2023-04-16 11:31:55,682 INFO [train.py:963] (1/4) Training started 2023-04-16 11:31:55,682 INFO [train.py:973] (1/4) Device: cuda:1 2023-04-16 11:31:55,685 INFO [train.py:982] (1/4) {'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': '266eaf80fe659f113b4392b20dc1e35812f677cd', 'k2-git-date': 'Tue Feb 14 11:19:02 2023', 'lhotse-version': '1.14.0.dev+git.b55313a.clean', 'torch-version': '1.13.1', 'torch-cuda-available': True, 'torch-cuda-version': '11.6', 'python-version': '3.1', 'icefall-git-branch': 'master', 'icefall-git-sha1': 'ba3c923-dirty', 'icefall-git-date': 'Sun Apr 16 10:42:45 2023', 'icefall-path': '/k2-dev/yangyifan/icefall-master', 'k2-path': '/k2-dev/yangyifan/anaconda3/envs/icefall/lib/python3.10/site-packages/k2-1.23.4.dev20230214+cuda11.6.torch1.13.1-py3.10-linux-x86_64.egg/k2/__init__.py', 'lhotse-path': '/k2-dev/yangyifan/anaconda3/envs/icefall/lib/python3.10/site-packages/lhotse-1.14.0.dev0+git.b55313a.clean-py3.10.egg/lhotse/__init__.py', 'hostname': 'de-74279-k2-train-6-1219221738-65dd59bbf8-2ghmr', 'IP address': '10.177.28.85'}, 'world_size': 4, 'master_port': 12354, 'tensorboard': True, 'num_epochs': 30, 'start_epoch': 1, 'start_batch': 0, 'exp_dir': PosixPath('pruned_transducer_stateless7/exp'), 'bpe_model': 'data/en/lang_bpe_500/bpe.model', 'base_lr': 0.05, 'lr_batches': 5000, 'lr_epochs': 3.5, 'context_size': 2, 'prune_range': 5, 'lm_scale': 0.25, 'am_scale': 0.0, 'simple_loss_scale': 0.5, 'seed': 42, 'print_diagnostics': False, 'inf_check': False, 'save_every_n': 2000, 'keep_last_k': 30, 'average_period': 200, 'use_fp16': True, '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, 'language': 'en', 'cv_manifest_dir': PosixPath('data/en/fbank'), 'manifest_dir': PosixPath('data/fbank'), 'max_duration': 550, 'bucketing_sampler': True, 'num_buckets': 30, '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} 2023-04-16 11:31:55,685 INFO [train.py:984] (1/4) About to create model 2023-04-16 11:31:56,575 INFO [zipformer.py:178] (1/4) 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. 2023-04-16 11:31:56,599 INFO [train.py:988] (1/4) Number of model parameters: 70369391 2023-04-16 11:31:59,918 INFO [train.py:1003] (1/4) Using DDP 2023-04-16 11:32:00,366 INFO [asr_datamodule.py:404] (1/4) About to get train cuts 2023-04-16 11:32:00,368 INFO [asr_datamodule.py:230] (1/4) Enable MUSAN 2023-04-16 11:32:00,368 INFO [asr_datamodule.py:231] (1/4) About to get Musan cuts 2023-04-16 11:32:03,375 INFO [asr_datamodule.py:255] (1/4) Enable SpecAugment 2023-04-16 11:32:03,375 INFO [asr_datamodule.py:256] (1/4) Time warp factor: 80 2023-04-16 11:32:03,375 INFO [asr_datamodule.py:266] (1/4) Num frame mask: 10 2023-04-16 11:32:03,375 INFO [asr_datamodule.py:279] (1/4) About to create train dataset 2023-04-16 11:32:03,376 INFO [asr_datamodule.py:306] (1/4) Using DynamicBucketingSampler. 2023-04-16 11:32:07,575 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 11:32:09,138 INFO [asr_datamodule.py:321] (1/4) About to create train dataloader 2023-04-16 11:32:09,139 INFO [asr_datamodule.py:411] (1/4) About to get dev cuts 2023-04-16 11:32:09,141 INFO [asr_datamodule.py:352] (1/4) About to create dev dataset 2023-04-16 11:32:10,067 INFO [asr_datamodule.py:369] (1/4) About to create dev dataloader 2023-04-16 11:32:10,068 INFO [train.py:1199] (1/4) Sanity check -- see if any of the batches in epoch 1 would cause OOM. 2023-04-16 11:32:14,453 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 11:32:20,405 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 11:32:31,338 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-16 11:32:31,338 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-16 11:32:31,338 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-16 11:32:31,345 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-16 11:32:31,360 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-16 11:32:31,379 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-16 11:32:31,388 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-16 11:33:18,297 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-16 11:34:11,997 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-16 11:34:46,480 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-16 11:34:48,484 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-16 11:36:00,627 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-16 11:36:19,369 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-16 11:37:02,545 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-16 11:37:25,858 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-16 11:37:43,920 INFO [train.py:1227] (1/4) Maximum memory allocated so far is 10794MB 2023-04-16 11:37:45,087 INFO [train.py:1227] (1/4) Maximum memory allocated so far is 11934MB 2023-04-16 11:37:46,438 INFO [train.py:1227] (1/4) Maximum memory allocated so far is 11934MB 2023-04-16 11:37:47,787 INFO [train.py:1227] (1/4) Maximum memory allocated so far is 11934MB 2023-04-16 11:37:49,469 INFO [train.py:1227] (1/4) Maximum memory allocated so far is 11934MB 2023-04-16 11:37:51,086 INFO [train.py:1227] (1/4) Maximum memory allocated so far is 11934MB 2023-04-16 11:38:07,909 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 11:38:11,915 INFO [train.py:893] (1/4) Epoch 1, batch 0, loss[loss=7.236, simple_loss=6.563, pruned_loss=6.713, over 13473.00 frames. ], tot_loss[loss=7.236, simple_loss=6.563, pruned_loss=6.713, over 13473.00 frames. ], batch size: 93, lr: 2.50e-02, grad_scale: 2.0 2023-04-16 11:38:11,915 INFO [train.py:918] (1/4) Computing validation loss 2023-04-16 11:38:34,890 INFO [train.py:927] (1/4) Epoch 1, validation: loss=6.498, simple_loss=5.86, pruned_loss=6.372, over 2446609.00 frames. 2023-04-16 11:38:34,891 INFO [train.py:928] (1/4) Maximum memory allocated so far is 11934MB 2023-04-16 11:38:35,112 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5042, 4.5041, 4.5041, 4.5040, 4.5039, 4.5041, 4.5041, 4.5040], device='cuda:1'), covar=tensor([0.0048, 0.0035, 0.0031, 0.0021, 0.0039, 0.0038, 0.0031, 0.0028], device='cuda:1'), in_proj_covar=tensor([0.0009, 0.0008, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009], device='cuda:1'), out_proj_covar=tensor([8.9688e-06, 8.7191e-06, 8.7382e-06, 8.6879e-06, 8.7628e-06, 8.6855e-06, 8.6903e-06, 8.8188e-06], device='cuda:1') 2023-04-16 11:38:37,649 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5.0, num_to_drop=2, layers_to_drop={2, 3} 2023-04-16 11:38:49,407 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 11:38:49,609 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=6.43 vs. limit=2.0 2023-04-16 11:39:02,785 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=139.62 vs. limit=5.0 2023-04-16 11:39:03,997 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=5.35 vs. limit=2.0 2023-04-16 11:39:08,540 INFO [train.py:893] (1/4) Epoch 1, batch 50, loss[loss=1.008, simple_loss=0.8942, pruned_loss=1.018, over 13507.00 frames. ], tot_loss[loss=1.887, simple_loss=1.717, pruned_loss=1.644, over 603691.77 frames. ], batch size: 76, lr: 2.75e-02, grad_scale: 2.0 2023-04-16 11:39:28,042 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-16 11:39:28,043 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-16 11:39:28,043 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-16 11:39:28,049 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-16 11:39:28,065 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-16 11:39:28,085 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-16 11:39:28,095 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-16 11:39:31,278 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=83.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 11:39:44,031 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 4.680e+01 1.295e+02 2.313e+02 7.962e+02 1.225e+04, threshold=4.625e+02, percent-clipped=0.0 2023-04-16 11:39:44,057 INFO [train.py:893] (1/4) Epoch 1, batch 100, loss[loss=0.8327, simple_loss=0.7164, pruned_loss=0.9235, over 13402.00 frames. ], tot_loss[loss=1.375, simple_loss=1.231, pruned_loss=1.298, over 1057268.39 frames. ], batch size: 65, lr: 3.00e-02, grad_scale: 2.0 2023-04-16 11:39:55,191 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1384, 4.1414, 4.1422, 4.1419, 4.1390, 4.1387, 4.1406, 4.1413], device='cuda:1'), covar=tensor([0.0020, 0.0015, 0.0012, 0.0015, 0.0016, 0.0025, 0.0017, 0.0020], device='cuda:1'), in_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009], device='cuda:1'), out_proj_covar=tensor([8.9606e-06, 9.0606e-06, 9.0096e-06, 8.9656e-06, 9.2977e-06, 9.0561e-06, 9.0831e-06, 9.1578e-06], device='cuda:1') 2023-04-16 11:39:59,248 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=29.33 vs. limit=5.0 2023-04-16 11:40:14,341 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=144.0, num_to_drop=2, layers_to_drop={2, 3} 2023-04-16 11:40:18,444 INFO [train.py:893] (1/4) Epoch 1, batch 150, loss[loss=0.828, simple_loss=0.7074, pruned_loss=0.877, over 13528.00 frames. ], tot_loss[loss=1.155, simple_loss=1.021, pruned_loss=1.136, over 1417078.45 frames. ], batch size: 72, lr: 3.25e-02, grad_scale: 2.0 2023-04-16 11:40:55,351 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 5.673e+01 1.416e+02 2.268e+02 3.806e+02 1.185e+03, threshold=4.536e+02, percent-clipped=12.0 2023-04-16 11:40:55,376 INFO [train.py:893] (1/4) Epoch 1, batch 200, loss[loss=0.8135, simple_loss=0.6944, pruned_loss=0.8088, over 13515.00 frames. ], tot_loss[loss=1.029, simple_loss=0.9015, pruned_loss=1.022, over 1693421.26 frames. ], batch size: 98, lr: 3.50e-02, grad_scale: 2.0 2023-04-16 11:41:07,900 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3250, 4.3254, 4.3256, 4.3262, 4.3247, 4.3228, 4.3244, 4.3261], device='cuda:1'), covar=tensor([0.0020, 0.0018, 0.0016, 0.0020, 0.0018, 0.0019, 0.0022, 0.0022], device='cuda:1'), in_proj_covar=tensor([0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009, 0.0009], device='cuda:1'), out_proj_covar=tensor([9.2305e-06, 9.4543e-06, 9.2822e-06, 9.1951e-06, 9.5140e-06, 9.3228e-06, 9.3267e-06, 9.3865e-06], device='cuda:1') 2023-04-16 11:41:08,727 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.42 vs. limit=2.0 2023-04-16 11:41:25,350 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.96 vs. limit=5.0 2023-04-16 11:41:29,423 INFO [train.py:893] (1/4) Epoch 1, batch 250, loss[loss=0.7953, simple_loss=0.6764, pruned_loss=0.7578, over 13202.00 frames. ], tot_loss[loss=0.9466, simple_loss=0.8238, pruned_loss=0.9352, over 1903714.60 frames. ], batch size: 117, lr: 3.75e-02, grad_scale: 2.0 2023-04-16 11:42:01,904 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=296.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 11:42:04,236 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=300.0, num_to_drop=2, layers_to_drop={1, 2} 2023-04-16 11:42:04,602 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 7.029e+01 1.011e+02 1.415e+02 2.248e+02 8.541e+02, threshold=2.829e+02, percent-clipped=4.0 2023-04-16 11:42:04,635 INFO [train.py:893] (1/4) Epoch 1, batch 300, loss[loss=0.7492, simple_loss=0.6305, pruned_loss=0.7057, over 13551.00 frames. ], tot_loss[loss=0.8938, simple_loss=0.7728, pruned_loss=0.8732, over 2071762.34 frames. ], batch size: 78, lr: 4.00e-02, grad_scale: 2.0 2023-04-16 11:42:38,586 INFO [train.py:893] (1/4) Epoch 1, batch 350, loss[loss=0.7899, simple_loss=0.6611, pruned_loss=0.7241, over 13529.00 frames. ], tot_loss[loss=0.8569, simple_loss=0.7358, pruned_loss=0.8269, over 2202603.42 frames. ], batch size: 98, lr: 4.25e-02, grad_scale: 2.0 2023-04-16 11:42:43,086 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=357.0, num_to_drop=2, layers_to_drop={1, 2} 2023-04-16 11:43:03,965 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=387.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 11:43:13,443 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 6.345e+01 9.312e+01 1.191e+02 1.814e+02 3.998e+02, threshold=2.382e+02, percent-clipped=6.0 2023-04-16 11:43:13,479 INFO [train.py:893] (1/4) Epoch 1, batch 400, loss[loss=0.7854, simple_loss=0.6557, pruned_loss=0.6969, over 13267.00 frames. ], tot_loss[loss=0.8304, simple_loss=0.7084, pruned_loss=0.7896, over 2303827.06 frames. ], batch size: 124, lr: 4.50e-02, grad_scale: 4.0 2023-04-16 11:43:39,289 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=439.0, num_to_drop=2, layers_to_drop={0, 3} 2023-04-16 11:43:46,657 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=448.0, num_to_drop=2, layers_to_drop={1, 2} 2023-04-16 11:43:48,216 INFO [train.py:893] (1/4) Epoch 1, batch 450, loss[loss=0.8037, simple_loss=0.6647, pruned_loss=0.706, over 13255.00 frames. ], tot_loss[loss=0.8123, simple_loss=0.6885, pruned_loss=0.7605, over 2381580.54 frames. ], batch size: 124, lr: 4.75e-02, grad_scale: 4.0 2023-04-16 11:44:01,678 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=8.22 vs. limit=5.0 2023-04-16 11:44:05,726 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-16 11:44:09,236 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=7.10 vs. limit=2.0 2023-04-16 11:44:21,852 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 7.092e+01 1.034e+02 1.217e+02 1.498e+02 2.544e+02, threshold=2.433e+02, percent-clipped=2.0 2023-04-16 11:44:21,876 INFO [train.py:893] (1/4) Epoch 1, batch 500, loss[loss=0.7432, simple_loss=0.6092, pruned_loss=0.6464, over 13488.00 frames. ], tot_loss[loss=0.7962, simple_loss=0.6707, pruned_loss=0.7337, over 2442655.95 frames. ], batch size: 93, lr: 4.99e-02, grad_scale: 4.0 2023-04-16 11:44:25,287 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=11.21 vs. limit=5.0 2023-04-16 11:44:57,636 INFO [train.py:893] (1/4) Epoch 1, batch 550, loss[loss=0.6917, simple_loss=0.57, pruned_loss=0.5754, over 13487.00 frames. ], tot_loss[loss=0.7828, simple_loss=0.6561, pruned_loss=0.7079, over 2490823.78 frames. ], batch size: 70, lr: 4.98e-02, grad_scale: 4.0 2023-04-16 11:44:59,195 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=8.73 vs. limit=5.0 2023-04-16 11:45:04,638 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=562.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 11:45:08,275 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=568.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 11:45:24,236 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=590.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 11:45:30,978 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=600.0, num_to_drop=2, layers_to_drop={0, 1} 2023-04-16 11:45:31,314 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 8.423e+01 1.230e+02 1.562e+02 1.964e+02 3.659e+02, threshold=3.124e+02, percent-clipped=9.0 2023-04-16 11:45:31,339 INFO [train.py:893] (1/4) Epoch 1, batch 600, loss[loss=0.6776, simple_loss=0.5612, pruned_loss=0.5419, over 12057.00 frames. ], tot_loss[loss=0.7659, simple_loss=0.6402, pruned_loss=0.6772, over 2529658.06 frames. ], batch size: 158, lr: 4.98e-02, grad_scale: 4.0 2023-04-16 11:45:40,754 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.09 vs. limit=2.0 2023-04-16 11:45:47,912 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=623.0, num_to_drop=2, layers_to_drop={0, 1} 2023-04-16 11:45:51,470 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=629.0, num_to_drop=2, layers_to_drop={2, 3} 2023-04-16 11:46:05,790 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=648.