 18663.00 frames. ], tot_loss[loss=0.1481, simple_loss=0.2303, pruned_loss=0.03296, over 3944668.86 frames. ], batch size: 564, lr: 3.37e-03, grad_scale: 8.0
+2023-03-29 21:16:15,607 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=84989.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:16:22,954 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=84992.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:17:16,957 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85012.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:17:49,520 INFO [train.py:892] (3/4) Epoch 46, batch 1550, loss[loss=0.1613, simple_loss=0.2398, pruned_loss=0.04138, over 19643.00 frames. ], tot_loss[loss=0.1484, simple_loss=0.2309, pruned_loss=0.03293, over 3945054.95 frames. ], batch size: 299, lr: 3.37e-03, grad_scale: 8.0
+2023-03-29 21:17:58,239 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7189, 3.6604, 3.5981, 3.3542, 3.7185, 2.6837, 3.0240, 1.7578],
+       device='cuda:3'), covar=tensor([0.0219, 0.0245, 0.0175, 0.0220, 0.0166, 0.1317, 0.0665, 0.1780],
+       device='cuda:3'), in_proj_covar=tensor([0.0111, 0.0156, 0.0119, 0.0143, 0.0125, 0.0140, 0.0147, 0.0134],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003],
+       device='cuda:3')
+2023-03-29 21:18:12,480 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.101e+02 3.669e+02 4.198e+02 5.006e+02 8.856e+02, threshold=8.396e+02, percent-clipped=3.0
+2023-03-29 21:18:26,368 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85040.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:19:20,693 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1019, 3.9675, 4.4231, 4.0007, 3.7821, 4.2682, 4.1023, 4.4943],
+       device='cuda:3'), covar=tensor([0.0960, 0.0456, 0.0431, 0.0460, 0.1124, 0.0679, 0.0523, 0.0407],
+       device='cuda:3'), in_proj_covar=tensor([0.0291, 0.0233, 0.0236, 0.0247, 0.0215, 0.0261, 0.0247, 0.0232],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 21:19:59,491 INFO [train.py:892] (3/4) Epoch 46, batch 1600, loss[loss=0.1603, simple_loss=0.2487, pruned_loss=0.03595, over 19710.00 frames. ], tot_loss[loss=0.1482, simple_loss=0.2306, pruned_loss=0.0329, over 3946987.41 frames. ], batch size: 310, lr: 3.37e-03, grad_scale: 8.0
+2023-03-29 21:21:21,544 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85107.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:21:42,284 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85115.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:22:09,289 INFO [train.py:892] (3/4) Epoch 46, batch 1650, loss[loss=0.1473, simple_loss=0.2393, pruned_loss=0.02761, over 19915.00 frames. ], tot_loss[loss=0.1479, simple_loss=0.2306, pruned_loss=0.03262, over 3946459.42 frames. ], batch size: 53, lr: 3.37e-03, grad_scale: 8.0
+2023-03-29 21:22:31,375 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.611e+02 3.617e+02 4.144e+02 5.075e+02 9.786e+02, threshold=8.288e+02, percent-clipped=2.0
+2023-03-29 21:23:27,156 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85156.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:23:57,604 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85168.0, num_to_drop=1, layers_to_drop={3}
+2023-03-29 21:24:15,563 INFO [train.py:892] (3/4) Epoch 46, batch 1700, loss[loss=0.1363, simple_loss=0.2126, pruned_loss=0.02999, over 19835.00 frames. ], tot_loss[loss=0.1471, simple_loss=0.2295, pruned_loss=0.03238, over 3947869.33 frames. ], batch size: 75, lr: 3.37e-03, grad_scale: 8.0
+2023-03-29 21:24:16,928 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85176.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:24:29,902 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85181.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:26:14,827 INFO [train.py:892] (3/4) Epoch 46, batch 1750, loss[loss=0.1503, simple_loss=0.2424, pruned_loss=0.02907, over 19716.00 frames. ], tot_loss[loss=0.1468, simple_loss=0.2293, pruned_loss=0.03217, over 3948096.87 frames. ], batch size: 54, lr: 3.37e-03, grad_scale: 8.0
+2023-03-29 21:26:22,878 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85229.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:26:35,388 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.122e+02 3.389e+02 3.888e+02 4.671e+02 1.140e+03, threshold=7.776e+02, percent-clipped=1.0
+2023-03-29 21:28:03,447 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6687, 2.7092, 1.8217, 3.0278, 2.8481, 2.9467, 3.0720, 2.4820],
+       device='cuda:3'), covar=tensor([0.0751, 0.0846, 0.1524, 0.0760, 0.0695, 0.0670, 0.0642, 0.0977],
+       device='cuda:3'), in_proj_covar=tensor([0.0151, 0.0152, 0.0148, 0.0162, 0.0140, 0.0147, 0.0156, 0.0154],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 21:28:04,586 INFO [train.py:892] (3/4) Epoch 46, batch 1800, loss[loss=0.1535, simple_loss=0.2308, pruned_loss=0.03813, over 19789.00 frames. ], tot_loss[loss=0.1473, simple_loss=0.2294, pruned_loss=0.03258, over 3947300.92 frames. ], batch size: 236, lr: 3.37e-03, grad_scale: 8.0
+2023-03-29 21:28:25,860 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1074, 2.4675, 3.3259, 2.7643, 2.8593, 2.7718, 2.1155, 2.2786],
+       device='cuda:3'), covar=tensor([0.1299, 0.2961, 0.0749, 0.1247, 0.1931, 0.1702, 0.2848, 0.2717],
+       device='cuda:3'), in_proj_covar=tensor([0.0361, 0.0407, 0.0358, 0.0301, 0.0383, 0.0405, 0.0393, 0.0370],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 21:28:30,792 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85289.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:29:00,501 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:29:37,980 INFO [train.py:892] (3/4) Epoch 46, batch 1850, loss[loss=0.1534, simple_loss=0.2462, pruned_loss=0.0303, over 19831.00 frames. ], tot_loss[loss=0.1482, simple_loss=0.2309, pruned_loss=0.03271, over 3947050.31 frames. ], batch size: 57, lr: 3.37e-03, grad_scale: 8.0
+2023-03-29 21:30:45,944 INFO [train.py:892] (3/4) Epoch 47, batch 0, loss[loss=0.1382, simple_loss=0.2081, pruned_loss=0.03418, over 19830.00 frames. ], tot_loss[loss=0.1382, simple_loss=0.2081, pruned_loss=0.03418, over 19830.00 frames. ], batch size: 127, lr: 3.33e-03, grad_scale: 8.0
+2023-03-29 21:30:45,945 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-29 21:31:20,079 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1274, 2.9478, 4.8571, 3.5560, 3.8321, 3.3597, 2.5464, 2.6947],
+       device='cuda:3'), covar=tensor([0.1025, 0.3624, 0.0402, 0.1080, 0.1957, 0.1767, 0.3166, 0.2962],
+       device='cuda:3'), in_proj_covar=tensor([0.0361, 0.0408, 0.0358, 0.0301, 0.0383, 0.0406, 0.0394, 0.0371],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 21:31:22,863 INFO [train.py:926] (3/4) Epoch 47, validation: loss=0.1894, simple_loss=0.2504, pruned_loss=0.06424, over 2883724.00 frames. 