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 11:46:07,486 INFO [train.py:893] (1/4) Epoch 1, batch 650, loss[loss=0.6952, simple_loss=0.5829, pruned_loss=0.5272, over 13392.00 frames. ], tot_loss[loss=0.747, simple_loss=0.624, pruned_loss=0.6436, over 2557226.46 frames. ], batch size: 113, lr: 4.98e-02, grad_scale: 4.0 2023-04-16 11:46:07,793 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=651.0, num_to_drop=2, layers_to_drop={1, 2} 2023-04-16 11:46:08,324 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=652.0, num_to_drop=2, layers_to_drop={2, 3} 2023-04-16 11:46:42,080 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.101e+02 2.199e+02 2.780e+02 3.736e+02 1.020e+03, threshold=5.560e+02, percent-clipped=40.0 2023-04-16 11:46:42,105 INFO [train.py:893] (1/4) Epoch 1, batch 700, loss[loss=0.6124, simple_loss=0.5193, pruned_loss=0.4432, over 13266.00 frames. ], tot_loss[loss=0.7266, simple_loss=0.6079, pruned_loss=0.6083, over 2581668.29 frames. ], batch size: 124, lr: 4.98e-02, grad_scale: 4.0 2023-04-16 11:46:51,626 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.85 vs. limit=2.0 2023-04-16 11:47:09,399 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=739.0, num_to_drop=2, layers_to_drop={0, 1} 2023-04-16 11:47:09,997 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1037, 3.1093, 3.5191, 3.0200, 3.4626, 3.8653, 3.1986, 3.5828], device='cuda:1'), covar=tensor([0.4784, 0.3920, 0.3054, 0.4206, 0.3695, 0.2241, 0.4654, 0.3377], device='cuda:1'), in_proj_covar=tensor([0.0027, 0.0025, 0.0024, 0.0030, 0.0026, 0.0022, 0.0026, 0.0025], device='cuda:1'), out_proj_covar=tensor([2.4527e-05, 2.4629e-05, 2.3110e-05, 2.5905e-05, 2.3945e-05, 2.1172e-05, 2.3305e-05, 2.0444e-05], device='cuda:1') 2023-04-16 11:47:11,709 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=743.0, num_to_drop=2, layers_to_drop={0, 3} 2023-04-16 11:47:16,213 INFO [train.py:893] (1/4) Epoch 1, batch 750, loss[loss=0.5477, simple_loss=0.4618, pruned_loss=0.3942, over 12559.00 frames. ], tot_loss[loss=0.7064, simple_loss=0.592, pruned_loss=0.5753, over 2596270.74 frames. ], batch size: 51, lr: 4.97e-02, grad_scale: 4.0 2023-04-16 11:47:20,906 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 2023-04-16 11:47:41,177 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=787.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 11:47:51,428 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.745e+02 2.893e+02 3.595e+02 5.208e+02 1.075e+03, threshold=7.189e+02, percent-clipped=20.0 2023-04-16 11:47:51,453 INFO [train.py:893] (1/4) Epoch 1, batch 800, loss[loss=0.5537, simple_loss=0.4628, pruned_loss=0.3987, over 12672.00 frames. ], tot_loss[loss=0.6886, simple_loss=0.5786, pruned_loss=0.545, over 2612812.16 frames. ], batch size: 52, lr: 4.97e-02, grad_scale: 8.0 2023-04-16 11:48:01,753 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 2023-04-16 11:48:08,370 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=2.01 vs. limit=2.0 2023-04-16 11:48:23,023 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=847.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 11:48:25,236 INFO [train.py:893] (1/4) Epoch 1, batch 850, loss[loss=0.6175, simple_loss=0.5273, pruned_loss=0.4191, over 13383.00 frames. ], tot_loss[loss=0.6708, simple_loss=0.5658, pruned_loss=0.5156, over 2628441.53 frames. ], batch size: 113, lr: 4.96e-02, grad_scale: 8.0 2023-04-16 11:48:39,215 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8642, 3.1322, 3.2827, 3.0326, 2.6410, 2.4446, 2.9887, 2.8926], device='cuda:1'), covar=tensor([0.3247, 0.2835, 0.2425, 0.2995, 0.3733, 0.4872, 0.2714, 0.2680], device='cuda:1'), in_proj_covar=tensor([0.0017, 0.0016, 0.0015, 0.0015, 0.0017, 0.0019, 0.0015, 0.0015], device='cuda:1'), out_proj_covar=tensor([1.3143e-05, 1.2693e-05, 1.1750e-05, 1.3565e-05, 1.4448e-05, 1.5889e-05, 1.2464e-05, 1.2399e-05], device='cuda:1') 2023-04-16 11:49:00,835 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.621e+02 2.908e+02 3.677e+02 4.771e+02 1.382e+03, threshold=7.354e+02, percent-clipped=2.0 2023-04-16 11:49:00,871 INFO [train.py:893] (1/4) Epoch 1, batch 900, loss[loss=0.591, simple_loss=0.5035, pruned_loss=0.3976, over 13038.00 frames. ], tot_loss[loss=0.6513, simple_loss=0.5509, pruned_loss=0.4877, over 2637283.59 frames. ], batch size: 142, lr: 4.96e-02, grad_scale: 8.0 2023-04-16 11:49:05,321 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=908.0, num_to_drop=2, layers_to_drop={1, 2} 2023-04-16 11:49:06,533 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.9877, 4.8780, 4.6841, 4.8317, 4.8886, 4.8528, 5.1133, 4.9054], device='cuda:1'), covar=tensor([0.0608, 0.0992, 0.1182, 0.0840, 0.0894, 0.0946, 0.0549, 0.0912], device='cuda:1'), in_proj_covar=tensor([0.0038, 0.0037, 0.0036, 0.0029, 0.0027, 0.0035, 0.0033, 0.0036], device='cuda:1'), out_proj_covar=tensor([3.3993e-05, 3.4834e-05, 3.3383e-05, 2.6627e-05, 2.5658e-05, 3.2629e-05, 3.0710e-05, 3.3459e-05], device='cuda:1') 2023-04-16 11:49:11,505 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=918.0, num_to_drop=2, layers_to_drop={1, 3} 2023-04-16 11:49:15,781 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=924.0, num_to_drop=2, layers_to_drop={0, 3} 2023-04-16 11:49:23,726 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-16 11:49:31,101 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=946.0, num_to_drop=2, layers_to_drop={1, 2} 2023-04-16 11:49:34,003 INFO [train.py:893] (1/4) Epoch 1, batch 950, loss[loss=0.5911, simple_loss=0.5081, pruned_loss=0.386, over 13534.00 frames. ], tot_loss[loss=0.6325, simple_loss=0.5368, pruned_loss=0.4619, over 2647439.62 frames. ], batch size: 83, lr: 4.96e-02, grad_scale: 8.0 2023-04-16 11:49:34,784 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=952.0, num_to_drop=2, layers_to_drop={1, 2} 2023-04-16 11:49:37,846 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.01 vs. limit=5.0 2023-04-16 11:50:09,401 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1000.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 11:50:09,806 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.606e+02 3.300e+02 3.832e+02 5.280e+02 1.324e+03, threshold=7.664e+02, percent-clipped=10.0 2023-04-16 11:50:09,830 INFO [train.py:893] (1/4) Epoch 1, batch 1000, loss[loss=0.596, simple_loss=0.5152, pruned_loss=0.381, over 13462.00 frames. ], tot_loss[loss=0.6154, simple_loss=0.5239, pruned_loss=0.4391, over 2645026.52 frames. ], batch size: 106, lr: 4.95e-02, grad_scale: 8.0 2023-04-16 11:50:24,000 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.85 vs. limit=2.0 2023-04-16 11:50:39,203 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1043.0, num_to_drop=2, layers_to_drop={1, 2} 2023-04-16 11:50:42,641 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 2023-04-16 11:50:44,818 INFO [train.py:893] (1/4) Epoch 1, batch 1050, loss[loss=0.5763, simple_loss=0.4984, pruned_loss=0.3646, over 13521.00 frames. ], tot_loss[loss=0.5985, simple_loss=0.5114, pruned_loss=0.4172, over 2649427.58 frames. ], batch size: 83, lr: 4.95e-02, grad_scale: 8.0 2023-04-16 11:51:14,985 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1091.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 11:51:21,210 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.895e+02 3.452e+02 4.416e+02 5.723e+02 1.913e+03, threshold=8.832e+02, percent-clipped=10.0 2023-04-16 11:51:21,234 INFO [train.py:893] (1/4) Epoch 1, batch 1100, loss[loss=0.5351, simple_loss=0.4711, pruned_loss=0.3252, over 13266.00 frames. ], tot_loss[loss=0.5855, simple_loss=0.5027, pruned_loss=0.3987, over 2654368.86 frames. ], batch size: 124, lr: 4.94e-02, grad_scale: 8.0 2023-04-16 11:51:57,241 INFO [train.py:893] (1/4) Epoch 1, batch 1150, loss[loss=0.5441, simple_loss=0.4776, pruned_loss=0.33, over 13271.00 frames. ], tot_loss[loss=0.5739, simple_loss=0.4951, pruned_loss=0.3821, over 2658462.19 frames. ], batch size: 124, lr: 4.94e-02, grad_scale: 8.0 2023-04-16 11:51:59,525 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1153.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 11:52:30,443 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=6.20 vs. limit=5.0 2023-04-16 11:52:32,378 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.264e+02 3.530e+02 4.274e+02 5.391e+02 1.250e+03, threshold=8.549e+02, percent-clipped=2.0 2023-04-16 11:52:32,403 INFO [train.py:893] (1/4) Epoch 1, batch 1200, loss[loss=0.5534, simple_loss=0.4861, pruned_loss=0.333, over 13397.00 frames. ], tot_loss[loss=0.5628, simple_loss=0.4871, pruned_loss=0.3678, over 2657255.04 frames. ], batch size: 113, lr: 4.93e-02, grad_scale: 8.0 2023-04-16 11:52:33,906 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1203.0, num_to_drop=2, layers_to_drop={1, 2} 2023-04-16 11:52:42,067 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1214.0, num_to_drop=2, layers_to_drop={1, 3} 2023-04-16 11:52:45,928 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1218.0, num_to_drop=2, layers_to_drop={0, 2} 2023-04-16 11:52:49,480 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1224.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 11:52:53,007 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-16 11:53:01,427 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-16 11:53:04,726 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1246.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 11:53:09,102 INFO [train.py:893] (1/4) Epoch 1, batch 1250, loss[loss=0.5637, simple_loss=0.4854, pruned_loss=0.3472, over 13084.00 frames. ], tot_loss[loss=0.5558, simple_loss=0.4821, pruned_loss=0.3574, over 2657126.85 frames. ], batch size: 142, lr: 4.92e-02, grad_scale: 8.0 2023-04-16 11:53:18,600 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1266.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 11:53:22,791 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1272.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 11:53:30,618 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5456, 4.7055, 4.6199, 4.5418, 4.8723, 4.6175, 4.8326, 4.7256], device='cuda:1'), covar=tensor([0.0628, 0.0611, 0.0661, 0.0396, 0.0475, 0.0543, 0.0669, 0.0633], device='cuda:1'), in_proj_covar=tensor([0.0039, 0.0040, 0.0039, 0.0037, 0.0039, 0.0037, 0.0031, 0.0035], device='cuda:1'), out_proj_covar=tensor([3.4461e-05, 3.5397e-05, 3.6920e-05, 3.2084e-05, 3.5471e-05, 3.3827e-05, 3.1589e-05, 3.1731e-05], device='cuda:1') 2023-04-16 11:53:39,152 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1294.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 11:53:44,032 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.345e+02 3.938e+02 4.877e+02 6.128e+02 1.226e+03, threshold=9.754e+02, percent-clipped=3.0 2023-04-16 11:53:44,057 INFO [train.py:893] (1/4) Epoch 1, batch 1300, loss[loss=0.4556, simple_loss=0.4018, pruned_loss=0.2692, over 12807.00 frames. ], tot_loss[loss=0.5497, simple_loss=0.4779, pruned_loss=0.3481, over 2655240.43 frames. ], batch size: 52, lr: 4.92e-02, grad_scale: 8.0 2023-04-16 11:54:16,685 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6606, 3.8669, 3.4516, 3.2705, 3.6860, 3.0024, 3.4259, 3.9678], device='cuda:1'), covar=tensor([0.0404, 0.0496, 0.0594, 0.0688, 0.0573, 0.0891, 0.0681, 0.0467], device='cuda:1'), in_proj_covar=tensor([0.0054, 0.0047, 0.0047, 0.0056, 0.0044, 0.0052, 0.0053, 0.0045], device='cuda:1'), out_proj_covar=tensor([4.3921e-05, 4.0047e-05, 3.8286e-05, 4.8529e-05, 3.6151e-05, 4.4476e-05, 4.5617e-05, 3.7228e-05], device='cuda:1') 2023-04-16 11:54:21,042 INFO [train.py:893] (1/4) Epoch 1, batch 1350, loss[loss=0.4982, simple_loss=0.4507, pruned_loss=0.2822, over 13358.00 frames. ], tot_loss[loss=0.5417, simple_loss=0.4725, pruned_loss=0.3378, over 2660245.40 frames. ], batch size: 73, lr: 4.91e-02, grad_scale: 8.0 2023-04-16 11:54:57,881 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.254e+02 3.467e+02 4.556e+02 5.828e+02 1.072e+03, threshold=9.111e+02, percent-clipped=1.0 2023-04-16 11:54:57,905 INFO [train.py:893] (1/4) Epoch 1, batch 1400, loss[loss=0.5063, simple_loss=0.4481, pruned_loss=0.2946, over 13521.00 frames. ], tot_loss[loss=0.5312, simple_loss=0.4649, pruned_loss=0.3268, over 2660649.22 frames. ], batch size: 70, lr: 4.91e-02, grad_scale: 8.0 2023-04-16 11:55:22,723 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1434.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 11:55:35,003 INFO [train.py:893] (1/4) Epoch 1, batch 1450, loss[loss=0.4451, simple_loss=0.3897, pruned_loss=0.2613, over 12649.00 frames. ], tot_loss[loss=0.5247, simple_loss=0.4603, pruned_loss=0.3188, over 2661194.17 frames. ], batch size: 52, lr: 4.90e-02, grad_scale: 8.0 2023-04-16 11:56:07,306 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.21 vs. limit=2.0 2023-04-16 11:56:08,363 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1495.0, num_to_drop=2, layers_to_drop={0, 3} 2023-04-16 11:56:12,847 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.702e+02 4.580e+02 5.694e+02 7.552e+02 1.475e+03, threshold=1.139e+03, percent-clipped=15.0 2023-04-16 11:56:12,883 INFO [train.py:893] (1/4) Epoch 1, batch 1500, loss[loss=0.5388, simple_loss=0.4836, pruned_loss=0.3052, over 13374.00 frames. ], tot_loss[loss=0.5196, simple_loss=0.4571, pruned_loss=0.312, over 2657311.61 frames. ], batch size: 109, lr: 4.89e-02, grad_scale: 8.0 2023-04-16 11:56:14,551 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1503.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 11:56:19,313 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1509.0, num_to_drop=2, layers_to_drop={1, 3} 2023-04-16 11:56:22,725 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.8025, 4.8921, 4.7157, 4.4966, 5.0265, 4.6052, 5.0997, 5.0138], device='cuda:1'), covar=tensor([0.0550, 0.0497, 0.0521, 0.0511, 0.0506, 0.0577, 0.0442, 0.0378], device='cuda:1'), in_proj_covar=tensor([0.0053, 0.0052, 0.0052, 0.0051, 0.0051, 0.0052, 0.0039, 0.0047], device='cuda:1'), out_proj_covar=tensor([4.8077e-05, 4.7877e-05, 4.9568e-05, 4.3898e-05, 4.8207e-05, 4.7999e-05, 4.4005e-05, 4.1517e-05], device='cuda:1') 2023-04-16 11:56:49,935 INFO [train.py:893] (1/4) Epoch 1, batch 1550, loss[loss=0.5178, simple_loss=0.4636, pruned_loss=0.2932, over 13444.00 frames. ], tot_loss[loss=0.5154, simple_loss=0.4546, pruned_loss=0.3062, over 2658756.08 frames. ], batch size: 95, lr: 4.89e-02, grad_scale: 8.0 2023-04-16 11:56:50,118 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1551.