+2023-03-29 21:31:22,865 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-29 21:31:31,557 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.328e+02 3.387e+02 4.010e+02 4.758e+02 1.602e+03, threshold=8.020e+02, percent-clipped=2.0
+2023-03-29 21:31:39,706 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85337.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:33:29,576 INFO [train.py:892] (3/4) Epoch 47, batch 50, loss[loss=0.1613, simple_loss=0.2454, pruned_loss=0.03861, over 19784.00 frames. ], tot_loss[loss=0.1493, simple_loss=0.2306, pruned_loss=0.03396, over 888624.06 frames. ], batch size: 241, lr: 3.33e-03, grad_scale: 8.0
+2023-03-29 21:35:26,323 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85428.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:35:31,485 INFO [train.py:892] (3/4) Epoch 47, batch 100, loss[loss=0.1456, simple_loss=0.2341, pruned_loss=0.02859, over 19750.00 frames. ], tot_loss[loss=0.1476, simple_loss=0.229, pruned_loss=0.03311, over 1566682.21 frames. ], batch size: 250, lr: 3.33e-03, grad_scale: 8.0
+2023-03-29 21:35:42,066 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.536e+02 3.357e+02 4.311e+02 4.946e+02 7.786e+02, threshold=8.621e+02, percent-clipped=0.0
+2023-03-29 21:36:34,056 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85456.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:36:50,171 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85463.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 21:37:09,001 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85471.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:37:33,150 INFO [train.py:892] (3/4) Epoch 47, batch 150, loss[loss=0.1321, simple_loss=0.2187, pruned_loss=0.0228, over 19747.00 frames. ], tot_loss[loss=0.1495, simple_loss=0.2314, pruned_loss=0.03384, over 2093976.60 frames. ], batch size: 97, lr: 3.33e-03, grad_scale: 8.0
+2023-03-29 21:37:37,906 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8236, 4.0929, 4.3420, 4.9109, 3.0343, 3.4953, 2.9955, 3.0119],
+       device='cuda:3'), covar=tensor([0.0424, 0.1702, 0.0723, 0.0325, 0.2073, 0.1165, 0.1312, 0.1523],
+       device='cuda:3'), in_proj_covar=tensor([0.0257, 0.0331, 0.0258, 0.0216, 0.0253, 0.0220, 0.0227, 0.0223],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 21:37:40,152 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85483.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:37:58,415 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85489.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:38:35,689 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85504.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:38:40,929 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0743, 4.2595, 2.4916, 4.4431, 4.6159, 1.9513, 3.9393, 3.3517],
+       device='cuda:3'), covar=tensor([0.0705, 0.0750, 0.2719, 0.0702, 0.0526, 0.2871, 0.0941, 0.0976],
+       device='cuda:3'), in_proj_covar=tensor([0.0245, 0.0272, 0.0241, 0.0293, 0.0272, 0.0210, 0.0249, 0.0213],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 21:39:40,834 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85530.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:39:42,161 INFO [train.py:892] (3/4) Epoch 47, batch 200, loss[loss=0.1338, simple_loss=0.2196, pruned_loss=0.02406, over 19660.00 frames. ], tot_loss[loss=0.1476, simple_loss=0.2298, pruned_loss=0.03271, over 2507022.16 frames. ], batch size: 58, lr: 3.33e-03, grad_scale: 8.0
+2023-03-29 21:39:43,521 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.8393, 3.8611, 2.3325, 4.0129, 4.2228, 1.9380, 3.5037, 3.2254],
+       device='cuda:3'), covar=tensor([0.0721, 0.0935, 0.2960, 0.0870, 0.0659, 0.2905, 0.1158, 0.0939],
+       device='cuda:3'), in_proj_covar=tensor([0.0246, 0.0273, 0.0242, 0.0295, 0.0274, 0.0212, 0.0251, 0.0214],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 21:39:50,775 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.307e+02 3.468e+02 4.248e+02 4.914e+02 6.825e+02, threshold=8.497e+02, percent-clipped=0.0
+2023-03-29 21:40:12,968 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85544.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:40:26,457 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.56 vs. limit=2.0
+2023-03-29 21:40:39,463 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=2.32 vs. limit=5.0
+2023-03-29 21:41:45,862 INFO [train.py:892] (3/4) Epoch 47, batch 250, loss[loss=0.1353, simple_loss=0.2142, pruned_loss=0.02825, over 19834.00 frames. ], tot_loss[loss=0.1469, simple_loss=0.2296, pruned_loss=0.03213, over 2827375.18 frames. ], batch size: 90, lr: 3.33e-03, grad_scale: 8.0
+2023-03-29 21:42:12,531 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85591.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:42:54,931 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85607.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:43:36,270 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.2203, 3.1164, 3.2937, 2.6760, 3.4038, 2.9738, 3.2506, 3.2959],
+       device='cuda:3'), covar=tensor([0.0586, 0.0520, 0.0582, 0.0775, 0.0378, 0.0474, 0.0448, 0.0428],
+       device='cuda:3'), in_proj_covar=tensor([0.0086, 0.0096, 0.0092, 0.0115, 0.0088, 0.0091, 0.0087, 0.0083],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 21:43:40,836 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85625.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:43:54,575 INFO [train.py:892] (3/4) Epoch 47, batch 300, loss[loss=0.1473, simple_loss=0.2335, pruned_loss=0.03055, over 19676.00 frames. ], tot_loss[loss=0.1472, simple_loss=0.23, pruned_loss=0.03224, over 3075546.11 frames. ], batch size: 64, lr: 3.33e-03, grad_scale: 8.0
+2023-03-29 21:44:06,337 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.290e+02 3.428e+02 3.993e+02 4.702e+02 7.624e+02, threshold=7.986e+02, percent-clipped=0.0
+2023-03-29 21:44:30,328 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0
+2023-03-29 21:44:50,631 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85655.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:45:13,920 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6245, 2.7046, 4.2341, 3.2124, 3.4264, 3.1419, 2.3876, 2.4658],
+       device='cuda:3'), covar=tensor([0.1271, 0.3534, 0.0552, 0.1237, 0.1950, 0.1862, 0.2963, 0.3116],
+       device='cuda:3'), in_proj_covar=tensor([0.0361, 0.0408, 0.0358, 0.0301, 0.0382, 0.0406, 0.0394, 0.0371],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 21:45:49,360 INFO [train.py:892] (3/4) Epoch 47, batch 350, loss[loss=0.1318, simple_loss=0.2086, pruned_loss=0.02746, over 19766.00 frames. ], tot_loss[loss=0.1469, simple_loss=0.2294, pruned_loss=0.03216, over 3268963.54 frames. ], batch size: 116, lr: 3.32e-03, grad_scale: 8.0
+2023-03-29 21:46:02,523 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85686.0, num_to_drop=1, layers_to_drop={3}
+2023-03-29 21:47:05,365 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.5515, 3.8302, 4.0188, 4.6658, 3.0186, 3.3351, 3.2088, 2.9376],
+       device='cuda:3'), covar=tensor([0.0546, 0.1903, 0.0943, 0.0433, 0.2193, 0.1334, 0.1169, 0.1616],