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 11:56:57,640 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=2.01 vs. limit=2.0 2023-04-16 11:57:02,064 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1566.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 11:57:29,611 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.538e+02 4.492e+02 5.897e+02 7.825e+02 1.707e+03, threshold=1.179e+03, percent-clipped=8.0 2023-04-16 11:57:29,635 INFO [train.py:893] (1/4) Epoch 1, batch 1600, loss[loss=0.5416, simple_loss=0.4749, pruned_loss=0.313, over 13546.00 frames. ], tot_loss[loss=0.5103, simple_loss=0.452, pruned_loss=0.2997, over 2660355.46 frames. ], batch size: 91, lr: 4.88e-02, grad_scale: 8.0 2023-04-16 11:57:31,856 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.7121, 5.2269, 5.3182, 5.5833, 5.5503, 5.3517, 5.4080, 5.4148], device='cuda:1'), covar=tensor([0.0375, 0.0469, 0.0353, 0.0365, 0.0312, 0.0480, 0.0373, 0.0402], device='cuda:1'), in_proj_covar=tensor([0.0074, 0.0081, 0.0074, 0.0074, 0.0060, 0.0078, 0.0074, 0.0073], device='cuda:1'), out_proj_covar=tensor([7.7222e-05, 8.0073e-05, 7.4890e-05, 7.1266e-05, 6.1140e-05, 7.4862e-05, 7.4469e-05, 7.0186e-05], device='cuda:1') 2023-04-16 11:57:36,693 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1611.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 11:57:49,451 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9043, 2.8820, 3.1179, 2.6123, 2.3293, 2.4936, 2.9451, 3.0154], device='cuda:1'), covar=tensor([0.0375, 0.0329, 0.0425, 0.0930, 0.0350, 0.0744, 0.0648, 0.0551], device='cuda:1'), in_proj_covar=tensor([0.0049, 0.0048, 0.0053, 0.0049, 0.0046, 0.0045, 0.0050, 0.0046], device='cuda:1'), out_proj_covar=tensor([4.2584e-05, 4.3413e-05, 4.5449e-05, 4.3142e-05, 4.1356e-05, 4.4706e-05, 4.5703e-05, 3.9752e-05], device='cuda:1') 2023-04-16 11:57:49,456 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1627.0, num_to_drop=2, layers_to_drop={1, 3} 2023-04-16 11:58:06,419 INFO [train.py:893] (1/4) Epoch 1, batch 1650, loss[loss=0.5145, simple_loss=0.4661, pruned_loss=0.2856, over 13445.00 frames. ], tot_loss[loss=0.5042, simple_loss=0.4482, pruned_loss=0.2932, over 2657453.91 frames. ], batch size: 106, lr: 4.87e-02, grad_scale: 8.0 2023-04-16 11:58:23,357 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1672.0, num_to_drop=2, layers_to_drop={0, 2} 2023-04-16 11:58:36,725 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.05 vs. limit=5.0 2023-04-16 11:58:41,012 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.1748, 5.2837, 5.2837, 5.0521, 5.5582, 5.0095, 5.6062, 5.4940], device='cuda:1'), covar=tensor([0.0361, 0.0320, 0.0320, 0.0349, 0.0312, 0.0361, 0.0222, 0.0232], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0065, 0.0061, 0.0062, 0.0065, 0.0066, 0.0045, 0.0059], device='cuda:1'), out_proj_covar=tensor([6.0237e-05, 6.1811e-05, 6.1395e-05, 5.5621e-05, 6.3973e-05, 6.3762e-05, 5.1558e-05, 5.3518e-05], device='cuda:1') 2023-04-16 11:58:45,612 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.181e+02 4.485e+02 5.687e+02 7.651e+02 1.834e+03, threshold=1.137e+03, percent-clipped=3.0 2023-04-16 11:58:45,636 INFO [train.py:893] (1/4) Epoch 1, batch 1700, loss[loss=0.5011, simple_loss=0.4549, pruned_loss=0.2769, over 13519.00 frames. ], tot_loss[loss=0.4977, simple_loss=0.4446, pruned_loss=0.2863, over 2657702.13 frames. ], batch size: 98, lr: 4.86e-02, grad_scale: 8.0 2023-04-16 11:59:21,487 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 2023-04-16 11:59:24,344 INFO [train.py:893] (1/4) Epoch 1, batch 1750, loss[loss=0.4645, simple_loss=0.4194, pruned_loss=0.2575, over 13461.00 frames. ], tot_loss[loss=0.4888, simple_loss=0.4386, pruned_loss=0.2785, over 2655259.08 frames. ], batch size: 100, lr: 4.86e-02, grad_scale: 8.0 2023-04-16 11:59:29,708 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.10 vs. limit=5.0 2023-04-16 11:59:54,511 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1790.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 11:59:56,299 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 2023-04-16 12:00:03,038 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.329e+02 4.138e+02 5.716e+02 7.236e+02 4.256e+03, threshold=1.143e+03, percent-clipped=8.0 2023-04-16 12:00:03,062 INFO [train.py:893] (1/4) Epoch 1, batch 1800, loss[loss=0.4286, simple_loss=0.4012, pruned_loss=0.2285, over 13354.00 frames. ], tot_loss[loss=0.4827, simple_loss=0.4347, pruned_loss=0.2727, over 2655298.81 frames. ], batch size: 73, lr: 4.85e-02, grad_scale: 8.0 2023-04-16 12:00:09,432 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=1809.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 12:00:32,077 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1838.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 12:00:42,430 INFO [train.py:893] (1/4) Epoch 1, batch 1850, loss[loss=0.4449, simple_loss=0.4056, pruned_loss=0.2433, over 13545.00 frames. ], tot_loss[loss=0.4775, simple_loss=0.4314, pruned_loss=0.2678, over 2658570.72 frames. ], batch size: 72, lr: 4.84e-02, grad_scale: 8.0 2023-04-16 12:00:43,962 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-16 12:00:47,683 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=1857.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:00:47,758 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1857.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 12:00:57,452 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=1870.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 12:01:20,453 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1899.0, num_to_drop=2, layers_to_drop={0, 2} 2023-04-16 12:01:21,558 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.333e+02 4.117e+02 5.326e+02 6.729e+02 1.222e+03, threshold=1.065e+03, percent-clipped=1.0 2023-04-16 12:01:21,584 INFO [train.py:893] (1/4) Epoch 1, batch 1900, loss[loss=0.4608, simple_loss=0.4188, pruned_loss=0.2522, over 13544.00 frames. ], tot_loss[loss=0.472, simple_loss=0.4281, pruned_loss=0.2629, over 2660081.49 frames. ], batch size: 83, lr: 4.83e-02, grad_scale: 8.0 2023-04-16 12:01:35,064 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1918.0, num_to_drop=2, layers_to_drop={0, 1} 2023-04-16 12:01:38,465 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1922.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 12:01:45,407 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=1931.0, num_to_drop=2, layers_to_drop={1, 3} 2023-04-16 12:02:00,339 INFO [train.py:893] (1/4) Epoch 1, batch 1950, loss[loss=0.4289, simple_loss=0.4034, pruned_loss=0.2273, over 13355.00 frames. ], tot_loss[loss=0.4674, simple_loss=0.425, pruned_loss=0.2588, over 2659726.34 frames. ], batch size: 73, lr: 4.83e-02, grad_scale: 8.0 2023-04-16 12:02:00,747 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.11 vs. limit=2.0 2023-04-16 12:02:13,043 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=1967.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:02:43,807 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.097e+02 4.764e+02 5.634e+02 7.381e+02 1.662e+03, threshold=1.127e+03, percent-clipped=9.0 2023-04-16 12:02:43,831 INFO [train.py:893] (1/4) Epoch 1, batch 2000, loss[loss=0.399, simple_loss=0.3621, pruned_loss=0.2179, over 12752.00 frames. ], tot_loss[loss=0.4653, simple_loss=0.4239, pruned_loss=0.2564, over 2654836.89 frames. ], batch size: 52, lr: 4.82e-02, grad_scale: 16.0 2023-04-16 12:02:50,065 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-16 12:03:26,602 INFO [train.py:893] (1/4) Epoch 1, batch 2050, loss[loss=0.3965, simple_loss=0.3871, pruned_loss=0.203, over 13486.00 frames. ], tot_loss[loss=0.4611, simple_loss=0.4227, pruned_loss=0.2521, over 2659255.09 frames. ], batch size: 93, lr: 4.81e-02, grad_scale: 16.0 2023-04-16 12:03:32,306 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3498, 3.8761, 3.4381, 3.6702, 3.7310, 3.5727, 3.6197, 3.7485], device='cuda:1'), covar=tensor([0.0284, 0.0215, 0.0337, 0.0485, 0.0174, 0.0413, 0.0347, 0.0355], device='cuda:1'), in_proj_covar=tensor([0.0045, 0.0038, 0.0040, 0.0043, 0.0037, 0.0038, 0.0045, 0.0040], device='cuda:1'), out_proj_covar=tensor([3.6629e-05, 3.1064e-05, 3.0964e-05, 3.6327e-05, 2.9573e-05, 3.2128e-05, 3.7892e-05, 3.2035e-05], device='cuda:1') 2023-04-16 12:04:00,145 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2090.0, num_to_drop=2, layers_to_drop={0, 1} 2023-04-16 12:04:09,338 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.349e+02 3.615e+02 4.558e+02 5.619e+02 1.123e+03, threshold=9.116e+02, percent-clipped=0.0 2023-04-16 12:04:09,363 INFO [train.py:893] (1/4) Epoch 1, batch 2100, loss[loss=0.3637, simple_loss=0.3534, pruned_loss=0.187, over 13220.00 frames. ], tot_loss[loss=0.4525, simple_loss=0.4175, pruned_loss=0.2456, over 2658690.25 frames. ], batch size: 58, lr: 4.80e-02, grad_scale: 16.0 2023-04-16 12:04:32,743 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 2023-04-16 12:04:39,538 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2138.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:04:50,554 INFO [train.py:893] (1/4) Epoch 1, batch 2150, loss[loss=0.4045, simple_loss=0.3873, pruned_loss=0.2108, over 13537.00 frames. ], tot_loss[loss=0.4457, simple_loss=0.4139, pruned_loss=0.2402, over 2662272.40 frames. ], batch size: 72, lr: 4.79e-02, grad_scale: 16.0 2023-04-16 12:05:10,885 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.9889, 4.5542, 4.5570, 4.6433, 4.6013, 4.6153, 4.7486, 4.5034], device='cuda:1'), covar=tensor([0.0554, 0.0604, 0.1068, 0.0844, 0.0497, 0.0587, 0.0653, 0.0504], device='cuda:1'), in_proj_covar=tensor([0.0098, 0.0102, 0.0102, 0.0105, 0.0074, 0.0104, 0.0099, 0.0086], device='cuda:1'), out_proj_covar=tensor([1.1197e-04, 1.0219e-04, 1.0917e-04, 1.0577e-04, 7.7778e-05, 1.0528e-04, 1.0796e-04, 8.5082e-05], device='cuda:1') 2023-04-16 12:05:26,252 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2194.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:05:32,314 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.396e+02 4.119e+02 5.010e+02 7.199e+02 1.601e+03, threshold=1.002e+03, percent-clipped=13.0 2023-04-16 12:05:32,337 INFO [train.py:893] (1/4) Epoch 1, batch 2200, loss[loss=0.4174, simple_loss=0.4062, pruned_loss=0.2143, over 13453.00 frames. ], tot_loss[loss=0.4364, simple_loss=0.4087, pruned_loss=0.2332, over 2664102.57 frames. ], batch size: 100, lr: 4.78e-02, grad_scale: 16.0 2023-04-16 12:05:42,930 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2213.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:05:50,505 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2222.0, num_to_drop=2, layers_to_drop={1, 2} 2023-04-16 12:05:53,470 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2226.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:06:00,403 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.2018, 3.3476, 2.6336, 2.3592, 2.7109, 3.0213, 3.0598, 3.0078], device='cuda:1'), covar=tensor([0.0182, 0.0184, 0.0295, 0.0442, 0.0257, 0.0283, 0.0230, 0.0221], device='cuda:1'), in_proj_covar=tensor([0.0025, 0.0023, 0.0026, 0.0027, 0.0026, 0.0028, 0.0028, 0.0023], device='cuda:1'), out_proj_covar=tensor([2.0642e-05, 1.7833e-05, 2.0517e-05, 2.2743e-05, 2.0697e-05, 2.1932e-05, 2.3570e-05, 1.8814e-05], device='cuda:1') 2023-04-16 12:06:15,127 INFO [train.py:893] (1/4) Epoch 1, batch 2250, loss[loss=0.3934, simple_loss=0.3752, pruned_loss=0.2058, over 13223.00 frames. ], tot_loss[loss=0.4305, simple_loss=0.4054, pruned_loss=0.2287, over 2667896.56 frames. ], batch size: 132, lr: 4.77e-02, grad_scale: 16.0 2023-04-16 12:06:25,340 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8060, 3.8891, 3.9510, 3.2407, 4.2707, 3.7007, 4.0957, 3.6848], device='cuda:1'), covar=tensor([0.0256, 0.0209, 0.0216, 0.0305, 0.0212, 0.0163, 0.0390, 0.0326], device='cuda:1'), in_proj_covar=tensor([0.0021, 0.0021, 0.0018, 0.0025, 0.0017, 0.0019, 0.0019, 0.0020], device='cuda:1'), out_proj_covar=tensor([1.8912e-05, 1.8034e-05, 1.5018e-05, 2.2596e-05, 1.4630e-05, 1.6632e-05, 1.6546e-05, 1.7259e-05], device='cuda:1') 2023-04-16 12:06:28,342 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2267.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 12:06:31,137 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2270.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:06:56,372 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.310e+02 3.867e+02 4.930e+02 5.960e+02 1.008e+03, threshold=9.860e+02, percent-clipped=1.0 2023-04-16 12:06:56,397 INFO [train.py:893] (1/4) Epoch 1, batch 2300, loss[loss=0.4235, simple_loss=0.4052, pruned_loss=0.2209, over 13554.00 frames. ], tot_loss[loss=0.4251, simple_loss=0.4022, pruned_loss=0.2246, over 2667083.40 frames. ], batch size: 78, lr: 4.77e-02, grad_scale: 16.0 2023-04-16 12:07:01,970 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7078, 3.5030, 3.5364, 3.1887, 3.9121, 3.5342, 3.9284, 3.7257], device='cuda:1'), covar=tensor([0.0186, 0.0222, 0.0213, 0.0218, 0.0194, 0.0170, 0.0295, 0.0190], device='cuda:1'), in_proj_covar=tensor([0.0021, 0.0021, 0.0018, 0.0024, 0.0017, 0.0019, 0.0018, 0.0019], device='cuda:1'), out_proj_covar=tensor([1.8396e-05, 1.8312e-05, 1.5233e-05, 2.1891e-05, 1.4908e-05, 1.6262e-05, 1.6050e-05, 1.6753e-05], device='cuda:1') 2023-04-16 12:07:09,526 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2315.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:07:31,059 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2340.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 12:07:32,833 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0 2023-04-16 12:07:32,942 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.69 vs. limit=5.0 2023-04-16 12:07:33,408 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4545, 4.3731, 4.5081, 3.6886, 4.7891, 4.2289, 4.4729, 4.0536], device='cuda:1'), covar=tensor([0.0142, 0.0145, 0.0128, 0.0225, 0.0093, 0.0139, 0.0280, 0.0273], device='cuda:1'), in_proj_covar=tensor([0.0020, 0.0020, 0.0018, 0.0023, 0.0016, 0.0019, 0.0018, 0.0019], device='cuda:1'), out_proj_covar=tensor([1.8075e-05, 1.7870e-05, 1.5115e-05, 2.1089e-05, 1.4717e-05, 1.6541e-05, 1.5762e-05, 1.6571e-05], device='cuda:1') 2023-04-16 12:07:38,961 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 2023-04-16 12:07:40,009 INFO [train.py:893] (1/4) Epoch 1, batch 2350, loss[loss=0.3794, simple_loss=0.3743, pruned_loss=0.1923, over 13455.00 frames. ], tot_loss[loss=0.418, simple_loss=0.3978, pruned_loss=0.2196, over 2665148.99 frames. ], batch size: 100, lr: 4.76e-02, grad_scale: 16.0 2023-04-16 12:08:00,412 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-16 12:08:07,081 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-04-16 12:08:22,532 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.956e+02 4.306e+02 5.227e+02 6.724e+02 1.071e+03, threshold=1.045e+03, percent-clipped=3.0 2023-04-16 12:08:22,556 INFO [train.py:893] (1/4) Epoch 1, batch 2400, loss[loss=0.3287, simple_loss=0.3244, pruned_loss=0.1664, over 12733.00 frames. ], tot_loss[loss=0.4142, simple_loss=0.3951, pruned_loss=0.2171, over 2666912.59 frames. ], batch size: 52, lr: 4.75e-02, grad_scale: 16.0 2023-04-16 12:08:22,890 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2401.0, num_to_drop=2, layers_to_drop={2, 3} 2023-04-16 12:08:33,169 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.5898, 5.0864, 5.0980, 5.3215, 4.9352, 5.0519, 5.3864, 5.0305], device='cuda:1'), covar=tensor([0.0504, 0.0522, 0.1012, 0.0943, 0.0458, 0.0628, 0.0578, 0.0522], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0109, 0.0115, 0.0121, 0.0082, 0.0115, 0.0115, 0.0096], device='cuda:1'), out_proj_covar=tensor([1.3614e-04, 1.1057e-04, 1.2333e-04, 1.2715e-04, 8.6841e-05, 1.1867e-04, 1.3071e-04, 9.5397e-05], device='cuda:1') 2023-04-16 12:08:44,526 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4772, 3.0208, 3.6789, 2.8336, 3.8269, 3.3736, 3.7240, 3.5753], device='cuda:1'), covar=tensor([0.0255, 0.0312, 0.0196, 0.0251, 0.0258, 0.0212, 0.0391, 0.0232], device='cuda:1'), in_proj_covar=tensor([0.0022, 0.0022, 0.0018, 0.0024, 0.0017, 0.0019, 0.0019, 0.0020], device='cuda:1'), out_proj_covar=tensor([1.9894e-05, 2.0251e-05, 1.6206e-05, 2.2973e-05, 1.5564e-05, 1.7252e-05, 1.7163e-05, 1.8024e-05], device='cuda:1') 2023-04-16 12:09:04,547 INFO [train.py:893] (1/4) Epoch 1, batch 2450, loss[loss=0.3929, simple_loss=0.3827, pruned_loss=0.2015, over 13484.00 frames. ], tot_loss[loss=0.4114, simple_loss=0.3937, pruned_loss=0.2149, over 2668589.07 frames. ], batch size: 81, lr: 4.74e-02, grad_scale: 16.0 2023-04-16 12:09:25,283 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.7313, 4.8549, 4.4959, 4.4887, 4.6031, 4.5146, 4.7932, 4.8534], device='cuda:1'), covar=tensor([0.0225, 0.0188, 0.0447, 0.0313, 0.0330, 0.0303, 0.0267, 0.0236], device='cuda:1'), in_proj_covar=tensor([0.0059, 0.0064, 0.0054, 0.0074, 0.0057, 0.0063, 0.0056, 0.0053], device='cuda:1'), out_proj_covar=tensor([6.2705e-05, 6.4278e-05, 5.4196e-05, 8.2498e-05, 6.0440e-05, 6.6046e-05, 5.6906e-05, 5.3831e-05], device='cuda:1') 2023-04-16 12:09:41,254 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2494.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 12:09:44,260 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6748, 4.2467, 3.7864, 4.0227, 4.2292, 4.4629, 4.1323, 4.6478], device='cuda:1'), covar=tensor([0.0442, 0.0281, 0.0292, 0.0287, 0.0189, 0.0237, 0.0311, 0.0128], device='cuda:1'), in_proj_covar=tensor([0.0048, 0.0040, 0.0039, 0.0048, 0.0038, 0.0042, 0.0037, 0.0037], device='cuda:1'), out_proj_covar=tensor([5.3446e-05, 4.4741e-05, 4.3833e-05, 5.5472e-05, 4.4391e-05, 4.6973e-05, 3.8805e-05, 3.6387e-05], device='cuda:1') 2023-04-16 12:09:47,165 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.2705, 4.6388, 4.6760, 4.8981, 4.6764, 4.7333, 5.0420, 4.7827], device='cuda:1'), covar=tensor([0.0726, 0.0745, 0.1481, 0.1303, 0.0637, 0.0742, 0.0919, 0.0555], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0114, 0.0122, 0.0127, 0.0085, 0.0121, 0.0123, 0.0098], device='cuda:1'), out_proj_covar=tensor([1.3988e-04, 1.1550e-04, 1.3037e-04, 1.3447e-04, 9.0540e-05, 1.2325e-04, 1.4141e-04, 9.6591e-05], device='cuda:1') 2023-04-16 12:09:47,709 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.314e+02 3.928e+02 4.819e+02 6.017e+02 8.631e+02, threshold=9.637e+02, percent-clipped=0.0 2023-04-16 12:09:47,734 INFO [train.py:893] (1/4) Epoch 1, batch 2500, loss[loss=0.4112, simple_loss=0.4019, pruned_loss=0.2102, over 13532.00 frames. ], tot_loss[loss=0.4062, simple_loss=0.3905, pruned_loss=0.2112, over 2661190.85 frames. ], batch size: 85, lr: 4.73e-02, grad_scale: 16.0 2023-04-16 12:09:57,654 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2513.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:10:08,471 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2526.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:10:23,017 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2542.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:10:30,390 INFO [train.py:893] (1/4) Epoch 1, batch 2550, loss[loss=0.3752, simple_loss=0.3768, pruned_loss=0.1868, over 13523.00 frames. ], tot_loss[loss=0.4054, simple_loss=0.3906, pruned_loss=0.2103, over 2663540.56 frames. ], batch size: 76, lr: 4.72e-02, grad_scale: 16.0 2023-04-16 12:10:39,737 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2561.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:10:41,573 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0 2023-04-16 12:10:48,961 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.08 vs. limit=2.0 2023-04-16 12:10:50,250 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=2574.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:10:50,577 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.90 vs. limit=2.0 2023-04-16 12:10:52,504 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-16 12:11:13,923 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.061e+02 4.539e+02 5.406e+02 6.708e+02 1.265e+03, threshold=1.081e+03, percent-clipped=3.0 2023-04-16 12:11:13,958 INFO [train.py:893] (1/4) Epoch 1, batch 2600, loss[loss=0.3974, simple_loss=0.3858, pruned_loss=0.2045, over 13539.00 frames. ], tot_loss[loss=0.4034, simple_loss=0.3889, pruned_loss=0.2091, over 2649344.57 frames. ], batch size: 85, lr: 4.71e-02, grad_scale: 16.0 2023-04-16 12:11:15,250 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 2023-04-16 12:11:53,079 INFO [train.py:893] (1/4) Epoch 1, batch 2650, loss[loss=0.3809, simple_loss=0.3765, pruned_loss=0.1926, over 13532.00 frames. ], tot_loss[loss=0.4026, simple_loss=0.3884, pruned_loss=0.2086, over 2645365.04 frames. ], batch size: 85, lr: 4.70e-02, grad_scale: 16.0 2023-04-16 12:12:17,947 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2685.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 12:12:25,466 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=2696.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 12:12:47,698 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 12:12:56,999 INFO [train.py:893] (1/4) Epoch 2, batch 0, loss[loss=0.4004, simple_loss=0.3872, pruned_loss=0.2068, over 13482.00 frames. ], tot_loss[loss=0.4004, simple_loss=0.3872, pruned_loss=0.2068, over 13482.00 frames. ], batch size: 81, lr: 4.60e-02, grad_scale: 16.0 2023-04-16 12:12:57,000 INFO [train.py:918] (1/4) Computing validation loss 2023-04-16 12:13:13,088 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.2109, 3.1823, 3.1884, 3.1246, 2.3787, 2.9312, 3.1639, 1.4206], device='cuda:1'), covar=tensor([0.0459, 0.1639, 0.0678, 0.0638, 0.3284, 0.0765, 0.1070, 0.6380], device='cuda:1'), in_proj_covar=tensor([0.0025, 0.0032, 0.0026, 0.0020, 0.0048, 0.0034, 0.0030, 0.0066], device='cuda:1'), out_proj_covar=tensor([1.9878e-05, 3.0872e-05, 2.2125e-05, 2.0432e-05, 4.5589e-05, 3.0185e-05, 2.9430e-05, 6.1212e-05], device='cuda:1') 2023-04-16 12:13:19,793 INFO [train.py:927] (1/4) Epoch 2, validation: loss=0.3243, simple_loss=0.342, pruned_loss=0.1533, over 2446609.00 frames. 2023-04-16 12:13:19,794 INFO [train.py:928] (1/4) Maximum memory allocated so far is 11934MB 2023-04-16 12:13:20,441 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.595e+02 4.408e+02 5.707e+02 7.369e+02 1.306e+03, threshold=1.141e+03, percent-clipped=2.0 2023-04-16 12:14:00,172 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2746.0, num_to_drop=2, layers_to_drop={1, 3} 2023-04-16 12:14:02,998 INFO [train.py:893] (1/4) Epoch 2, batch 50, loss[loss=0.3613, simple_loss=0.3574, pruned_loss=0.1826, over 13501.00 frames. ], tot_loss[loss=0.3842, simple_loss=0.3737, pruned_loss=0.1973, over 603353.53 frames. ], batch size: 76, lr: 4.59e-02, grad_scale: 16.0 2023-04-16 12:14:27,336 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-16 12:14:27,337 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-16 12:14:27,337 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-16 12:14:27,343 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-16 12:14:27,357 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-16 12:14:28,016 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-16 12:14:28,027 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-16 12:14:47,272 INFO [train.py:893] (1/4) Epoch 2, batch 100, loss[loss=0.3289, simple_loss=0.2998, pruned_loss=0.179, over 7829.00 frames. ], tot_loss[loss=0.3907, simple_loss=0.3786, pruned_loss=0.2014, over 1057988.04 frames. ], batch size: 29, lr: 4.58e-02, grad_scale: 16.0 2023-04-16 12:14:48,356 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.351e+02 4.141e+02 5.422e+02 6.663e+02 1.362e+03, threshold=1.084e+03, percent-clipped=1.0 2023-04-16 12:15:08,925 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2824.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 12:15:31,829 INFO [train.py:893] (1/4) Epoch 2, batch 150, loss[loss=0.392, simple_loss=0.3716, pruned_loss=0.2062, over 13374.00 frames. ], tot_loss[loss=0.3927, simple_loss=0.381, pruned_loss=0.2022, over 1398831.95 frames. ], batch size: 67, lr: 4.57e-02, grad_scale: 16.0 2023-04-16 12:15:40,572 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7731, 3.8089, 3.4751, 3.0095, 1.8495, 2.7025, 3.9063, 4.0615], device='cuda:1'), covar=tensor([0.0302, 0.0756, 0.1307, 0.2733, 0.3695, 0.2039, 0.0417, 0.0267], device='cuda:1'), in_proj_covar=tensor([0.0030, 0.0029, 0.0031, 0.0053, 0.0060, 0.0048, 0.0024, 0.0021], device='cuda:1'), out_proj_covar=tensor([2.4162e-05, 2.5251e-05, 2.8532e-05, 4.9059e-05, 5.3009e-05, 4.3204e-05, 2.1626e-05, 1.8221e-05], device='cuda:1') 2023-04-16 12:15:57,410 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5828, 3.8702, 3.4925, 3.9323, 4.0910, 4.5165, 4.0038, 4.5413], device='cuda:1'), covar=tensor([0.0418, 0.0459, 0.0339, 0.0372, 0.0235, 0.0304, 0.0321, 0.0114], device='cuda:1'), in_proj_covar=tensor([0.0046, 0.0041, 0.0041, 0.0050, 0.0038, 0.0044, 0.0038, 0.0039], device='cuda:1'), out_proj_covar=tensor([5.6445e-05, 4.9812e-05, 5.2697e-05, 6.6274e-05, 5.1999e-05, 5.4644e-05, 4.4850e-05, 4.4013e-05], device='cuda:1') 2023-04-16 12:16:02,150 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2885.0, num_to_drop=2, layers_to_drop={0, 1} 2023-04-16 12:16:14,801 INFO [train.py:893] (1/4) Epoch 2, batch 200, loss[loss=0.4132, simple_loss=0.4059, pruned_loss=0.2102, over 13523.00 frames. ], tot_loss[loss=0.396, simple_loss=0.384, pruned_loss=0.204, over 1680263.97 frames. ], batch size: 85, lr: 4.56e-02, grad_scale: 16.0 2023-04-16 12:16:15,563 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.311e+02 4.527e+02 5.410e+02 6.609e+02 1.468e+03, threshold=1.082e+03, percent-clipped=2.0 2023-04-16 12:16:41,458 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0515, 3.2122, 2.8698, 3.5438, 3.7098, 3.0028, 3.1668, 2.7895], device='cuda:1'), covar=tensor([0.0283, 0.0265, 0.0265, 0.0137, 0.0115, 0.0283, 0.0245, 0.0400], device='cuda:1'), in_proj_covar=tensor([0.0033, 0.0032, 0.0030, 0.0029, 0.0028, 0.0026, 0.0034, 0.0036], device='cuda:1'), out_proj_covar=tensor([3.2318e-05, 3.0095e-05, 2.8359e-05, 2.6763e-05, 2.6694e-05, 2.6149e-05, 3.2058e-05, 3.3496e-05], device='cuda:1') 2023-04-16 12:16:47,627 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=2937.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 12:16:58,404 INFO [train.py:893] (1/4) Epoch 2, batch 250, loss[loss=0.4027, simple_loss=0.3904, pruned_loss=0.2076, over 13534.00 frames. ], tot_loss[loss=0.3944, simple_loss=0.3839, pruned_loss=0.2024, over 1901210.65 frames. ], batch size: 85, lr: 4.55e-02, grad_scale: 16.0 2023-04-16 12:17:02,149 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.07 vs. limit=2.0 2023-04-16 12:17:28,167 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0288, 2.3688, 1.9787, 2.3599, 2.2910, 2.0964, 2.5033, 1.9345], device='cuda:1'), covar=tensor([0.0266, 0.0473, 0.0278, 0.0287, 0.0374, 0.0237, 0.0263, 0.0561], device='cuda:1'), in_proj_covar=tensor([0.0017, 0.0016, 0.0019, 0.0018, 0.0020, 0.0017, 0.0019, 0.0019], device='cuda:1'), out_proj_covar=tensor([1.5389e-05, 1.4163e-05, 1.6629e-05, 1.6319e-05, 1.8206e-05, 1.5325e-05, 1.7360e-05, 1.7440e-05], device='cuda:1') 2023-04-16 12:17:38,738 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=2996.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 12:17:41,018 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=2998.0, num_to_drop=2, layers_to_drop={1, 2} 2023-04-16 12:17:42,240 INFO [train.py:893] (1/4) Epoch 2, batch 300, loss[loss=0.3719, simple_loss=0.373, pruned_loss=0.1854, over 13514.00 frames. ], tot_loss[loss=0.3914, simple_loss=0.3829, pruned_loss=0.2, over 2073079.78 frames. ], batch size: 70, lr: 4.54e-02, grad_scale: 16.0 2023-04-16 12:17:43,330 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.732e+02 4.238e+02 5.185e+02 6.684e+02 1.179e+03, threshold=1.037e+03, percent-clipped=2.0 2023-04-16 12:17:44,594 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.27 vs. limit=2.0 2023-04-16 12:17:59,161 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3019.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 12:18:09,250 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5979, 4.0623, 4.2291, 3.3931, 4.2912, 2.4747, 4.3476, 2.8985], device='cuda:1'), covar=tensor([0.1872, 0.1135, 0.0468, 0.1458, 0.0317, 0.4631, 0.0458, 0.3895], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0047, 0.0049, 0.0069, 0.0046, 0.0091, 0.0039, 0.0078], device='cuda:1'), out_proj_covar=tensor([7.5546e-05, 5.3111e-05, 5.0144e-05, 7.2173e-05, 4.6154e-05, 9.4262e-05, 4.2177e-05, 8.5093e-05], device='cuda:1') 2023-04-16 12:18:17,121 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0364, 2.3217, 2.3979, 2.3278, 1.9620, 1.5985, 2.0913, 2.2556], device='cuda:1'), covar=tensor([0.0136, 0.0370, 0.0396, 0.0320, 0.0198, 0.0373, 0.0563, 0.0368], device='cuda:1'), in_proj_covar=tensor([0.0036, 0.0038, 0.0041, 0.0034, 0.0032, 0.0030, 0.0036, 0.0035], device='cuda:1'), out_proj_covar=tensor([3.1569e-05, 3.6398e-05, 3.8038e-05, 2.9567e-05, 2.9012e-05, 3.4322e-05, 3.3304e-05, 3.0405e-05], device='cuda:1') 2023-04-16 12:18:18,472 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3041.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:18:20,928 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3044.