+       device='cuda:3'), in_proj_covar=tensor([0.0259, 0.0333, 0.0260, 0.0218, 0.0255, 0.0221, 0.0228, 0.0224],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 21:47:46,153 INFO [train.py:892] (3/4) Epoch 47, batch 400, loss[loss=0.1537, simple_loss=0.2293, pruned_loss=0.03907, over 19566.00 frames. ], tot_loss[loss=0.148, simple_loss=0.2302, pruned_loss=0.0329, over 3420300.23 frames. ], batch size: 47, lr: 3.32e-03, grad_scale: 8.0
+2023-03-29 21:47:56,618 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.491e+02 3.390e+02 3.969e+02 4.785e+02 1.194e+03, threshold=7.938e+02, percent-clipped=3.0
+2023-03-29 21:48:51,852 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9923, 3.0179, 2.0128, 3.5774, 3.2653, 3.4786, 3.5734, 2.9113],
+       device='cuda:3'), covar=tensor([0.0739, 0.0803, 0.1643, 0.0650, 0.0716, 0.0633, 0.0676, 0.0852],
+       device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0154, 0.0150, 0.0164, 0.0143, 0.0149, 0.0159, 0.0157],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004],
+       device='cuda:3')
+2023-03-29 21:49:02,614 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85763.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:49:20,574 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=85771.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:49:46,489 INFO [train.py:892] (3/4) Epoch 47, batch 450, loss[loss=0.1345, simple_loss=0.217, pruned_loss=0.02603, over 19837.00 frames. ], tot_loss[loss=0.1496, simple_loss=0.2316, pruned_loss=0.03378, over 3537628.89 frames. ], batch size: 239, lr: 3.32e-03, grad_scale: 8.0
+2023-03-29 21:49:53,681 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85784.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:50:56,273 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85811.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:51:16,885 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=85819.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:51:43,867 INFO [train.py:892] (3/4) Epoch 47, batch 500, loss[loss=0.1199, simple_loss=0.1967, pruned_loss=0.02159, over 19835.00 frames. ], tot_loss[loss=0.1484, simple_loss=0.2306, pruned_loss=0.03312, over 3626721.69 frames. ], batch size: 90, lr: 3.32e-03, grad_scale: 8.0
+2023-03-29 21:51:52,538 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.359e+02 3.439e+02 4.112e+02 5.000e+02 9.207e+02, threshold=8.225e+02, percent-clipped=1.0
+2023-03-29 21:52:01,475 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85839.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:53:18,281 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6643, 4.9065, 4.9704, 4.8297, 4.6391, 4.9568, 4.4593, 4.4677],
+       device='cuda:3'), covar=tensor([0.0515, 0.0527, 0.0499, 0.0437, 0.0655, 0.0486, 0.0702, 0.0956],
+       device='cuda:3'), in_proj_covar=tensor([0.0298, 0.0318, 0.0326, 0.0285, 0.0297, 0.0277, 0.0289, 0.0337],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 21:53:23,931 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:53:36,925 INFO [train.py:892] (3/4) Epoch 47, batch 550, loss[loss=0.1382, simple_loss=0.2159, pruned_loss=0.03026, over 19780.00 frames. ], tot_loss[loss=0.1487, simple_loss=0.2308, pruned_loss=0.03327, over 3698509.55 frames. ], batch size: 191, lr: 3.32e-03, grad_scale: 8.0
+2023-03-29 21:53:50,282 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:55:33,600 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85930.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 21:55:34,541 INFO [train.py:892] (3/4) Epoch 47, batch 600, loss[loss=0.1395, simple_loss=0.2066, pruned_loss=0.03622, over 19834.00 frames. ], tot_loss[loss=0.1486, simple_loss=0.2303, pruned_loss=0.03345, over 3754755.07 frames. ], batch size: 128, lr: 3.32e-03, grad_scale: 16.0
+2023-03-29 21:55:42,776 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.322e+02 3.636e+02 4.237e+02 5.117e+02 1.424e+03, threshold=8.474e+02, percent-clipped=3.0
+2023-03-29 21:55:45,997 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:56:37,255 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=85957.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:56:49,429 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.2418, 3.1501, 4.9108, 3.5918, 3.8482, 3.5833, 2.6412, 2.8573],
+       device='cuda:3'), covar=tensor([0.0929, 0.3007, 0.0392, 0.1089, 0.1827, 0.1586, 0.2717, 0.2507],
+       device='cuda:3'), in_proj_covar=tensor([0.0360, 0.0408, 0.0358, 0.0301, 0.0382, 0.0406, 0.0393, 0.0370],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 21:57:01,946 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.5251, 5.9728, 6.0589, 5.7749, 5.6748, 5.8048, 5.7166, 5.5368],
+       device='cuda:3'), covar=tensor([0.1437, 0.1134, 0.0753, 0.1351, 0.0628, 0.0659, 0.1772, 0.1960],
+       device='cuda:3'), in_proj_covar=tensor([0.0309, 0.0359, 0.0389, 0.0320, 0.0296, 0.0301, 0.0380, 0.0412],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 21:57:31,387 INFO [train.py:892] (3/4) Epoch 47, batch 650, loss[loss=0.142, simple_loss=0.2288, pruned_loss=0.02761, over 19846.00 frames. ], tot_loss[loss=0.1479, simple_loss=0.2297, pruned_loss=0.03301, over 3796324.53 frames. ], batch size: 190, lr: 3.32e-03, grad_scale: 16.0
+2023-03-29 21:57:32,310 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=85981.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 21:57:58,543 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=85991.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 21:58:07,141 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7958, 4.5668, 5.1868, 4.6479, 4.2385, 4.8995, 4.7979, 5.2854],
+       device='cuda:3'), covar=tensor([0.0890, 0.0434, 0.0351, 0.0426, 0.0838, 0.0599, 0.0451, 0.0340],
+       device='cuda:3'), in_proj_covar=tensor([0.0294, 0.0234, 0.0237, 0.0249, 0.0217, 0.0264, 0.0249, 0.0235],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 21:58:30,614 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86002.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:58:47,956 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0
+2023-03-29 21:59:06,142 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86018.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 21:59:15,052 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4896, 2.5564, 2.6735, 2.5752, 2.6482, 2.7082, 2.6300, 2.7154],
+       device='cuda:3'), covar=tensor([0.0451, 0.0438, 0.0415, 0.0386, 0.0509, 0.0354, 0.0437, 0.0393],
+       device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0093, 0.0096, 0.0091, 0.0103, 0.0096, 0.0112, 0.0084],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 21:59:35,618 INFO [train.py:892] (3/4) Epoch 47, batch 700, loss[loss=0.1469, simple_loss=0.2364, pruned_loss=0.02866, over 19806.00 frames. ], tot_loss[loss=0.1475, simple_loss=0.2298, pruned_loss=0.03257, over 3830583.63 frames. ], batch size: 114, lr: 3.32e-03, grad_scale: 16.0
+2023-03-29 21:59:44,265 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.350e+02 3.531e+02 4.077e+02 5.107e+02 6.974e+02, threshold=8.153e+02, percent-clipped=0.0
+2023-03-29 22:00:24,910 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86052.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:00:53,749 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86063.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:01:00,243 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.9597, 2.8748, 3.1066, 2.4667, 3.1605, 2.6260, 3.0154, 3.0785],
+       device='cuda:3'), covar=tensor([0.0593, 0.0626, 0.0510, 0.0871, 0.0455, 0.0581, 0.0537, 0.0445],