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 12:18:25,570 INFO [train.py:893] (1/4) Epoch 2, batch 350, loss[loss=0.4267, simple_loss=0.4116, pruned_loss=0.2208, over 11860.00 frames. ], tot_loss[loss=0.3918, simple_loss=0.3833, pruned_loss=0.2001, over 2204825.17 frames. ], batch size: 157, lr: 4.53e-02, grad_scale: 16.0 2023-04-16 12:18:25,800 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.3926, 4.7434, 4.8717, 4.9846, 4.7975, 4.6995, 5.1084, 4.7931], device='cuda:1'), covar=tensor([0.0606, 0.0691, 0.1293, 0.1428, 0.0493, 0.0834, 0.0861, 0.0673], device='cuda:1'), in_proj_covar=tensor([0.0138, 0.0127, 0.0142, 0.0150, 0.0095, 0.0132, 0.0147, 0.0113], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-16 12:18:52,585 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3080.0, num_to_drop=2, layers_to_drop={1, 3} 2023-04-16 12:19:10,159 INFO [train.py:893] (1/4) Epoch 2, batch 400, loss[loss=0.4308, simple_loss=0.4172, pruned_loss=0.2222, over 13250.00 frames. ], tot_loss[loss=0.3881, simple_loss=0.3811, pruned_loss=0.1975, over 2306063.55 frames. ], batch size: 124, lr: 4.52e-02, grad_scale: 16.0 2023-04-16 12:19:10,828 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.297e+02 4.068e+02 5.091e+02 6.517e+02 1.067e+03, threshold=1.018e+03, percent-clipped=1.0 2023-04-16 12:19:12,011 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.8712, 2.1612, 2.1206, 2.2372, 1.5570, 1.7874, 2.0225, 2.2538], device='cuda:1'), covar=tensor([0.0290, 0.0478, 0.0672, 0.0161, 0.0253, 0.0298, 0.0524, 0.0328], device='cuda:1'), in_proj_covar=tensor([0.0033, 0.0035, 0.0038, 0.0032, 0.0030, 0.0027, 0.0033, 0.0032], device='cuda:1'), out_proj_covar=tensor([2.8955e-05, 3.3193e-05, 3.6105e-05, 2.7313e-05, 2.7842e-05, 3.1122e-05, 2.9963e-05, 2.8321e-05], device='cuda:1') 2023-04-16 12:19:54,900 INFO [train.py:893] (1/4) Epoch 2, batch 450, loss[loss=0.4986, simple_loss=0.4629, pruned_loss=0.2671, over 13440.00 frames. ], tot_loss[loss=0.3902, simple_loss=0.383, pruned_loss=0.1987, over 2385244.47 frames. ], batch size: 106, lr: 4.51e-02, grad_scale: 16.0 2023-04-16 12:20:13,871 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.10 vs. limit=2.0 2023-04-16 12:20:18,776 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-16 12:20:21,147 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3180.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:20:29,268 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 2023-04-16 12:20:38,210 INFO [train.py:893] (1/4) Epoch 2, batch 500, loss[loss=0.4207, simple_loss=0.4133, pruned_loss=0.214, over 13382.00 frames. ], tot_loss[loss=0.39, simple_loss=0.3834, pruned_loss=0.1983, over 2446083.25 frames. ], batch size: 109, lr: 4.50e-02, grad_scale: 16.0 2023-04-16 12:20:39,286 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.435e+02 4.634e+02 5.605e+02 7.324e+02 2.141e+03, threshold=1.121e+03, percent-clipped=6.0 2023-04-16 12:20:44,831 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7697, 2.8963, 3.5928, 2.5476, 3.8597, 3.2892, 3.9765, 3.3520], device='cuda:1'), covar=tensor([0.0122, 0.0237, 0.0187, 0.0243, 0.0174, 0.0221, 0.0160, 0.0170], device='cuda:1'), in_proj_covar=tensor([0.0023, 0.0023, 0.0020, 0.0025, 0.0017, 0.0022, 0.0019, 0.0020], device='cuda:1'), out_proj_covar=tensor([2.7554e-05, 2.8442e-05, 2.3773e-05, 3.2399e-05, 2.1412e-05, 2.7089e-05, 2.2326e-05, 2.5166e-05], device='cuda:1') 2023-04-16 12:21:22,353 INFO [train.py:893] (1/4) Epoch 2, batch 550, loss[loss=0.3767, simple_loss=0.3802, pruned_loss=0.1866, over 13433.00 frames. ], tot_loss[loss=0.3877, simple_loss=0.3822, pruned_loss=0.1966, over 2494903.04 frames. ], batch size: 95, lr: 4.49e-02, grad_scale: 16.0 2023-04-16 12:21:42,618 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 2023-04-16 12:21:50,135 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5279, 3.2889, 3.6834, 2.7625, 3.9374, 3.5621, 4.0249, 3.6175], device='cuda:1'), covar=tensor([0.0165, 0.0203, 0.0174, 0.0222, 0.0143, 0.0167, 0.0147, 0.0144], device='cuda:1'), in_proj_covar=tensor([0.0024, 0.0024, 0.0021, 0.0025, 0.0018, 0.0022, 0.0020, 0.0020], device='cuda:1'), out_proj_covar=tensor([2.9495e-05, 3.0430e-05, 2.5190e-05, 3.3700e-05, 2.2504e-05, 2.8934e-05, 2.3320e-05, 2.6199e-05], device='cuda:1') 2023-04-16 12:21:55,505 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3288.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 12:22:00,159 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3293.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:22:05,324 INFO [train.py:893] (1/4) Epoch 2, batch 600, loss[loss=0.4156, simple_loss=0.3983, pruned_loss=0.2164, over 13532.00 frames. ], tot_loss[loss=0.3843, simple_loss=0.3792, pruned_loss=0.1947, over 2533044.95 frames. ], batch size: 76, lr: 4.48e-02, grad_scale: 16.0 2023-04-16 12:22:06,085 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.737e+02 4.280e+02 5.504e+02 6.489e+02 1.365e+03, threshold=1.101e+03, percent-clipped=1.0 2023-04-16 12:22:23,401 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 2023-04-16 12:22:42,639 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3341.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:22:49,854 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3349.0, num_to_drop=2, layers_to_drop={1, 3} 2023-04-16 12:22:50,296 INFO [train.py:893] (1/4) Epoch 2, batch 650, loss[loss=0.3487, simple_loss=0.3595, pruned_loss=0.1689, over 13411.00 frames. ], tot_loss[loss=0.38, simple_loss=0.3759, pruned_loss=0.1921, over 2558562.94 frames. ], batch size: 113, lr: 4.47e-02, grad_scale: 16.0 2023-04-16 12:22:56,964 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 2023-04-16 12:23:12,817 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3375.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:23:24,523 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3389.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:23:34,088 INFO [train.py:893] (1/4) Epoch 2, batch 700, loss[loss=0.3791, simple_loss=0.3799, pruned_loss=0.1891, over 13408.00 frames. ], tot_loss[loss=0.3765, simple_loss=0.3737, pruned_loss=0.1897, over 2575610.53 frames. ], batch size: 113, lr: 4.46e-02, grad_scale: 16.0 2023-04-16 12:23:35,207 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.882e+02 4.256e+02 4.913e+02 6.838e+02 1.348e+03, threshold=9.827e+02, percent-clipped=2.0 2023-04-16 12:24:11,606 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-04-16 12:24:18,237 INFO [train.py:893] (1/4) Epoch 2, batch 750, loss[loss=0.3895, simple_loss=0.3952, pruned_loss=0.1918, over 13425.00 frames. ], tot_loss[loss=0.3762, simple_loss=0.3727, pruned_loss=0.1899, over 2593644.04 frames. ], batch size: 95, lr: 4.45e-02, grad_scale: 16.0 2023-04-16 12:24:45,329 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3480.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:24:48,450 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.3377, 2.0772, 1.8631, 2.2484, 1.9105, 1.4547, 1.5863, 2.3197], device='cuda:1'), covar=tensor([0.0164, 0.0524, 0.1326, 0.0107, 0.0129, 0.0451, 0.0722, 0.0360], device='cuda:1'), in_proj_covar=tensor([0.0026, 0.0029, 0.0033, 0.0025, 0.0025, 0.0027, 0.0028, 0.0026], device='cuda:1'), out_proj_covar=tensor([2.4309e-05, 2.7791e-05, 3.1758e-05, 2.0586e-05, 2.3221e-05, 3.0312e-05, 2.6551e-05, 2.3615e-05], device='cuda:1') 2023-04-16 12:25:01,844 INFO [train.py:893] (1/4) Epoch 2, batch 800, loss[loss=0.3722, simple_loss=0.3775, pruned_loss=0.1835, over 13457.00 frames. ], tot_loss[loss=0.3791, simple_loss=0.376, pruned_loss=0.1911, over 2613252.92 frames. ], batch size: 79, lr: 4.44e-02, grad_scale: 16.0 2023-04-16 12:25:02,610 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.861e+02 5.123e+02 6.281e+02 8.038e+02 1.655e+03, threshold=1.256e+03, percent-clipped=13.0 2023-04-16 12:25:27,271 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3528.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:25:29,803 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4973, 3.4026, 3.1631, 3.3700, 2.4495, 2.8382, 3.1819, 1.7374], device='cuda:1'), covar=tensor([0.0209, 0.0851, 0.0199, 0.0250, 0.1977, 0.0590, 0.0846, 0.3166], device='cuda:1'), in_proj_covar=tensor([0.0035, 0.0044, 0.0036, 0.0028, 0.0078, 0.0048, 0.0045, 0.0093], device='cuda:1'), out_proj_covar=tensor([3.0129e-05, 4.6645e-05, 3.3268e-05, 2.9319e-05, 7.7017e-05, 4.4104e-05, 4.5452e-05, 8.9550e-05], device='cuda:1') 2023-04-16 12:25:41,664 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 2023-04-16 12:25:43,005 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3546.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 12:25:45,908 INFO [train.py:893] (1/4) Epoch 2, batch 850, loss[loss=0.4033, simple_loss=0.3906, pruned_loss=0.208, over 13444.00 frames. ], tot_loss[loss=0.3806, simple_loss=0.3776, pruned_loss=0.1919, over 2626393.84 frames. ], batch size: 95, lr: 4.43e-02, grad_scale: 16.0 2023-04-16 12:25:49,969 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2821, 4.5104, 4.3487, 4.1388, 4.1674, 3.9855, 4.4362, 4.5881], device='cuda:1'), covar=tensor([0.0228, 0.0210, 0.0247, 0.0410, 0.0432, 0.0331, 0.0254, 0.0220], device='cuda:1'), in_proj_covar=tensor([0.0068, 0.0078, 0.0061, 0.0081, 0.0064, 0.0072, 0.0063, 0.0061], device='cuda:1'), out_proj_covar=tensor([8.5399e-05, 9.6506e-05, 7.8064e-05, 1.0953e-04, 8.8699e-05, 9.3890e-05, 8.1784e-05, 7.7020e-05], device='cuda:1') 2023-04-16 12:26:25,107 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3593.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:26:31,155 INFO [train.py:893] (1/4) Epoch 2, batch 900, loss[loss=0.3554, simple_loss=0.3542, pruned_loss=0.1784, over 13416.00 frames. ], tot_loss[loss=0.3787, simple_loss=0.3759, pruned_loss=0.1908, over 2636974.60 frames. ], batch size: 65, lr: 4.42e-02, grad_scale: 16.0 2023-04-16 12:26:32,230 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.963e+02 5.281e+02 6.987e+02 9.717e+02 1.667e+03, threshold=1.397e+03, percent-clipped=8.0 2023-04-16 12:26:38,202 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3607.0, num_to_drop=2, layers_to_drop={1, 3} 2023-04-16 12:27:01,646 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-16 12:27:08,105 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3641.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:27:10,644 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3644.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 12:27:12,881 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0225, 3.0573, 3.4228, 3.5606, 3.8146, 3.2554, 3.6371, 3.7041], device='cuda:1'), covar=tensor([0.0234, 0.0120, 0.0235, 0.0376, 0.0153, 0.0192, 0.0160, 0.0220], device='cuda:1'), in_proj_covar=tensor([0.0038, 0.0024, 0.0033, 0.0040, 0.0035, 0.0034, 0.0035, 0.0029], device='cuda:1'), out_proj_covar=tensor([5.8557e-05, 4.3099e-05, 5.3219e-05, 6.2653e-05, 5.6358e-05, 5.2476e-05, 5.1571e-05, 4.8279e-05], device='cuda:1') 2023-04-16 12:27:15,900 INFO [train.py:893] (1/4) Epoch 2, batch 950, loss[loss=0.3632, simple_loss=0.3613, pruned_loss=0.1826, over 13210.00 frames. ], tot_loss[loss=0.3765, simple_loss=0.3732, pruned_loss=0.1899, over 2646555.09 frames. ], batch size: 132, lr: 4.41e-02, grad_scale: 16.0 2023-04-16 12:27:18,661 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8712, 3.5679, 4.0698, 3.1858, 4.2422, 3.5579, 4.2097, 3.7009], device='cuda:1'), covar=tensor([0.0152, 0.0163, 0.0117, 0.0156, 0.0157, 0.0166, 0.0155, 0.0131], device='cuda:1'), in_proj_covar=tensor([0.0024, 0.0026, 0.0022, 0.0025, 0.0019, 0.0023, 0.0019, 0.0021], device='cuda:1'), out_proj_covar=tensor([3.3028e-05, 3.9031e-05, 3.0551e-05, 3.8663e-05, 2.7711e-05, 3.3326e-05, 2.6534e-05, 3.1117e-05], device='cuda:1') 2023-04-16 12:27:19,449 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1290, 4.2942, 3.9240, 4.1308, 3.8929, 3.7585, 4.2535, 4.3228], device='cuda:1'), covar=tensor([0.0200, 0.0199, 0.0435, 0.0255, 0.0372, 0.0331, 0.0282, 0.0181], device='cuda:1'), in_proj_covar=tensor([0.0069, 0.0080, 0.0065, 0.0083, 0.0065, 0.0075, 0.0064, 0.0062], device='cuda:1'), out_proj_covar=tensor([8.8260e-05, 1.0071e-04, 8.6121e-05, 1.1372e-04, 9.2338e-05, 9.8857e-05, 8.6997e-05, 7.8178e-05], device='cuda:1') 2023-04-16 12:27:38,894 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3675.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:27:40,436 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3676.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 12:28:02,052 INFO [train.py:893] (1/4) Epoch 2, batch 1000, loss[loss=0.3465, simple_loss=0.3529, pruned_loss=0.1701, over 13503.00 frames. ], tot_loss[loss=0.3713, simple_loss=0.3697, pruned_loss=0.1865, over 2650381.68 frames. ], batch size: 91, lr: 4.40e-02, grad_scale: 16.0 2023-04-16 12:28:02,797 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.113e+02 4.777e+02 5.956e+02 7.807e+02 2.216e+03, threshold=1.191e+03, percent-clipped=1.0 2023-04-16 12:28:16,235 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7383, 1.8562, 3.2063, 2.8379, 2.8423, 2.3857, 2.4876, 2.4915], device='cuda:1'), covar=tensor([0.0817, 0.1727, 0.0204, 0.0133, 0.0126, 0.0221, 0.0181, 0.0424], device='cuda:1'), in_proj_covar=tensor([0.0049, 0.0081, 0.0040, 0.0034, 0.0033, 0.0039, 0.0032, 0.0042], device='cuda:1'), out_proj_covar=tensor([4.3615e-05, 8.1536e-05, 3.5694e-05, 2.9261e-05, 2.7953e-05, 3.4460e-05, 2.8958e-05, 3.7693e-05], device='cuda:1') 2023-04-16 12:28:23,451 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3723.