+       device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0097, 0.0093, 0.0117, 0.0089, 0.0092, 0.0089, 0.0084],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 22:01:07,337 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86069.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:01:36,238 INFO [train.py:892] (3/4) Epoch 47, batch 750, loss[loss=0.1362, simple_loss=0.2172, pruned_loss=0.02758, over 19866.00 frames. ], tot_loss[loss=0.1476, simple_loss=0.23, pruned_loss=0.03258, over 3857153.76 frames. ], batch size: 104, lr: 3.32e-03, grad_scale: 16.0
+2023-03-29 22:01:43,102 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86084.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:01:49,022 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.3606, 4.1770, 4.1996, 3.9418, 4.3917, 3.1155, 3.6880, 2.0561],
+       device='cuda:3'), covar=tensor([0.0201, 0.0236, 0.0156, 0.0209, 0.0148, 0.1014, 0.0670, 0.1579],
+       device='cuda:3'), in_proj_covar=tensor([0.0111, 0.0157, 0.0120, 0.0143, 0.0126, 0.0141, 0.0148, 0.0134],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003],
+       device='cuda:3')
+2023-03-29 22:02:29,672 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5203, 2.6808, 4.1890, 3.0894, 3.3536, 3.0931, 2.3724, 2.4729],
+       device='cuda:3'), covar=tensor([0.1285, 0.3355, 0.0545, 0.1200, 0.1932, 0.1754, 0.2782, 0.3002],
+       device='cuda:3'), in_proj_covar=tensor([0.0363, 0.0411, 0.0361, 0.0303, 0.0385, 0.0409, 0.0396, 0.0372],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 22:02:47,920 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86113.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:03:24,242 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86130.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:03:25,167 INFO [train.py:892] (3/4) Epoch 47, batch 800, loss[loss=0.1406, simple_loss=0.2184, pruned_loss=0.03139, over 19764.00 frames. ], tot_loss[loss=0.1474, simple_loss=0.2296, pruned_loss=0.03255, over 3878507.94 frames. ], batch size: 188, lr: 3.32e-03, grad_scale: 16.0
+2023-03-29 22:03:28,114 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86132.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:03:35,194 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.984e+02 3.596e+02 4.313e+02 5.169e+02 1.005e+03, threshold=8.627e+02, percent-clipped=4.0
+2023-03-29 22:03:46,759 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86139.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:05:10,092 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3087, 3.4619, 3.4906, 3.5563, 3.2726, 3.4496, 3.2450, 3.3535],
+       device='cuda:3'), covar=tensor([0.0267, 0.0328, 0.0341, 0.0246, 0.0376, 0.0327, 0.0368, 0.0490],
+       device='cuda:3'), in_proj_covar=tensor([0.0100, 0.0094, 0.0097, 0.0091, 0.0104, 0.0097, 0.0112, 0.0085],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 22:05:24,293 INFO [train.py:892] (3/4) Epoch 47, batch 850, loss[loss=0.1424, simple_loss=0.2229, pruned_loss=0.03096, over 19735.00 frames. ], tot_loss[loss=0.1479, simple_loss=0.2306, pruned_loss=0.03262, over 3894971.41 frames. ], batch size: 95, lr: 3.31e-03, grad_scale: 16.0
+2023-03-29 22:05:37,661 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86186.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:05:40,999 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86187.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:07:21,650 INFO [train.py:892] (3/4) Epoch 47, batch 900, loss[loss=0.2538, simple_loss=0.3283, pruned_loss=0.08971, over 19208.00 frames. ], tot_loss[loss=0.1486, simple_loss=0.231, pruned_loss=0.03308, over 3906320.11 frames. ], batch size: 452, lr: 3.31e-03, grad_scale: 16.0
+2023-03-29 22:07:22,522 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86231.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:07:29,403 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:07:30,573 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.206e+02 3.419e+02 4.069e+02 4.713e+02 1.195e+03, threshold=8.139e+02, percent-clipped=3.0
+2023-03-29 22:07:56,405 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86246.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:09:15,484 INFO [train.py:892] (3/4) Epoch 47, batch 950, loss[loss=0.1638, simple_loss=0.2401, pruned_loss=0.04381, over 19478.00 frames. ], tot_loss[loss=0.1482, simple_loss=0.2309, pruned_loss=0.0328, over 3915919.08 frames. ], batch size: 43, lr: 3.31e-03, grad_scale: 16.0
+2023-03-29 22:09:16,603 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86281.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:09:16,934 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6195, 2.5645, 4.4039, 3.1492, 3.3232, 3.0117, 2.3301, 2.4559],
+       device='cuda:3'), covar=tensor([0.1417, 0.4214, 0.0504, 0.1328, 0.2446, 0.2147, 0.3192, 0.3098],
+       device='cuda:3'), in_proj_covar=tensor([0.0361, 0.0410, 0.0360, 0.0301, 0.0383, 0.0407, 0.0395, 0.0371],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 22:09:24,391 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.2590, 5.0551, 5.0369, 5.3173, 5.0771, 5.5513, 5.3780, 5.6092],
+       device='cuda:3'), covar=tensor([0.0705, 0.0442, 0.0412, 0.0395, 0.0644, 0.0412, 0.0442, 0.0374],
+       device='cuda:3'), in_proj_covar=tensor([0.0166, 0.0192, 0.0213, 0.0192, 0.0189, 0.0172, 0.0166, 0.0213],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 22:09:27,185 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0
+2023-03-29 22:09:29,123 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86286.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 22:10:16,842 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86307.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:10:18,951 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.6194, 2.6877, 1.8856, 2.9383, 2.7314, 2.8317, 2.9589, 2.4461],
+       device='cuda:3'), covar=tensor([0.0759, 0.0812, 0.1444, 0.0756, 0.0770, 0.0665, 0.0679, 0.1000],
+       device='cuda:3'), in_proj_covar=tensor([0.0152, 0.0152, 0.0148, 0.0162, 0.0141, 0.0148, 0.0157, 0.0155],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0003],
+       device='cuda:3')
+2023-03-29 22:10:29,666 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86313.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:10:44,079 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3695, 2.4929, 4.0825, 2.9552, 3.1032, 2.9235, 2.2493, 2.4097],
+       device='cuda:3'), covar=tensor([0.1540, 0.3926, 0.0572, 0.1493, 0.2701, 0.2147, 0.3159, 0.3219],
+       device='cuda:3'), in_proj_covar=tensor([0.0362, 0.0411, 0.0360, 0.0302, 0.0384, 0.0408, 0.0396, 0.0372],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 22:11:06,213 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0
+2023-03-29 22:11:07,539 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86329.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:11:10,606 INFO [train.py:892] (3/4) Epoch 47, batch 1000, loss[loss=0.1405, simple_loss=0.2168, pruned_loss=0.03211, over 19638.00 frames. ], tot_loss[loss=0.1478, simple_loss=0.2304, pruned_loss=0.03261, over 3924435.81 frames. ], batch size: 79, lr: 3.31e-03, grad_scale: 16.0
+2023-03-29 22:11:20,551 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.209e+02 3.279e+02 4.003e+02 4.723e+02 7.348e+02, threshold=8.007e+02, percent-clipped=0.0