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:28:35,864 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3737.0, num_to_drop=2, layers_to_drop={2, 3} 2023-04-16 12:28:43,145 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.4506, 2.2246, 1.8235, 1.7629, 1.5346, 1.8053, 1.5986, 2.1766], device='cuda:1'), covar=tensor([0.0459, 0.0443, 0.1517, 0.0269, 0.0242, 0.0208, 0.0554, 0.0322], device='cuda:1'), in_proj_covar=tensor([0.0035, 0.0033, 0.0034, 0.0034, 0.0025, 0.0030, 0.0041, 0.0036], device='cuda:1'), out_proj_covar=tensor([3.2426e-05, 3.3490e-05, 3.4921e-05, 2.8753e-05, 2.9284e-05, 2.4009e-05, 3.7730e-05, 3.1430e-05], device='cuda:1') 2023-04-16 12:28:47,101 INFO [train.py:893] (1/4) Epoch 2, batch 1050, loss[loss=0.3534, simple_loss=0.3638, pruned_loss=0.1715, over 13378.00 frames. ], tot_loss[loss=0.3653, simple_loss=0.3659, pruned_loss=0.1823, over 2654720.12 frames. ], batch size: 113, lr: 4.39e-02, grad_scale: 16.0 2023-04-16 12:29:02,790 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 2023-04-16 12:29:13,928 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.84 vs. limit=2.0 2023-04-16 12:29:23,814 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8278, 3.4043, 3.6497, 2.9949, 4.1144, 3.4511, 4.1548, 3.6440], device='cuda:1'), covar=tensor([0.0157, 0.0199, 0.0305, 0.0156, 0.0220, 0.0201, 0.0235, 0.0148], device='cuda:1'), in_proj_covar=tensor([0.0025, 0.0026, 0.0023, 0.0026, 0.0019, 0.0024, 0.0019, 0.0021], device='cuda:1'), out_proj_covar=tensor([3.5874e-05, 3.9838e-05, 3.2914e-05, 4.0184e-05, 2.9031e-05, 3.6762e-05, 2.7577e-05, 3.1812e-05], device='cuda:1') 2023-04-16 12:29:32,636 INFO [train.py:893] (1/4) Epoch 2, batch 1100, loss[loss=0.3362, simple_loss=0.3483, pruned_loss=0.162, over 13529.00 frames. ], tot_loss[loss=0.3637, simple_loss=0.3651, pruned_loss=0.1812, over 2655403.80 frames. ], batch size: 70, lr: 4.37e-02, grad_scale: 16.0 2023-04-16 12:29:33,668 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.108e+02 5.042e+02 6.518e+02 8.032e+02 1.563e+03, threshold=1.304e+03, percent-clipped=1.0 2023-04-16 12:30:11,855 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7066, 4.2204, 3.7133, 3.1614, 2.2687, 3.2525, 4.0247, 4.1851], device='cuda:1'), covar=tensor([0.0265, 0.0367, 0.0531, 0.1320, 0.1551, 0.0864, 0.0171, 0.0209], device='cuda:1'), in_proj_covar=tensor([0.0036, 0.0031, 0.0035, 0.0063, 0.0065, 0.0051, 0.0027, 0.0025], device='cuda:1'), out_proj_covar=tensor([3.2398e-05, 2.9542e-05, 3.3394e-05, 5.9854e-05, 6.0183e-05, 4.8508e-05, 2.4529e-05, 2.2009e-05], device='cuda:1') 2023-04-16 12:30:18,727 INFO [train.py:893] (1/4) Epoch 2, batch 1150, loss[loss=0.3932, simple_loss=0.3932, pruned_loss=0.1966, over 13428.00 frames. ], tot_loss[loss=0.361, simple_loss=0.3641, pruned_loss=0.1789, over 2659309.05 frames. ], batch size: 95, lr: 4.36e-02, grad_scale: 16.0 2023-04-16 12:30:35,973 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0920, 1.6105, 3.3100, 3.0607, 2.8988, 2.7963, 2.8117, 2.2683], device='cuda:1'), covar=tensor([0.0714, 0.1897, 0.0169, 0.0109, 0.0096, 0.0175, 0.0151, 0.0505], device='cuda:1'), in_proj_covar=tensor([0.0056, 0.0090, 0.0042, 0.0039, 0.0037, 0.0045, 0.0034, 0.0048], device='cuda:1'), out_proj_covar=tensor([5.0628e-05, 9.1076e-05, 3.7426e-05, 3.3362e-05, 3.1744e-05, 4.0054e-05, 3.0149e-05, 4.2882e-05], device='cuda:1') 2023-04-16 12:30:45,686 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.1529, 2.0843, 1.8426, 1.8950, 2.1100, 2.3813, 2.2738, 1.7318], device='cuda:1'), covar=tensor([0.0179, 0.0741, 0.0243, 0.0227, 0.0316, 0.0152, 0.0370, 0.0540], device='cuda:1'), in_proj_covar=tensor([0.0024, 0.0023, 0.0027, 0.0023, 0.0031, 0.0023, 0.0024, 0.0034], device='cuda:1'), out_proj_covar=tensor([2.5890e-05, 2.5771e-05, 2.8103e-05, 2.3369e-05, 3.4262e-05, 2.3754e-05, 2.6107e-05, 3.6353e-05], device='cuda:1') 2023-04-16 12:31:04,124 INFO [train.py:893] (1/4) Epoch 2, batch 1200, loss[loss=0.4112, simple_loss=0.4047, pruned_loss=0.2088, over 13489.00 frames. ], tot_loss[loss=0.3609, simple_loss=0.3643, pruned_loss=0.1787, over 2658851.10 frames. ], batch size: 93, lr: 4.35e-02, grad_scale: 16.0 2023-04-16 12:31:04,948 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.864e+02 4.882e+02 5.871e+02 7.581e+02 1.368e+03, threshold=1.174e+03, percent-clipped=2.0 2023-04-16 12:31:05,877 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=3902.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 12:31:27,898 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=3926.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:31:30,989 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-16 12:31:43,953 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-16 12:31:45,048 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=3944.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 12:31:50,590 INFO [train.py:893] (1/4) Epoch 2, batch 1250, loss[loss=0.3722, simple_loss=0.3814, pruned_loss=0.1814, over 13421.00 frames. ], tot_loss[loss=0.3616, simple_loss=0.3644, pruned_loss=0.1794, over 2649609.51 frames. ], batch size: 103, lr: 4.34e-02, grad_scale: 16.0 2023-04-16 12:32:24,491 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=3987.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:32:29,290 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=3992.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 12:32:33,596 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-04-16 12:32:35,659 INFO [train.py:893] (1/4) Epoch 2, batch 1300, loss[loss=0.403, simple_loss=0.3866, pruned_loss=0.2097, over 11815.00 frames. ], tot_loss[loss=0.3623, simple_loss=0.3651, pruned_loss=0.1797, over 2651824.29 frames. ], batch size: 158, lr: 4.33e-02, grad_scale: 32.0 2023-04-16 12:32:40,628 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.644e+02 4.834e+02 5.973e+02 8.685e+02 1.398e+03, threshold=1.195e+03, percent-clipped=5.0 2023-04-16 12:33:07,041 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.09 vs. limit=2.0 2023-04-16 12:33:10,695 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4032.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 12:33:27,001 INFO [train.py:893] (1/4) Epoch 2, batch 1350, loss[loss=0.3585, simple_loss=0.368, pruned_loss=0.1746, over 13391.00 frames. ], tot_loss[loss=0.3643, simple_loss=0.3666, pruned_loss=0.181, over 2652598.26 frames. ], batch size: 109, lr: 4.32e-02, grad_scale: 32.0 2023-04-16 12:33:33,101 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9238, 4.0421, 4.2480, 3.5557, 4.2819, 2.3698, 4.4304, 2.7359], device='cuda:1'), covar=tensor([0.0825, 0.0437, 0.0268, 0.0708, 0.0238, 0.2230, 0.0194, 0.2356], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0041, 0.0043, 0.0057, 0.0044, 0.0067, 0.0032, 0.0068], device='cuda:1'), out_proj_covar=tensor([6.9811e-05, 4.6318e-05, 4.5431e-05, 6.2901e-05, 4.5959e-05, 7.2846e-05, 3.5037e-05, 7.8696e-05], device='cuda:1') 2023-04-16 12:34:13,917 INFO [train.py:893] (1/4) Epoch 2, batch 1400, loss[loss=0.3637, simple_loss=0.3669, pruned_loss=0.1802, over 13373.00 frames. ], tot_loss[loss=0.3613, simple_loss=0.3644, pruned_loss=0.1791, over 2649452.62 frames. ], batch size: 109, lr: 4.31e-02, grad_scale: 32.0 2023-04-16 12:34:14,700 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.422e+02 4.905e+02 6.254e+02 7.280e+02 1.526e+03, threshold=1.251e+03, percent-clipped=2.0 2023-04-16 12:34:44,045 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 2023-04-16 12:34:51,934 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4141.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:34:52,645 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.0911, 4.5332, 4.5518, 4.4562, 4.2183, 4.5330, 4.8265, 4.5054], device='cuda:1'), covar=tensor([0.0638, 0.0780, 0.1504, 0.1922, 0.0785, 0.0956, 0.0989, 0.0684], device='cuda:1'), in_proj_covar=tensor([0.0146, 0.0139, 0.0180, 0.0186, 0.0108, 0.0166, 0.0186, 0.0125], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001], device='cuda:1') 2023-04-16 12:34:59,847 INFO [train.py:893] (1/4) Epoch 2, batch 1450, loss[loss=0.3809, simple_loss=0.3846, pruned_loss=0.1887, over 13377.00 frames. ], tot_loss[loss=0.3591, simple_loss=0.3626, pruned_loss=0.1778, over 2653588.59 frames. ], batch size: 109, lr: 4.30e-02, grad_scale: 32.0 2023-04-16 12:35:13,482 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.10 vs. limit=5.0 2023-04-16 12:35:47,015 INFO [train.py:893] (1/4) Epoch 2, batch 1500, loss[loss=0.3588, simple_loss=0.3673, pruned_loss=0.1752, over 13539.00 frames. ], tot_loss[loss=0.358, simple_loss=0.3619, pruned_loss=0.1771, over 2652750.56 frames. ], batch size: 87, lr: 4.29e-02, grad_scale: 32.0 2023-04-16 12:35:48,169 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.727e+02 4.901e+02 5.855e+02 6.848e+02 1.405e+03, threshold=1.171e+03, percent-clipped=2.0 2023-04-16 12:35:49,210 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4202.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 12:35:49,266 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4202.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:36:32,957 INFO [train.py:893] (1/4) Epoch 2, batch 1550, loss[loss=0.36, simple_loss=0.3708, pruned_loss=0.1746, over 13537.00 frames. ], tot_loss[loss=0.3567, simple_loss=0.3612, pruned_loss=0.1761, over 2653098.17 frames. ], batch size: 83, lr: 4.28e-02, grad_scale: 32.0 2023-04-16 12:36:33,147 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4250.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 12:36:58,716 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6034, 3.7530, 3.3812, 3.6784, 2.2038, 3.0843, 3.4348, 1.7960], device='cuda:1'), covar=tensor([0.0071, 0.0423, 0.0238, 0.0117, 0.1898, 0.0366, 0.0481, 0.2672], device='cuda:1'), in_proj_covar=tensor([0.0040, 0.0051, 0.0050, 0.0035, 0.0100, 0.0066, 0.0055, 0.0113], device='cuda:1'), out_proj_covar=tensor([3.9569e-05, 6.0303e-05, 5.2889e-05, 4.0461e-05, 1.0634e-04, 6.7590e-05, 6.3379e-05, 1.1408e-04], device='cuda:1') 2023-04-16 12:37:02,836 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4282.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:37:09,399 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=4288.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:37:19,995 INFO [train.py:893] (1/4) Epoch 2, batch 1600, loss[loss=0.3668, simple_loss=0.3766, pruned_loss=0.1785, over 13422.00 frames. ], tot_loss[loss=0.3578, simple_loss=0.3628, pruned_loss=0.1764, over 2653354.92 frames. ], batch size: 106, lr: 4.27e-02, grad_scale: 32.0 2023-04-16 12:37:20,819 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.949e+02 4.864e+02 6.312e+02 7.658e+02 1.439e+03, threshold=1.262e+03, percent-clipped=4.0 2023-04-16 12:37:50,358 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4332.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 12:38:06,049 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=4349.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:38:06,563 INFO [train.py:893] (1/4) Epoch 2, batch 1650, loss[loss=0.3355, simple_loss=0.3509, pruned_loss=0.16, over 13273.00 frames. ], tot_loss[loss=0.3543, simple_loss=0.3608, pruned_loss=0.1739, over 2655662.63 frames. ], batch size: 124, lr: 4.26e-02, grad_scale: 32.0 2023-04-16 12:38:28,535 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.6921, 1.7892, 3.3114, 2.8989, 2.8566, 3.1061, 2.9347, 2.1221], device='cuda:1'), covar=tensor([0.1289, 0.1891, 0.0139, 0.0268, 0.0171, 0.0142, 0.0114, 0.0723], device='cuda:1'), in_proj_covar=tensor([0.0076, 0.0104, 0.0050, 0.0047, 0.0045, 0.0049, 0.0039, 0.0063], device='cuda:1'), out_proj_covar=tensor([7.1409e-05, 1.0557e-04, 4.5156e-05, 4.2768e-05, 3.9269e-05, 4.5291e-05, 3.5485e-05, 5.8447e-05], device='cuda:1') 2023-04-16 12:38:34,802 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4380.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 12:38:52,491 INFO [train.py:893] (1/4) Epoch 2, batch 1700, loss[loss=0.3509, simple_loss=0.3631, pruned_loss=0.1693, over 13521.00 frames. ], tot_loss[loss=0.3552, simple_loss=0.3621, pruned_loss=0.1742, over 2658669.16 frames. ], batch size: 91, lr: 4.25e-02, grad_scale: 16.0 2023-04-16 12:38:54,507 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.588e+02 4.984e+02 6.280e+02 7.975e+02 1.860e+03, threshold=1.256e+03, percent-clipped=4.0 2023-04-16 12:38:55,822 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-04-16 12:39:02,885 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.84 vs. limit=2.0 2023-04-16 12:39:18,782 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=2.00 vs. limit=2.0 2023-04-16 12:39:39,337 INFO [train.py:893] (1/4) Epoch 2, batch 1750, loss[loss=0.3146, simple_loss=0.3325, pruned_loss=0.1484, over 13475.00 frames. ], tot_loss[loss=0.3528, simple_loss=0.3601, pruned_loss=0.1728, over 2661803.76 frames. ], batch size: 79, lr: 4.24e-02, grad_scale: 16.0 2023-04-16 12:39:50,393 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1827, 3.4332, 3.6659, 3.8963, 4.3483, 3.6477, 3.7161, 4.1883], device='cuda:1'), covar=tensor([0.0342, 0.0199, 0.0207, 0.0387, 0.0154, 0.0192, 0.0196, 0.0161], device='cuda:1'), in_proj_covar=tensor([0.0039, 0.0024, 0.0034, 0.0045, 0.0037, 0.0034, 0.0037, 0.0029], device='cuda:1'), out_proj_covar=tensor([7.2627e-05, 5.2191e-05, 6.8080e-05, 8.4958e-05, 7.3579e-05, 6.6109e-05, 6.9002e-05, 5.8241e-05], device='cuda:1') 2023-04-16 12:40:01,112 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2717, 3.5605, 4.3290, 3.6388, 4.6603, 4.0864, 4.4246, 4.0551], device='cuda:1'), covar=tensor([0.0102, 0.0172, 0.0098, 0.0103, 0.0095, 0.0086, 0.0157, 0.0092], device='cuda:1'), in_proj_covar=tensor([0.0023, 0.0027, 0.0024, 0.0025, 0.0018, 0.0024, 0.0019, 0.0021], device='cuda:1'), out_proj_covar=tensor([4.1959e-05, 5.0457e-05, 4.1654e-05, 4.6676e-05, 3.3119e-05, 4.4429e-05, 3.2874e-05, 3.7942e-05], device='cuda:1') 2023-04-16 12:40:22,234 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4497.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:40:24,551 INFO [train.py:893] (1/4) Epoch 2, batch 1800, loss[loss=0.3407, simple_loss=0.3539, pruned_loss=0.1638, over 13388.00 frames. ], tot_loss[loss=0.3489, simple_loss=0.3575, pruned_loss=0.1702, over 2662207.43 frames. ], batch size: 109, lr: 4.23e-02, grad_scale: 16.0 2023-04-16 12:40:26,849 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.000e+02 4.797e+02 5.759e+02 6.677e+02 1.553e+03, threshold=1.152e+03, percent-clipped=1.0 2023-04-16 12:40:31,182 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8269, 2.5590, 2.0492, 2.9518, 3.2802, 3.2682, 2.2708, 1.7477], device='cuda:1'), covar=tensor([0.0271, 0.0560, 0.0475, 0.0192, 0.0117, 0.0151, 0.0378, 0.1059], device='cuda:1'), in_proj_covar=tensor([0.0040, 0.0062, 0.0047, 0.0039, 0.0035, 0.0035, 0.0046, 0.0069], device='cuda:1'), out_proj_covar=tensor([5.1090e-05, 7.0071e-05, 5.7955e-05, 4.5236e-05, 4.5964e-05, 4.5901e-05, 5.5076e-05, 7.5444e-05], device='cuda:1') 2023-04-16 12:40:52,739 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.3581, 4.8385, 4.8864, 4.7982, 4.5661, 4.7750, 5.1270, 4.7230], device='cuda:1'), covar=tensor([0.0804, 0.0697, 0.1958, 0.2215, 0.0634, 0.0844, 0.0924, 0.0778], device='cuda:1'), in_proj_covar=tensor([0.0151, 0.0144, 0.0196, 0.0197, 0.0107, 0.0169, 0.0191, 0.0132], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001], device='cuda:1') 2023-04-16 12:41:11,400 INFO [train.py:893] (1/4) Epoch 2, batch 1850, loss[loss=0.3417, simple_loss=0.3411, pruned_loss=0.1712, over 13345.00 frames. ], tot_loss[loss=0.3472, simple_loss=0.3561, pruned_loss=0.1691, over 2657722.92 frames. ], batch size: 67, lr: 4.22e-02, grad_scale: 16.0 2023-04-16 12:41:16,382 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-16 12:41:23,457 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4905, 3.7419, 3.8887, 2.9507, 1.9564, 2.8812, 3.7928, 3.9201], device='cuda:1'), covar=tensor([0.0271, 0.0334, 0.0325, 0.1012, 0.1552, 0.0846, 0.0156, 0.0134], device='cuda:1'), in_proj_covar=tensor([0.0051, 0.0042, 0.0048, 0.0086, 0.0090, 0.0074, 0.0037, 0.0033], device='cuda:1'), out_proj_covar=tensor([5.1887e-05, 4.8626e-05, 5.1343e-05, 9.1004e-05, 9.0842e-05, 7.7308e-05, 3.9929e-05, 3.6041e-05], device='cuda:1') 2023-04-16 12:41:30,542 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.0169, 4.6649, 4.5657, 4.4244, 4.2519, 4.5119, 4.8870, 4.5241], device='cuda:1'), covar=tensor([0.0622, 0.0548, 0.1412, 0.1988, 0.0613, 0.0904, 0.0741, 0.0527], device='cuda:1'), in_proj_covar=tensor([0.0154, 0.0145, 0.0190, 0.0194, 0.0109, 0.0171, 0.0193, 0.0131], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001], device='cuda:1') 2023-04-16 12:41:41,068 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4582.