+2023-03-29 22:12:16,899 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86358.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:13:07,774 INFO [train.py:892] (3/4) Epoch 47, batch 1050, loss[loss=0.1398, simple_loss=0.2112, pruned_loss=0.03422, over 19788.00 frames. ], tot_loss[loss=0.1487, simple_loss=0.2314, pruned_loss=0.03303, over 3929981.98 frames. ], batch size: 65, lr: 3.31e-03, grad_scale: 16.0
+2023-03-29 22:14:11,845 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86408.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:14:51,544 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86425.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:14:58,763 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.6673, 2.7230, 2.9403, 2.5985, 3.0713, 3.0569, 3.5277, 3.8463],
+       device='cuda:3'), covar=tensor([0.0655, 0.1662, 0.1665, 0.2262, 0.1641, 0.1488, 0.0736, 0.0591],
+       device='cuda:3'), in_proj_covar=tensor([0.0267, 0.0250, 0.0280, 0.0267, 0.0314, 0.0269, 0.0244, 0.0277],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 22:15:05,586 INFO [train.py:892] (3/4) Epoch 47, batch 1100, loss[loss=0.1476, simple_loss=0.2182, pruned_loss=0.03852, over 19814.00 frames. ], tot_loss[loss=0.149, simple_loss=0.2315, pruned_loss=0.03329, over 3933837.66 frames. ], batch size: 147, lr: 3.31e-03, grad_scale: 16.0
+2023-03-29 22:15:16,016 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.572e+02 3.379e+02 4.084e+02 4.897e+02 1.462e+03, threshold=8.168e+02, percent-clipped=5.0
+2023-03-29 22:15:27,867 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86440.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:15:40,310 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86445.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:17:00,644 INFO [train.py:892] (3/4) Epoch 47, batch 1150, loss[loss=0.1343, simple_loss=0.2122, pruned_loss=0.02823, over 19953.00 frames. ], tot_loss[loss=0.1486, simple_loss=0.231, pruned_loss=0.03308, over 3937888.50 frames. ], batch size: 53, lr: 3.31e-03, grad_scale: 16.0
+2023-03-29 22:17:19,487 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.0957, 3.0828, 4.7712, 3.5611, 3.7155, 3.4557, 2.5917, 2.7623],
+       device='cuda:3'), covar=tensor([0.0992, 0.3006, 0.0375, 0.1082, 0.1854, 0.1675, 0.2752, 0.2601],
+       device='cuda:3'), in_proj_covar=tensor([0.0360, 0.0408, 0.0357, 0.0300, 0.0382, 0.0405, 0.0393, 0.0370],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 22:17:46,973 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86501.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:17:59,870 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86506.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 22:18:58,205 INFO [train.py:892] (3/4) Epoch 47, batch 1200, loss[loss=0.154, simple_loss=0.2364, pruned_loss=0.03585, over 19750.00 frames. ], tot_loss[loss=0.1479, simple_loss=0.23, pruned_loss=0.03294, over 3940164.71 frames. ], batch size: 250, lr: 3.31e-03, grad_scale: 16.0
+2023-03-29 22:18:59,395 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86531.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:19:07,258 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.118e+02 3.301e+02 3.835e+02 4.475e+02 8.063e+02, threshold=7.670e+02, percent-clipped=0.0
+2023-03-29 22:19:40,162 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0
+2023-03-29 22:20:44,963 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86579.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:20:48,873 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86580.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:20:49,861 INFO [train.py:892] (3/4) Epoch 47, batch 1250, loss[loss=0.161, simple_loss=0.2441, pruned_loss=0.03892, over 19569.00 frames. ], tot_loss[loss=0.1475, simple_loss=0.2295, pruned_loss=0.0327, over 3942631.96 frames. ], batch size: 60, lr: 3.31e-03, grad_scale: 16.0
+2023-03-29 22:21:04,076 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86586.0, num_to_drop=1, layers_to_drop={2}
+2023-03-29 22:21:41,983 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86602.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:22:06,576 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86613.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:22:49,113 INFO [train.py:892] (3/4) Epoch 47, batch 1300, loss[loss=0.1362, simple_loss=0.2256, pruned_loss=0.02343, over 19927.00 frames. ], tot_loss[loss=0.1471, simple_loss=0.2292, pruned_loss=0.03246, over 3944693.86 frames. ], batch size: 51, lr: 3.31e-03, grad_scale: 16.0
+2023-03-29 22:22:57,972 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86634.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 22:22:59,158 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.242e+02 3.251e+02 3.887e+02 4.565e+02 9.819e+02, threshold=7.775e+02, percent-clipped=3.0
+2023-03-29 22:23:12,437 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86641.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:23:21,953 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.4108, 2.7014, 2.4139, 1.8937, 2.4457, 2.6719, 2.6426, 2.6398],
+       device='cuda:3'), covar=tensor([0.0477, 0.0354, 0.0399, 0.0656, 0.0476, 0.0353, 0.0354, 0.0330],
+       device='cuda:3'), in_proj_covar=tensor([0.0119, 0.0112, 0.0112, 0.0112, 0.0116, 0.0102, 0.0104, 0.0102],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 22:23:53,864 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86658.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:24:00,156 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86661.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:24:43,575 INFO [train.py:892] (3/4) Epoch 47, batch 1350, loss[loss=0.1388, simple_loss=0.2309, pruned_loss=0.02336, over 19875.00 frames. ], tot_loss[loss=0.1464, simple_loss=0.2285, pruned_loss=0.03211, over 3946708.67 frames. ], batch size: 53, lr: 3.30e-03, grad_scale: 16.0
+2023-03-29 22:25:44,662 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86706.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:25:48,901 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86708.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:26:27,225 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86725.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:26:40,943 INFO [train.py:892] (3/4) Epoch 47, batch 1400, loss[loss=0.1413, simple_loss=0.2326, pruned_loss=0.02498, over 19795.00 frames. ], tot_loss[loss=0.147, simple_loss=0.2292, pruned_loss=0.03246, over 3946781.97 frames. ], batch size: 83, lr: 3.30e-03, grad_scale: 16.0
+2023-03-29 22:26:49,082 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.330e+02 3.187e+02 3.920e+02 5.050e+02 7.981e+02, threshold=7.841e+02, percent-clipped=2.0
+2023-03-29 22:27:36,116 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86756.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:28:14,311 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86773.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:28:29,053 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0
+2023-03-29 22:28:33,036 INFO [train.py:892] (3/4) Epoch 47, batch 1450, loss[loss=0.1315, simple_loss=0.2128, pruned_loss=0.02513, over 19728.00 frames. ], tot_loss[loss=0.1465, simple_loss=0.2287, pruned_loss=0.03219, over 3948554.03 frames. ], batch size: 104, lr: 3.30e-03, grad_scale: 16.0