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:41:57,804 INFO [train.py:893] (1/4) Epoch 2, batch 1900, loss[loss=0.3572, simple_loss=0.3667, pruned_loss=0.1739, over 13416.00 frames. ], tot_loss[loss=0.3468, simple_loss=0.3552, pruned_loss=0.1692, over 2658295.78 frames. ], batch size: 95, lr: 4.21e-02, grad_scale: 16.0 2023-04-16 12:41:59,810 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.107e+02 4.525e+02 5.790e+02 7.364e+02 1.699e+03, threshold=1.158e+03, percent-clipped=5.0 2023-04-16 12:42:15,045 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.84 vs. limit=2.0 2023-04-16 12:42:25,425 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4630.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:42:38,614 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=4644.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:42:44,056 INFO [train.py:893] (1/4) Epoch 2, batch 1950, loss[loss=0.3732, simple_loss=0.3742, pruned_loss=0.1861, over 13217.00 frames. ], tot_loss[loss=0.3448, simple_loss=0.3538, pruned_loss=0.168, over 2661229.35 frames. ], batch size: 132, lr: 4.20e-02, grad_scale: 16.0 2023-04-16 12:43:29,734 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 2023-04-16 12:43:30,749 INFO [train.py:893] (1/4) Epoch 2, batch 2000, loss[loss=0.3414, simple_loss=0.3457, pruned_loss=0.1685, over 13443.00 frames. ], tot_loss[loss=0.3482, simple_loss=0.3565, pruned_loss=0.17, over 2661705.54 frames. ], batch size: 65, lr: 4.19e-02, grad_scale: 16.0 2023-04-16 12:43:32,435 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.843e+02 4.530e+02 5.357e+02 6.358e+02 1.307e+03, threshold=1.071e+03, percent-clipped=1.0 2023-04-16 12:43:38,382 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-16 12:43:51,512 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.40 vs. limit=5.0 2023-04-16 12:44:17,028 INFO [train.py:893] (1/4) Epoch 2, batch 2050, loss[loss=0.3659, simple_loss=0.3767, pruned_loss=0.1775, over 13433.00 frames. ], tot_loss[loss=0.3491, simple_loss=0.3576, pruned_loss=0.1703, over 2666125.05 frames. ], batch size: 106, lr: 4.17e-02, grad_scale: 16.0 2023-04-16 12:44:29,703 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9861, 2.3893, 2.6905, 1.6738, 2.2783, 2.4269, 2.9219, 1.4071], device='cuda:1'), covar=tensor([0.0228, 0.0886, 0.0195, 0.0513, 0.0468, 0.0265, 0.0502, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0032, 0.0028, 0.0033, 0.0033, 0.0043, 0.0031, 0.0028, 0.0045], device='cuda:1'), out_proj_covar=tensor([3.7334e-05, 3.7520e-05, 3.7708e-05, 3.7667e-05, 5.2918e-05, 3.6266e-05, 3.5916e-05, 5.4163e-05], device='cuda:1') 2023-04-16 12:44:49,597 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9937, 2.4129, 1.9753, 1.6907, 1.6678, 2.0608, 1.6237, 2.1638], device='cuda:1'), covar=tensor([0.0383, 0.0379, 0.0888, 0.0507, 0.0478, 0.0230, 0.0737, 0.0478], device='cuda:1'), in_proj_covar=tensor([0.0042, 0.0038, 0.0036, 0.0042, 0.0031, 0.0037, 0.0053, 0.0042], device='cuda:1'), out_proj_covar=tensor([4.0894e-05, 3.7885e-05, 3.7185e-05, 3.5977e-05, 3.6249e-05, 3.0487e-05, 5.0911e-05, 3.8928e-05], device='cuda:1') 2023-04-16 12:45:01,087 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4797.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:45:03,348 INFO [train.py:893] (1/4) Epoch 2, batch 2100, loss[loss=0.339, simple_loss=0.3611, pruned_loss=0.1585, over 13459.00 frames. ], tot_loss[loss=0.3465, simple_loss=0.3554, pruned_loss=0.1688, over 2665353.75 frames. ], batch size: 103, lr: 4.16e-02, grad_scale: 16.0 2023-04-16 12:45:05,358 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.512e+02 4.812e+02 5.608e+02 7.076e+02 1.162e+03, threshold=1.122e+03, percent-clipped=2.0 2023-04-16 12:45:31,456 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6640, 4.1279, 4.1306, 3.4027, 3.8879, 2.5804, 4.3276, 3.1122], device='cuda:1'), covar=tensor([0.0930, 0.0333, 0.0199, 0.0598, 0.0328, 0.2548, 0.0166, 0.2338], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0074, 0.0077, 0.0102, 0.0083, 0.0125, 0.0058, 0.0147], device='cuda:1'), out_proj_covar=tensor([1.4281e-04, 8.6815e-05, 8.7671e-05, 1.2033e-04, 9.3411e-05, 1.4354e-04, 6.9537e-05, 1.8595e-04], device='cuda:1') 2023-04-16 12:45:45,932 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4845.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:45:50,053 INFO [train.py:893] (1/4) Epoch 2, batch 2150, loss[loss=0.33, simple_loss=0.3572, pruned_loss=0.1514, over 13504.00 frames. ], tot_loss[loss=0.3438, simple_loss=0.3542, pruned_loss=0.1667, over 2666127.44 frames. ], batch size: 93, lr: 4.15e-02, grad_scale: 16.0 2023-04-16 12:46:36,470 INFO [train.py:893] (1/4) Epoch 2, batch 2200, loss[loss=0.3546, simple_loss=0.3567, pruned_loss=0.1762, over 13440.00 frames. ], tot_loss[loss=0.3432, simple_loss=0.3539, pruned_loss=0.1663, over 2664815.85 frames. ], batch size: 65, lr: 4.14e-02, grad_scale: 16.0 2023-04-16 12:46:38,212 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.919e+02 3.986e+02 4.975e+02 7.024e+02 1.362e+03, threshold=9.950e+02, percent-clipped=2.0 2023-04-16 12:46:46,800 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.4211, 1.8218, 1.8003, 1.5813, 1.2059, 1.3160, 1.7836, 1.5928], device='cuda:1'), covar=tensor([0.0352, 0.0331, 0.0423, 0.0257, 0.0241, 0.0234, 0.0390, 0.0239], device='cuda:1'), in_proj_covar=tensor([0.0031, 0.0029, 0.0030, 0.0030, 0.0027, 0.0029, 0.0029, 0.0027], device='cuda:1'), out_proj_covar=tensor([3.3004e-05, 3.4051e-05, 3.4339e-05, 3.0592e-05, 2.7570e-05, 3.6332e-05, 3.0568e-05, 2.8364e-05], device='cuda:1') 2023-04-16 12:46:49,076 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.2990, 4.0095, 3.5536, 3.7782, 4.0249, 4.4118, 4.2036, 4.3048], device='cuda:1'), covar=tensor([0.0485, 0.0285, 0.0350, 0.0430, 0.0193, 0.0187, 0.0192, 0.0236], device='cuda:1'), in_proj_covar=tensor([0.0055, 0.0049, 0.0050, 0.0065, 0.0045, 0.0055, 0.0047, 0.0055], device='cuda:1'), out_proj_covar=tensor([1.0468e-04, 9.6913e-05, 1.0171e-04, 1.2075e-04, 9.7097e-05, 1.1019e-04, 9.0578e-05, 1.0484e-04], device='cuda:1') 2023-04-16 12:47:16,984 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=4944.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:47:23,251 INFO [train.py:893] (1/4) Epoch 2, batch 2250, loss[loss=0.3001, simple_loss=0.3213, pruned_loss=0.1395, over 13541.00 frames. ], tot_loss[loss=0.342, simple_loss=0.3525, pruned_loss=0.1657, over 2662908.04 frames. ], batch size: 72, lr: 4.13e-02, grad_scale: 16.0 2023-04-16 12:48:01,028 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=4992.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 12:48:08,817 INFO [train.py:893] (1/4) Epoch 2, batch 2300, loss[loss=0.3973, simple_loss=0.3891, pruned_loss=0.2027, over 13227.00 frames. ], tot_loss[loss=0.3403, simple_loss=0.3512, pruned_loss=0.1647, over 2662640.79 frames. ], batch size: 124, lr: 4.12e-02, grad_scale: 16.0 2023-04-16 12:48:10,811 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.990e+02 4.129e+02 4.819e+02 6.882e+02 1.653e+03, threshold=9.638e+02, percent-clipped=3.0 2023-04-16 12:48:21,651 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.8727, 2.1206, 3.3156, 3.0401, 3.0643, 2.6877, 2.8033, 2.2423], device='cuda:1'), covar=tensor([0.1284, 0.1734, 0.0201, 0.0325, 0.0233, 0.0290, 0.0154, 0.0850], device='cuda:1'), in_proj_covar=tensor([0.0093, 0.0113, 0.0055, 0.0053, 0.0054, 0.0060, 0.0046, 0.0076], device='cuda:1'), out_proj_covar=tensor([8.9362e-05, 1.1389e-04, 5.0702e-05, 5.1110e-05, 4.9213e-05, 5.6530e-05, 4.4595e-05, 7.3007e-05], device='cuda:1') 2023-04-16 12:48:28,525 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6816, 3.7868, 3.9661, 2.9718, 2.4952, 2.8729, 3.9844, 4.0691], device='cuda:1'), covar=tensor([0.0286, 0.0473, 0.0274, 0.1268, 0.1503, 0.1065, 0.0151, 0.0118], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0053, 0.0059, 0.0103, 0.0107, 0.0087, 0.0043, 0.0039], device='cuda:1'), out_proj_covar=tensor([7.1876e-05, 6.5207e-05, 6.8266e-05, 1.1385e-04, 1.1402e-04, 9.6943e-05, 5.0129e-05, 4.5310e-05], device='cuda:1') 2023-04-16 12:48:55,277 INFO [train.py:893] (1/4) Epoch 2, batch 2350, loss[loss=0.3353, simple_loss=0.3453, pruned_loss=0.1626, over 13087.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3504, pruned_loss=0.164, over 2661639.85 frames. ], batch size: 142, lr: 4.11e-02, grad_scale: 16.0 2023-04-16 12:49:08,627 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9919, 4.3350, 4.4812, 3.7091, 4.0024, 2.6809, 4.5710, 3.1161], device='cuda:1'), covar=tensor([0.0912, 0.0335, 0.0196, 0.0701, 0.0534, 0.2454, 0.0180, 0.3543], device='cuda:1'), in_proj_covar=tensor([0.0127, 0.0077, 0.0085, 0.0111, 0.0092, 0.0133, 0.0060, 0.0159], device='cuda:1'), out_proj_covar=tensor([1.5330e-04, 9.2360e-05, 9.8014e-05, 1.3195e-04, 1.0487e-04, 1.5477e-04, 7.2521e-05, 2.0418e-04], device='cuda:1') 2023-04-16 12:49:20,459 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-16 12:49:34,068 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.4767, 2.1276, 3.3973, 3.0384, 3.2070, 2.7790, 2.7975, 2.0354], device='cuda:1'), covar=tensor([0.1511, 0.1646, 0.0155, 0.0425, 0.0340, 0.0187, 0.0158, 0.0972], device='cuda:1'), in_proj_covar=tensor([0.0097, 0.0118, 0.0057, 0.0056, 0.0055, 0.0062, 0.0049, 0.0079], device='cuda:1'), out_proj_covar=tensor([9.4343e-05, 1.1909e-04, 5.1900e-05, 5.3646e-05, 5.1053e-05, 5.8684e-05, 4.6214e-05, 7.6304e-05], device='cuda:1') 2023-04-16 12:49:41,168 INFO [train.py:893] (1/4) Epoch 2, batch 2400, loss[loss=0.3258, simple_loss=0.3421, pruned_loss=0.1547, over 13338.00 frames. ], tot_loss[loss=0.339, simple_loss=0.3498, pruned_loss=0.1641, over 2662349.65 frames. ], batch size: 73, lr: 4.10e-02, grad_scale: 16.0 2023-04-16 12:49:43,540 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.916e+02 4.826e+02 5.829e+02 7.201e+02 1.538e+03, threshold=1.166e+03, percent-clipped=7.0 2023-04-16 12:50:24,723 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0237, 2.4970, 2.6055, 3.4780, 3.4743, 3.5809, 2.7878, 2.0492], device='cuda:1'), covar=tensor([0.0288, 0.0797, 0.0480, 0.0081, 0.0187, 0.0120, 0.0328, 0.1050], device='cuda:1'), in_proj_covar=tensor([0.0043, 0.0076, 0.0062, 0.0041, 0.0036, 0.0039, 0.0055, 0.0076], device='cuda:1'), out_proj_covar=tensor([6.0287e-05, 9.2217e-05, 7.8784e-05, 5.1311e-05, 5.1285e-05, 5.3292e-05, 7.0184e-05, 9.0153e-05], device='cuda:1') 2023-04-16 12:50:27,561 INFO [train.py:893] (1/4) Epoch 2, batch 2450, loss[loss=0.2777, simple_loss=0.2934, pruned_loss=0.1311, over 12756.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3498, pruned_loss=0.1643, over 2661604.18 frames. ], batch size: 52, lr: 4.09e-02, grad_scale: 16.0 2023-04-16 12:50:28,813 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5266, 3.2599, 3.2635, 3.4067, 2.1003, 2.7214, 3.0970, 1.7558], device='cuda:1'), covar=tensor([0.0090, 0.0407, 0.0248, 0.0142, 0.1587, 0.0520, 0.0525, 0.3138], device='cuda:1'), in_proj_covar=tensor([0.0049, 0.0064, 0.0063, 0.0046, 0.0119, 0.0085, 0.0077, 0.0133], device='cuda:1'), out_proj_covar=tensor([5.4730e-05, 8.3060e-05, 7.5213e-05, 5.8734e-05, 1.3534e-04, 9.6588e-05, 9.5586e-05, 1.4570e-04], device='cuda:1') 2023-04-16 12:50:29,728 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.55 vs. limit=5.0 2023-04-16 12:50:30,413 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.4812, 2.0402, 1.5847, 1.6049, 1.3641, 1.5976, 1.4591, 1.7378], device='cuda:1'), covar=tensor([0.0358, 0.0395, 0.0570, 0.0284, 0.0193, 0.0194, 0.0399, 0.0285], device='cuda:1'), in_proj_covar=tensor([0.0042, 0.0039, 0.0036, 0.0041, 0.0032, 0.0036, 0.0049, 0.0040], device='cuda:1'), out_proj_covar=tensor([4.1477e-05, 3.8352e-05, 3.7492e-05, 3.5973e-05, 3.6987e-05, 3.0080e-05, 4.7131e-05, 3.7483e-05], device='cuda:1') 2023-04-16 12:50:36,948 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.33 vs. limit=5.0 2023-04-16 12:50:42,520 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0284, 2.6260, 2.8625, 3.4881, 3.6330, 3.6608, 2.6650, 2.2529], device='cuda:1'), covar=tensor([0.0291, 0.0694, 0.0365, 0.0138, 0.0107, 0.0104, 0.0369, 0.0790], device='cuda:1'), in_proj_covar=tensor([0.0041, 0.0075, 0.0061, 0.0041, 0.0035, 0.0038, 0.0055, 0.0074], device='cuda:1'), out_proj_covar=tensor([5.8764e-05, 9.0440e-05, 7.6914e-05, 5.0630e-05, 4.9769e-05, 5.2481e-05, 7.0238e-05, 8.8108e-05], device='cuda:1') 2023-04-16 12:51:12,664 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5082, 3.7171, 3.6240, 2.7112, 1.9131, 2.7549, 3.5942, 3.9702], device='cuda:1'), covar=tensor([0.0226, 0.0293, 0.0297, 0.1014, 0.1537, 0.0771, 0.0181, 0.0088], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0053, 0.0060, 0.0101, 0.0106, 0.0086, 0.0045, 0.0041], device='cuda:1'), out_proj_covar=tensor([7.4434e-05, 6.6615e-05, 7.0359e-05, 1.1521e-04, 1.1457e-04, 9.8431e-05, 5.3163e-05, 4.8834e-05], device='cuda:1') 2023-04-16 12:51:13,102 INFO [train.py:893] (1/4) Epoch 2, batch 2500, loss[loss=0.3599, simple_loss=0.3612, pruned_loss=0.1793, over 13324.00 frames. ], tot_loss[loss=0.3369, simple_loss=0.3483, pruned_loss=0.1628, over 2662892.86 frames. ], batch size: 118, lr: 4.08e-02, grad_scale: 16.0 2023-04-16 12:51:15,140 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.258e+02 4.389e+02 5.492e+02 6.426e+02 1.391e+03, threshold=1.098e+03, percent-clipped=1.0 2023-04-16 12:51:40,856 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2013, 4.4307, 4.3093, 3.3514, 2.4582, 3.1276, 4.3283, 4.4576], device='cuda:1'), covar=tensor([0.0231, 0.0297, 0.0287, 0.0891, 0.1482, 0.0830, 0.0225, 0.0226], device='cuda:1'), in_proj_covar=tensor([0.0066, 0.0053, 0.0060, 0.0100, 0.0106, 0.0087, 0.0045, 0.0042], device='cuda:1'), out_proj_covar=tensor([7.5830e-05, 6.6498e-05, 7.0433e-05, 1.1521e-04, 1.1482e-04, 9.9302e-05, 5.3497e-05, 4.9381e-05], device='cuda:1') 2023-04-16 12:51:59,502 INFO [train.py:893] (1/4) Epoch 2, batch 2550, loss[loss=0.3214, simple_loss=0.3417, pruned_loss=0.1506, over 13496.00 frames. ], tot_loss[loss=0.3363, simple_loss=0.3478, pruned_loss=0.1624, over 2661188.34 frames. ], batch size: 70, lr: 4.07e-02, grad_scale: 16.0 2023-04-16 12:52:24,131 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-16 12:52:46,263 INFO [train.py:893] (1/4) Epoch 2, batch 2600, loss[loss=0.3794, simple_loss=0.3741, pruned_loss=0.1924, over 13363.00 frames. ], tot_loss[loss=0.3376, simple_loss=0.3487, pruned_loss=0.1633, over 2662442.88 frames. ], batch size: 73, lr: 4.06e-02, grad_scale: 16.0 2023-04-16 12:52:47,915 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.769e+02 4.568e+02 5.360e+02 6.678e+02 1.569e+03, threshold=1.072e+03, percent-clipped=3.0 2023-04-16 12:53:24,050 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0498, 3.7539, 4.3290, 3.3359, 4.6784, 3.6044, 4.5978, 4.1357], device='cuda:1'), covar=tensor([0.0108, 0.0111, 0.0098, 0.0097, 0.0091, 0.0127, 0.0093, 0.0071], device='cuda:1'), in_proj_covar=tensor([0.0022, 0.0025, 0.0024, 0.0024, 0.0018, 0.0024, 0.0018, 0.0020], device='cuda:1'), out_proj_covar=tensor([4.9983e-05, 5.7228e-05, 5.3145e-05, 5.5856e-05, 4.0418e-05, 5.4496e-05, 3.9664e-05, 4.6293e-05], device='cuda:1') 2023-04-16 12:53:28,392 INFO [train.py:893] (1/4) Epoch 2, batch 2650, loss[loss=0.3166, simple_loss=0.3402, pruned_loss=0.1465, over 13453.00 frames. ], tot_loss[loss=0.3378, simple_loss=0.3484, pruned_loss=0.1636, over 2654545.64 frames. ], batch size: 79, lr: 4.05e-02, grad_scale: 16.0 2023-04-16 12:53:47,413 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.77 vs. limit=5.0 2023-04-16 12:54:01,698 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.87 vs. limit=2.0 2023-04-16 12:54:25,449 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 12:54:35,446 INFO [train.py:893] (1/4) Epoch 3, batch 0, loss[loss=0.3191, simple_loss=0.3255, pruned_loss=0.1563, over 13386.00 frames. ], tot_loss[loss=0.3191, simple_loss=0.3255, pruned_loss=0.1563, over 13386.00 frames. ], batch size: 62, lr: 3.84e-02, grad_scale: 16.0 2023-04-16 12:54:35,446 INFO [train.py:918] (1/4) Computing validation loss 2023-04-16 12:54:57,765 INFO [train.py:927] (1/4) Epoch 3, validation: loss=0.2472, simple_loss=0.2811, pruned_loss=0.1067, over 2446609.00 frames. 