+2023-03-29 22:29:06,648 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86796.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:29:21,079 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86801.0, num_to_drop=1, layers_to_drop={2}
+2023-03-29 22:29:49,765 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86814.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:30:28,356 INFO [train.py:892] (3/4) Epoch 47, batch 1500, loss[loss=0.1418, simple_loss=0.2214, pruned_loss=0.03115, over 19877.00 frames. ], tot_loss[loss=0.1469, simple_loss=0.2289, pruned_loss=0.03242, over 3948988.85 frames. ], batch size: 134, lr: 3.30e-03, grad_scale: 16.0
+2023-03-29 22:30:37,005 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.591e+02 3.342e+02 4.163e+02 4.864e+02 6.569e+02, threshold=8.326e+02, percent-clipped=0.0
+2023-03-29 22:31:04,879 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.6295, 4.4534, 4.9634, 4.4938, 4.1363, 4.7199, 4.6262, 5.0429],
+       device='cuda:3'), covar=tensor([0.0801, 0.0388, 0.0353, 0.0389, 0.0919, 0.0562, 0.0510, 0.0358],
+       device='cuda:3'), in_proj_covar=tensor([0.0292, 0.0234, 0.0236, 0.0246, 0.0215, 0.0263, 0.0248, 0.0235],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 22:31:22,986 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.4606, 5.9035, 5.9723, 5.7205, 5.6714, 5.7269, 5.6181, 5.4558],
+       device='cuda:3'), covar=tensor([0.1443, 0.1262, 0.0843, 0.1147, 0.0713, 0.0634, 0.1807, 0.1946],
+       device='cuda:3'), in_proj_covar=tensor([0.0310, 0.0360, 0.0390, 0.0322, 0.0297, 0.0304, 0.0382, 0.0412],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004],
+       device='cuda:3')
+2023-03-29 22:32:11,883 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86875.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:32:18,229 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=86878.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:32:23,356 INFO [train.py:892] (3/4) Epoch 47, batch 1550, loss[loss=0.1427, simple_loss=0.2271, pruned_loss=0.02912, over 19796.00 frames. ], tot_loss[loss=0.1459, simple_loss=0.228, pruned_loss=0.03197, over 3950467.63 frames. ], batch size: 86, lr: 3.30e-03, grad_scale: 16.0
+2023-03-29 22:33:11,393 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=86902.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:33:57,177 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.08 vs. limit=2.0
+2023-03-29 22:34:19,995 INFO [train.py:892] (3/4) Epoch 47, batch 1600, loss[loss=0.1865, simple_loss=0.2678, pruned_loss=0.05262, over 19643.00 frames. ], tot_loss[loss=0.1469, simple_loss=0.2289, pruned_loss=0.03243, over 3950355.49 frames. ], batch size: 57, lr: 3.30e-03, grad_scale: 16.0
+2023-03-29 22:34:27,993 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.263e+02 3.386e+02 4.039e+02 4.750e+02 1.112e+03, threshold=8.078e+02, percent-clipped=1.0
+2023-03-29 22:34:30,941 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=86936.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:34:37,220 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=86939.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:35:03,422 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=86950.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:36:13,837 INFO [train.py:892] (3/4) Epoch 47, batch 1650, loss[loss=0.1441, simple_loss=0.236, pruned_loss=0.02605, over 19855.00 frames. ], tot_loss[loss=0.1464, simple_loss=0.2287, pruned_loss=0.03203, over 3949932.08 frames. ], batch size: 56, lr: 3.30e-03, grad_scale: 16.0
+2023-03-29 22:37:07,487 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87002.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 22:38:05,844 INFO [train.py:892] (3/4) Epoch 47, batch 1700, loss[loss=0.156, simple_loss=0.2539, pruned_loss=0.0291, over 19851.00 frames. ], tot_loss[loss=0.1467, simple_loss=0.2291, pruned_loss=0.03211, over 3948639.25 frames. ], batch size: 56, lr: 3.30e-03, grad_scale: 16.0
+2023-03-29 22:38:14,348 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 1.986e+02 3.322e+02 4.069e+02 4.865e+02 7.645e+02, threshold=8.138e+02, percent-clipped=0.0
+2023-03-29 22:38:47,224 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0
+2023-03-29 22:39:20,702 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87063.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 22:39:56,707 INFO [train.py:892] (3/4) Epoch 47, batch 1750, loss[loss=0.1328, simple_loss=0.2159, pruned_loss=0.02481, over 19777.00 frames. ], tot_loss[loss=0.1468, simple_loss=0.2291, pruned_loss=0.03224, over 3949730.34 frames. ], batch size: 163, lr: 3.30e-03, grad_scale: 16.0
+2023-03-29 22:40:26,998 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87096.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:40:36,203 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87101.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 22:41:33,782 INFO [train.py:892] (3/4) Epoch 47, batch 1800, loss[loss=0.1515, simple_loss=0.2325, pruned_loss=0.0353, over 19738.00 frames. ], tot_loss[loss=0.1462, simple_loss=0.2285, pruned_loss=0.03198, over 3950658.20 frames. ], batch size: 92, lr: 3.30e-03, grad_scale: 16.0
+2023-03-29 22:41:41,061 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.449e+02 3.587e+02 4.161e+02 4.922e+02 1.323e+03, threshold=8.323e+02, percent-clipped=3.0
+2023-03-29 22:41:59,487 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87144.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:42:08,354 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87149.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:42:12,243 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.7653, 3.5464, 3.8628, 3.0621, 3.9847, 3.3455, 3.6205, 3.9369],
+       device='cuda:3'), covar=tensor([0.0583, 0.0437, 0.0518, 0.0753, 0.0354, 0.0412, 0.0479, 0.0406],
+       device='cuda:3'), in_proj_covar=tensor([0.0087, 0.0097, 0.0093, 0.0117, 0.0089, 0.0092, 0.0088, 0.0084],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 22:42:36,087 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.1443, 3.0683, 4.9664, 3.5362, 3.6579, 3.5019, 2.5489, 2.7960],
+       device='cuda:3'), covar=tensor([0.1024, 0.3106, 0.0365, 0.1170, 0.2043, 0.1613, 0.2689, 0.2603],
+       device='cuda:3'), in_proj_covar=tensor([0.0362, 0.0408, 0.0358, 0.0302, 0.0382, 0.0405, 0.0393, 0.0371],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 22:42:47,378 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87170.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:43:06,782 INFO [train.py:892] (3/4) Epoch 47, batch 1850, loss[loss=0.1833, simple_loss=0.2721, pruned_loss=0.04726, over 19814.00 frames. ], tot_loss[loss=0.148, simple_loss=0.2317, pruned_loss=0.03215, over 3947352.60 frames. ], batch size: 57, lr: 3.30e-03, grad_scale: 16.0
+2023-03-29 22:44:09,146 INFO [train.py:892] (3/4) Epoch 48, batch 0, loss[loss=0.1462, simple_loss=0.2299, pruned_loss=0.03123, over 19679.00 frames. ], tot_loss[loss=0.1462, simple_loss=0.2299, pruned_loss=0.03123, over 19679.00 frames. ], batch size: 55, lr: 3.26e-03, grad_scale: 16.0
+2023-03-29 22:44:09,147 INFO [train.py:917] (3/4) Computing validation loss
+2023-03-29 22:44:44,075 INFO [train.py:926] (3/4) Epoch 48, validation: loss=0.1901, simple_loss=0.2508, pruned_loss=0.06469, over 2883724.00 frames. 
+2023-03-29 22:44:44,076 INFO [train.py:927] (3/4) Maximum memory allocated so far is 22394MB
+2023-03-29 22:46:30,413 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87230.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:46:40,240 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87234.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:46:41,401 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.137e+02 3.310e+02 3.743e+02 4.381e+02 8.999e+02, threshold=7.487e+02, percent-clipped=2.0
+2023-03-29 22:46:45,787 INFO [train.py:892] (3/4) Epoch 48, batch 50, loss[loss=0.14, simple_loss=0.214, pruned_loss=0.03306, over 19866.00 frames. ], tot_loss[loss=0.1445, simple_loss=0.226, pruned_loss=0.03149, over 890526.72 frames. ], batch size: 129, lr: 3.26e-03, grad_scale: 16.0
+2023-03-29 22:46:47,095 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87236.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:48:06,449 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.1696, 2.7537, 3.2762, 3.3270, 3.9267, 4.4288, 4.2587, 4.2825],
+       device='cuda:3'), covar=tensor([0.1004, 0.1618, 0.1252, 0.0765, 0.0486, 0.0321, 0.0389, 0.0466],
+       device='cuda:3'), in_proj_covar=tensor([0.0169, 0.0173, 0.0184, 0.0161, 0.0147, 0.0142, 0.0137, 0.0126],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 22:48:06,531 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.7629, 4.0869, 3.9723, 4.9141, 3.2664, 3.6213, 3.2988, 3.0071],
+       device='cuda:3'), covar=tensor([0.0472, 0.1787, 0.0957, 0.0356, 0.1907, 0.1000, 0.1157, 0.1509],
+       device='cuda:3'), in_proj_covar=tensor([0.0255, 0.0331, 0.0258, 0.0217, 0.0253, 0.0218, 0.0227, 0.0223],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 22:48:38,932 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87284.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:48:42,201 INFO [train.py:892] (3/4) Epoch 48, batch 100, loss[loss=0.1343, simple_loss=0.215, pruned_loss=0.02679, over 19696.00 frames. ], tot_loss[loss=0.1431, simple_loss=0.2246, pruned_loss=0.03081, over 1569627.44 frames. ], batch size: 85, lr: 3.26e-03, grad_scale: 16.0
+2023-03-29 22:48:53,991 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87291.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:50:36,653 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.455e+02 3.250e+02 3.994e+02 4.855e+02 8.208e+02, threshold=7.988e+02, percent-clipped=2.0
+2023-03-29 22:50:39,172 INFO [train.py:892] (3/4) Epoch 48, batch 150, loss[loss=0.1391, simple_loss=0.2199, pruned_loss=0.02912, over 19821.00 frames. ], tot_loss[loss=0.1434, simple_loss=0.2252, pruned_loss=0.03086, over 2097896.53 frames. ], batch size: 147, lr: 3.26e-03, grad_scale: 16.0
+2023-03-29 22:50:44,607 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.5339, 2.7299, 3.9614, 3.1605, 3.1829, 3.0826, 2.3135, 2.5419],
+       device='cuda:3'), covar=tensor([0.1311, 0.3498, 0.0630, 0.1277, 0.2126, 0.1833, 0.2949, 0.2993],
+       device='cuda:3'), in_proj_covar=tensor([0.0364, 0.0409, 0.0358, 0.0303, 0.0383, 0.0407, 0.0395, 0.0373],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 22:51:30,891 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.8242, 4.4988, 4.4946, 4.2444, 4.8121, 3.1645, 3.9379, 2.3889],
+       device='cuda:3'), covar=tensor([0.0163, 0.0232, 0.0164, 0.0215, 0.0151, 0.0986, 0.0834, 0.1492],
+       device='cuda:3'), in_proj_covar=tensor([0.0111, 0.0156, 0.0120, 0.0143, 0.0126, 0.0140, 0.0148, 0.0134],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003],
+       device='cuda:3')
+2023-03-29 22:51:32,862 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87358.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 22:51:41,904 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87362.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:52:34,018 INFO [train.py:892] (3/4) Epoch 48, batch 200, loss[loss=0.1186, simple_loss=0.1971, pruned_loss=0.01998, over 19732.00 frames. ], tot_loss[loss=0.1444, simple_loss=0.2269, pruned_loss=0.03097, over 2507531.97 frames. ], batch size: 118, lr: 3.26e-03, grad_scale: 16.0
+2023-03-29 22:52:38,438 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.8723, 2.5467, 4.1915, 3.6607, 4.0531, 4.1548, 3.9484, 3.8670],
+       device='cuda:3'), covar=tensor([0.0675, 0.1005, 0.0116, 0.0582, 0.0168, 0.0239, 0.0183, 0.0211],
+       device='cuda:3'), in_proj_covar=tensor([0.0105, 0.0108, 0.0093, 0.0154, 0.0092, 0.0106, 0.0094, 0.0092],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 22:54:01,008 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87423.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:54:01,428 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.69 vs. limit=5.0
+2023-03-29 22:54:24,402 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.319e+02 3.416e+02 4.017e+02 4.841e+02 7.070e+02, threshold=8.034e+02, percent-clipped=0.0
+2023-03-29 22:54:26,341 INFO [train.py:892] (3/4) Epoch 48, batch 250, loss[loss=0.1564, simple_loss=0.2437, pruned_loss=0.03454, over 19750.00 frames. ], tot_loss[loss=0.1438, simple_loss=0.2257, pruned_loss=0.03098, over 2827951.53 frames. ], batch size: 250, lr: 3.26e-03, grad_scale: 16.0
+2023-03-29 22:55:23,605 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87461.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:55:45,401 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87470.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:56:15,049 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0
+2023-03-29 22:56:19,966 INFO [train.py:892] (3/4) Epoch 48, batch 300, loss[loss=0.1333, simple_loss=0.2157, pruned_loss=0.02545, over 19807.00 frames. ], tot_loss[loss=0.1439, simple_loss=0.2257, pruned_loss=0.03101, over 3077781.07 frames. ], batch size: 117, lr: 3.25e-03, grad_scale: 16.0
+2023-03-29 22:56:41,414 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.9463, 4.6812, 4.7345, 4.9937, 4.7767, 5.1403, 5.0952, 5.2724],
+       device='cuda:3'), covar=tensor([0.0692, 0.0428, 0.0459, 0.0352, 0.0627, 0.0448, 0.0421, 0.0290],
+       device='cuda:3'), in_proj_covar=tensor([0.0165, 0.0190, 0.0211, 0.0189, 0.0187, 0.0171, 0.0163, 0.0209],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003],
+       device='cuda:3')
+2023-03-29 22:56:59,115 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.08 vs. limit=2.0
+2023-03-29 22:57:07,914 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87505.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:57:20,132 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.3658, 2.4467, 1.6611, 2.6432, 2.4510, 2.4827, 2.6045, 2.1839],
+       device='cuda:3'), covar=tensor([0.0825, 0.0847, 0.1490, 0.0735, 0.0790, 0.0754, 0.0744, 0.1062],
+       device='cuda:3'), in_proj_covar=tensor([0.0153, 0.0154, 0.0149, 0.0164, 0.0143, 0.0149, 0.0159, 0.0156],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004],
+       device='cuda:3')
+2023-03-29 22:57:37,237 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87518.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:57:45,610 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87522.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:58:14,252 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87534.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 22:58:15,402 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.562e+02 3.502e+02 4.019e+02 4.844e+02 7.845e+02, threshold=8.037e+02, percent-clipped=0.0
+2023-03-29 22:58:18,688 INFO [train.py:892] (3/4) Epoch 48, batch 350, loss[loss=0.1796, simple_loss=0.2617, pruned_loss=0.04869, over 19704.00 frames. ], tot_loss[loss=0.1463, simple_loss=0.2284, pruned_loss=0.0321, over 3270299.84 frames. ], batch size: 325, lr: 3.25e-03, grad_scale: 16.0
+2023-03-29 22:59:31,530 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87566.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 23:00:10,465 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87582.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 23:00:14,876 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87584.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 23:00:17,788 INFO [train.py:892] (3/4) Epoch 48, batch 400, loss[loss=0.1526, simple_loss=0.2386, pruned_loss=0.03326, over 19833.00 frames. ], tot_loss[loss=0.1467, simple_loss=0.2294, pruned_loss=0.03198, over 3419809.62 frames. ], batch size: 58, lr: 3.25e-03, grad_scale: 16.0
+2023-03-29 23:00:18,753 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87586.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 23:00:25,606 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87589.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 23:00:41,008 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0
+2023-03-29 23:00:57,008 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=192, metric=1.45 vs. limit=2.0
+2023-03-29 23:01:05,051 INFO [zipformer.py:625] (3/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=87605.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 23:02:10,863 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.523e+02 3.389e+02 3.932e+02 4.658e+02 8.787e+02, threshold=7.864e+02, percent-clipped=1.0
+2023-03-29 23:02:13,080 INFO [train.py:892] (3/4) Epoch 48, batch 450, loss[loss=0.1353, simple_loss=0.2169, pruned_loss=0.02681, over 19773.00 frames. ], tot_loss[loss=0.146, simple_loss=0.2286, pruned_loss=0.03165, over 3537313.16 frames. ], batch size: 108, lr: 3.25e-03, grad_scale: 16.0
+2023-03-29 23:02:37,089 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87645.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 23:02:48,809 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87650.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 23:03:07,341 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87658.0, num_to_drop=1, layers_to_drop={2}
+2023-03-29 23:03:24,737 INFO [zipformer.py:625] (3/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=87666.0, num_to_drop=1, layers_to_drop={2}
+2023-03-29 23:04:11,533 INFO [train.py:892] (3/4) Epoch 48, batch 500, loss[loss=0.1282, simple_loss=0.2039, pruned_loss=0.02623, over 19817.00 frames. ], tot_loss[loss=0.1465, simple_loss=0.2288, pruned_loss=0.03212, over 3629197.32 frames. ], batch size: 133, lr: 3.25e-03, grad_scale: 16.0
+2023-03-29 23:04:58,732 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87706.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 23:05:28,382 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87718.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 23:06:06,493 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.147e+02 3.363e+02 3.924e+02 4.897e+02 8.586e+02, threshold=7.848e+02, percent-clipped=2.0
+2023-03-29 23:06:09,076 INFO [train.py:892] (3/4) Epoch 48, batch 550, loss[loss=0.1531, simple_loss=0.2467, pruned_loss=0.02969, over 19678.00 frames. ], tot_loss[loss=0.1466, simple_loss=0.229, pruned_loss=0.03211, over 3700496.38 frames. ], batch size: 49, lr: 3.25e-03, grad_scale: 16.0
+2023-03-29 23:07:06,870 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([4.4535, 3.3407, 5.1648, 3.8327, 3.9272, 3.8061, 2.7810, 3.0533],
+       device='cuda:3'), covar=tensor([0.1013, 0.3265, 0.0370, 0.1161, 0.1953, 0.1609, 0.2880, 0.2654],
+       device='cuda:3'), in_proj_covar=tensor([0.0364, 0.0410, 0.0359, 0.0304, 0.0384, 0.0409, 0.0395, 0.0373],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 23:08:07,521 INFO [train.py:892] (3/4) Epoch 48, batch 600, loss[loss=0.1309, simple_loss=0.2104, pruned_loss=0.02564, over 19745.00 frames. ], tot_loss[loss=0.1473, simple_loss=0.2296, pruned_loss=0.03247, over 3755050.85 frames. ], batch size: 106, lr: 3.25e-03, grad_scale: 16.0
+2023-03-29 23:08:15,513 INFO [scaling.py:679] (3/4) Whitening: num_groups=8, num_channels=96, metric=1.08 vs. limit=2.0
+2023-03-29 23:09:19,035 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87817.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 23:09:28,025 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.89 vs. limit=5.0
+2023-03-29 23:09:59,425 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.332e+02 3.329e+02 4.109e+02 5.019e+02 8.243e+02, threshold=8.218e+02, percent-clipped=1.0
+2023-03-29 23:10:01,437 INFO [train.py:892] (3/4) Epoch 48, batch 650, loss[loss=0.1653, simple_loss=0.2465, pruned_loss=0.04202, over 19711.00 frames. ], tot_loss[loss=0.1467, simple_loss=0.2291, pruned_loss=0.03214, over 3796101.00 frames. ], batch size: 305, lr: 3.25e-03, grad_scale: 16.0
+2023-03-29 23:10:07,564 INFO [scaling.py:679] (3/4) Whitening: num_groups=1, num_channels=384, metric=4.53 vs. limit=5.0
+2023-03-29 23:11:01,546 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87861.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 23:11:56,093 INFO [train.py:892] (3/4) Epoch 48, batch 700, loss[loss=0.1621, simple_loss=0.2383, pruned_loss=0.04292, over 19793.00 frames. ], tot_loss[loss=0.1466, simple_loss=0.229, pruned_loss=0.03215, over 3830408.09 frames. ], batch size: 45, lr: 3.25e-03, grad_scale: 16.0
+2023-03-29 23:11:57,169 INFO [zipformer.py:625] (3/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=87886.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 23:12:56,926 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([5.0876, 5.3640, 5.4038, 5.2775, 5.0832, 5.3775, 4.9047, 4.8652],
+       device='cuda:3'), covar=tensor([0.0466, 0.0488, 0.0470, 0.0449, 0.0602, 0.0463, 0.0705, 0.1004],
+       device='cuda:3'), in_proj_covar=tensor([0.0300, 0.0316, 0.0328, 0.0286, 0.0296, 0.0279, 0.0287, 0.0337],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')
+2023-03-29 23:13:47,288 INFO [zipformer.py:625] (3/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=87934.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 23:13:48,379 INFO [optim.py:368] (3/4) Clipping_scale=2.0, grad-norm quartiles 2.639e+02 3.556e+02 4.149e+02 4.938e+02 1.091e+03, threshold=8.298e+02, percent-clipped=2.0
+2023-03-29 23:13:50,818 INFO [train.py:892] (3/4) Epoch 48, batch 750, loss[loss=0.1421, simple_loss=0.2317, pruned_loss=0.02622, over 19665.00 frames. ], tot_loss[loss=0.147, simple_loss=0.2293, pruned_loss=0.03237, over 3857220.49 frames. ], batch size: 50, lr: 3.25e-03, grad_scale: 32.0
+2023-03-29 23:14:39,411 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87940.0, num_to_drop=1, layers_to_drop={1}
+2023-03-29 23:14:50,944 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87945.0, num_to_drop=0, layers_to_drop=set()
+2023-03-29 23:15:26,482 INFO [zipformer.py:625] (3/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=87961.0, num_to_drop=1, layers_to_drop={0}
+2023-03-29 23:17:04,764 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([2.5852, 1.9482, 2.3386, 2.7254, 3.1051, 3.2376, 3.0842, 3.1386],
+       device='cuda:3'), covar=tensor([0.1122, 0.1834, 0.1506, 0.0854, 0.0559, 0.0429, 0.0551, 0.0534],
+       device='cuda:3'), in_proj_covar=tensor([0.0170, 0.0174, 0.0186, 0.0162, 0.0148, 0.0143, 0.0138, 0.0126],
+       device='cuda:3'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002],
+       device='cuda:3')
+2023-03-29 23:17:10,257 INFO [train.py:892] (3/4) Epoch 48, batch 800, loss[loss=0.1358, simple_loss=0.2127, pruned_loss=0.02944, over 19812.00 frames. ], tot_loss[loss=0.1466, simple_loss=0.2288, pruned_loss=0.03222, over 3878188.96 frames. ], batch size: 132, lr: 3.25e-03, grad_scale: 32.0
+2023-03-29 23:17:40,225 INFO [zipformer.py:1454] (3/4) attn_weights_entropy = tensor([3.3390, 2.5883, 3.7232, 2.9558, 3.1429, 2.9702, 2.2353, 2.3813],
+       device='cuda:3'), covar=tensor([0.1366, 0.3240, 0.0674, 0.1308, 0.1935, 0.1733, 0.2838, 0.2845],
+       device='cuda:3'), in_proj_covar=tensor([0.0364, 0.0411, 0.0359, 0.0304, 0.0385, 0.0410, 0.0397, 0.0374],
+       device='cuda:3'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003],
+       device='cuda:3')