2023-04-16 12:54:57,766 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12250MB 2023-04-16 12:55:00,478 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.744e+02 4.537e+02 5.513e+02 6.512e+02 1.233e+03, threshold=1.103e+03, percent-clipped=2.0 2023-04-16 12:55:44,987 INFO [train.py:893] (1/4) Epoch 3, batch 50, loss[loss=0.3381, simple_loss=0.3609, pruned_loss=0.1577, over 13519.00 frames. ], tot_loss[loss=0.3228, simple_loss=0.3354, pruned_loss=0.155, over 603165.98 frames. ], batch size: 98, lr: 3.83e-02, grad_scale: 16.0 2023-04-16 12:55:57,076 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0994, 2.7272, 2.8925, 3.7860, 3.7480, 3.5681, 2.8958, 2.2418], device='cuda:1'), covar=tensor([0.0300, 0.0728, 0.0451, 0.0065, 0.0091, 0.0262, 0.0354, 0.0989], device='cuda:1'), in_proj_covar=tensor([0.0045, 0.0081, 0.0066, 0.0041, 0.0036, 0.0041, 0.0058, 0.0078], device='cuda:1'), out_proj_covar=tensor([6.4867e-05, 9.9642e-05, 8.4890e-05, 5.1740e-05, 5.3110e-05, 5.7292e-05, 7.4807e-05, 9.4459e-05], device='cuda:1') 2023-04-16 12:56:09,247 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-16 12:56:09,247 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-16 12:56:09,247 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-16 12:56:09,255 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-16 12:56:09,271 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-16 12:56:09,293 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-16 12:56:09,302 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-16 12:56:27,549 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 2023-04-16 12:56:31,872 INFO [train.py:893] (1/4) Epoch 3, batch 100, loss[loss=0.3604, simple_loss=0.3667, pruned_loss=0.1771, over 13428.00 frames. ], tot_loss[loss=0.3298, simple_loss=0.34, pruned_loss=0.1598, over 1053664.07 frames. ], batch size: 95, lr: 3.82e-02, grad_scale: 16.0 2023-04-16 12:56:33,000 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.9094, 4.8115, 5.0992, 4.8628, 5.3321, 4.7352, 5.3632, 5.2718], device='cuda:1'), covar=tensor([0.0281, 0.0398, 0.0419, 0.0324, 0.0395, 0.0543, 0.0352, 0.0344], device='cuda:1'), in_proj_covar=tensor([0.0107, 0.0129, 0.0119, 0.0092, 0.0154, 0.0131, 0.0092, 0.0103], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001], device='cuda:1') 2023-04-16 12:56:34,460 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.338e+02 4.385e+02 5.122e+02 6.357e+02 1.277e+03, threshold=1.024e+03, percent-clipped=2.0 2023-04-16 12:57:18,782 INFO [train.py:893] (1/4) Epoch 3, batch 150, loss[loss=0.3261, simple_loss=0.3443, pruned_loss=0.154, over 13344.00 frames. ], tot_loss[loss=0.3342, simple_loss=0.3428, pruned_loss=0.1628, over 1397226.70 frames. ], batch size: 118, lr: 3.81e-02, grad_scale: 16.0 2023-04-16 12:57:32,842 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0 2023-04-16 12:58:06,407 INFO [train.py:893] (1/4) Epoch 3, batch 200, loss[loss=0.358, simple_loss=0.3496, pruned_loss=0.1832, over 13385.00 frames. ], tot_loss[loss=0.3349, simple_loss=0.3443, pruned_loss=0.1627, over 1669590.71 frames. ], batch size: 62, lr: 3.80e-02, grad_scale: 16.0 2023-04-16 12:58:09,202 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.881e+02 4.242e+02 5.295e+02 6.523e+02 1.110e+03, threshold=1.059e+03, percent-clipped=3.0 2023-04-16 12:58:52,256 INFO [train.py:893] (1/4) Epoch 3, batch 250, loss[loss=0.3195, simple_loss=0.3407, pruned_loss=0.1492, over 13498.00 frames. ], tot_loss[loss=0.3366, simple_loss=0.3465, pruned_loss=0.1633, over 1889597.42 frames. ], batch size: 91, lr: 3.79e-02, grad_scale: 16.0 2023-04-16 12:58:53,499 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5650.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 12:59:22,400 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8890, 3.8123, 2.5198, 3.8153, 3.7515, 1.9478, 3.1699, 2.4677], device='cuda:1'), covar=tensor([0.0190, 0.0377, 0.1317, 0.0140, 0.0311, 0.1356, 0.1207, 0.1994], device='cuda:1'), in_proj_covar=tensor([0.0061, 0.0073, 0.0124, 0.0052, 0.0075, 0.0105, 0.0099, 0.0151], device='cuda:1'), out_proj_covar=tensor([9.5542e-05, 1.0872e-04, 1.6419e-04, 7.8101e-05, 1.0958e-04, 1.4053e-04, 1.4226e-04, 1.8665e-04], device='cuda:1') 2023-04-16 12:59:24,023 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0448, 2.0097, 2.5006, 3.5540, 3.4004, 3.4330, 2.6110, 1.8735], device='cuda:1'), covar=tensor([0.0258, 0.1081, 0.0569, 0.0088, 0.0149, 0.0136, 0.0416, 0.1187], device='cuda:1'), in_proj_covar=tensor([0.0045, 0.0083, 0.0072, 0.0043, 0.0038, 0.0040, 0.0060, 0.0081], device='cuda:1'), out_proj_covar=tensor([6.6439e-05, 1.0453e-04, 9.3062e-05, 5.3997e-05, 5.5878e-05, 5.7016e-05, 7.9159e-05, 9.9612e-05], device='cuda:1') 2023-04-16 12:59:40,030 INFO [train.py:893] (1/4) Epoch 3, batch 300, loss[loss=0.3345, simple_loss=0.3628, pruned_loss=0.1531, over 13451.00 frames. ], tot_loss[loss=0.3355, simple_loss=0.3467, pruned_loss=0.1621, over 2062660.06 frames. ], batch size: 100, lr: 3.78e-02, grad_scale: 16.0 2023-04-16 12:59:42,576 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.716e+02 4.500e+02 5.480e+02 6.361e+02 1.148e+03, threshold=1.096e+03, percent-clipped=2.0 2023-04-16 12:59:51,174 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5711.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 13:00:27,082 INFO [train.py:893] (1/4) Epoch 3, batch 350, loss[loss=0.3507, simple_loss=0.346, pruned_loss=0.1777, over 13472.00 frames. ], tot_loss[loss=0.3373, simple_loss=0.3481, pruned_loss=0.1633, over 2193977.90 frames. ], batch size: 100, lr: 3.77e-02, grad_scale: 16.0 2023-04-16 13:00:32,521 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5755.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:00:58,246 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2023-04-16 13:01:02,297 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6248, 4.0294, 3.8589, 2.7588, 2.0184, 2.9075, 3.8533, 4.0746], device='cuda:1'), covar=tensor([0.0325, 0.0328, 0.0305, 0.1292, 0.1733, 0.0963, 0.0130, 0.0075], device='cuda:1'), in_proj_covar=tensor([0.0077, 0.0061, 0.0069, 0.0114, 0.0120, 0.0098, 0.0052, 0.0047], device='cuda:1'), out_proj_covar=tensor([9.5472e-05, 8.1407e-05, 8.6340e-05, 1.3745e-04, 1.3541e-04, 1.1818e-04, 6.5509e-05, 5.8009e-05], device='cuda:1') 2023-04-16 13:01:13,815 INFO [train.py:893] (1/4) Epoch 3, batch 400, loss[loss=0.3312, simple_loss=0.3325, pruned_loss=0.165, over 13403.00 frames. ], tot_loss[loss=0.3368, simple_loss=0.3483, pruned_loss=0.1626, over 2296128.16 frames. ], batch size: 62, lr: 3.76e-02, grad_scale: 16.0 2023-04-16 13:01:16,258 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7871, 4.4860, 4.4462, 3.7890, 4.2672, 2.5528, 4.7630, 3.3391], device='cuda:1'), covar=tensor([0.0929, 0.0193, 0.0206, 0.0538, 0.0281, 0.2097, 0.0096, 0.2261], device='cuda:1'), in_proj_covar=tensor([0.0145, 0.0092, 0.0102, 0.0127, 0.0107, 0.0148, 0.0074, 0.0177], device='cuda:1'), out_proj_covar=tensor([1.8355e-04, 1.1375e-04, 1.2151e-04, 1.5219e-04, 1.2597e-04, 1.7822e-04, 9.1360e-05, 2.3287e-04], device='cuda:1') 2023-04-16 13:01:16,660 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.858e+02 4.189e+02 5.044e+02 6.148e+02 1.450e+03, threshold=1.009e+03, percent-clipped=1.0 2023-04-16 13:01:30,405 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5816.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:01:52,549 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.2591, 2.2939, 2.1964, 1.8764, 2.0749, 2.2455, 2.2886, 1.4407], device='cuda:1'), covar=tensor([0.0336, 0.0592, 0.0338, 0.0464, 0.0663, 0.0333, 0.0816, 0.1007], device='cuda:1'), in_proj_covar=tensor([0.0042, 0.0035, 0.0042, 0.0041, 0.0054, 0.0039, 0.0034, 0.0058], device='cuda:1'), out_proj_covar=tensor([5.4906e-05, 5.1060e-05, 5.2129e-05, 5.2214e-05, 7.2984e-05, 4.9370e-05, 4.9134e-05, 7.6120e-05], device='cuda:1') 2023-04-16 13:02:00,367 INFO [train.py:893] (1/4) Epoch 3, batch 450, loss[loss=0.3249, simple_loss=0.3407, pruned_loss=0.1546, over 13492.00 frames. ], tot_loss[loss=0.3413, simple_loss=0.352, pruned_loss=0.1653, over 2377969.95 frames. ], batch size: 81, lr: 3.76e-02, grad_scale: 16.0 2023-04-16 13:02:08,865 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5858.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:02:25,935 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-16 13:02:46,672 INFO [train.py:893] (1/4) Epoch 3, batch 500, loss[loss=0.2903, simple_loss=0.3201, pruned_loss=0.1303, over 13535.00 frames. ], tot_loss[loss=0.3392, simple_loss=0.3507, pruned_loss=0.1639, over 2438754.86 frames. ], batch size: 98, lr: 3.75e-02, grad_scale: 16.0 2023-04-16 13:02:49,925 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.506e+02 4.151e+02 5.175e+02 6.712e+02 1.423e+03, threshold=1.035e+03, percent-clipped=8.0 2023-04-16 13:03:05,852 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=5919.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:03:33,422 INFO [train.py:893] (1/4) Epoch 3, batch 550, loss[loss=0.3188, simple_loss=0.3394, pruned_loss=0.1491, over 13445.00 frames. ], tot_loss[loss=0.3357, simple_loss=0.3486, pruned_loss=0.1614, over 2490304.80 frames. ], batch size: 103, lr: 3.74e-02, grad_scale: 16.0 2023-04-16 13:04:11,534 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=5990.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:04:14,695 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.1236, 4.6052, 4.6537, 4.4379, 4.4425, 4.6032, 4.9630, 4.5399], device='cuda:1'), covar=tensor([0.0619, 0.0731, 0.1476, 0.2352, 0.0621, 0.1008, 0.0818, 0.0754], device='cuda:1'), in_proj_covar=tensor([0.0167, 0.0169, 0.0228, 0.0230, 0.0123, 0.0200, 0.0212, 0.0151], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 13:04:19,894 INFO [train.py:893] (1/4) Epoch 3, batch 600, loss[loss=0.3368, simple_loss=0.3516, pruned_loss=0.161, over 13449.00 frames. ], tot_loss[loss=0.3331, simple_loss=0.3462, pruned_loss=0.16, over 2522558.28 frames. ], batch size: 100, lr: 3.73e-02, grad_scale: 16.0 2023-04-16 13:04:26,350 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.607e+02 4.651e+02 5.653e+02 7.517e+02 1.178e+03, threshold=1.131e+03, percent-clipped=4.0 2023-04-16 13:04:29,952 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6006.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 13:04:31,685 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 2023-04-16 13:04:45,017 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 2023-04-16 13:05:03,924 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6043.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:05:06,277 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.1941, 1.9113, 1.4423, 1.4120, 1.1629, 1.6141, 1.2128, 1.9348], device='cuda:1'), covar=tensor([0.0542, 0.0453, 0.1197, 0.0625, 0.0313, 0.0204, 0.0649, 0.0447], device='cuda:1'), in_proj_covar=tensor([0.0046, 0.0043, 0.0042, 0.0051, 0.0038, 0.0040, 0.0054, 0.0045], device='cuda:1'), out_proj_covar=tensor([4.4769e-05, 4.0103e-05, 4.4027e-05, 4.4594e-05, 4.4476e-05, 3.5071e-05, 5.2591e-05, 4.1143e-05], device='cuda:1') 2023-04-16 13:05:08,423 INFO [train.py:893] (1/4) Epoch 3, batch 650, loss[loss=0.2912, simple_loss=0.3244, pruned_loss=0.129, over 13430.00 frames. ], tot_loss[loss=0.3314, simple_loss=0.345, pruned_loss=0.159, over 2556586.91 frames. ], batch size: 88, lr: 3.72e-02, grad_scale: 16.0 2023-04-16 13:05:11,129 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6051.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:05:13,813 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.42 vs. limit=5.0 2023-04-16 13:05:17,129 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.19 vs. limit=2.0 2023-04-16 13:05:23,754 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.83 vs. limit=2.0 2023-04-16 13:05:35,634 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.4490, 4.9187, 4.8235, 4.7355, 4.3511, 4.8365, 5.2791, 5.0334], device='cuda:1'), covar=tensor([0.0651, 0.0869, 0.2037, 0.2654, 0.0798, 0.1120, 0.0908, 0.0710], device='cuda:1'), in_proj_covar=tensor([0.0161, 0.0168, 0.0225, 0.0226, 0.0120, 0.0195, 0.0214, 0.0145], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 13:05:55,396 INFO [train.py:893] (1/4) Epoch 3, batch 700, loss[loss=0.304, simple_loss=0.3308, pruned_loss=0.1386, over 13439.00 frames. ], tot_loss[loss=0.3268, simple_loss=0.3414, pruned_loss=0.1561, over 2580797.29 frames. ], batch size: 106, lr: 3.71e-02, grad_scale: 16.0 2023-04-16 13:05:58,626 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.252e+02 4.009e+02 5.106e+02 6.344e+02 2.444e+03, threshold=1.021e+03, percent-clipped=2.0 2023-04-16 13:06:00,641 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6104.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:06:06,323 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6111.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:06:16,049 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7694, 3.5983, 3.8770, 3.8053, 4.3193, 3.9446, 3.7796, 4.3693], device='cuda:1'), covar=tensor([0.0204, 0.0170, 0.0170, 0.0346, 0.0120, 0.0129, 0.0198, 0.0091], device='cuda:1'), in_proj_covar=tensor([0.0040, 0.0023, 0.0036, 0.0048, 0.0039, 0.0034, 0.0037, 0.0029], device='cuda:1'), out_proj_covar=tensor([1.0056e-04, 7.0249e-05, 9.5388e-05, 1.2361e-04, 1.0841e-04, 8.9416e-05, 9.6768e-05, 7.9541e-05], device='cuda:1') 2023-04-16 13:06:29,466 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6135.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:06:41,879 INFO [train.py:893] (1/4) Epoch 3, batch 750, loss[loss=0.3106, simple_loss=0.3174, pruned_loss=0.1519, over 13385.00 frames. ], tot_loss[loss=0.3271, simple_loss=0.3413, pruned_loss=0.1564, over 2598404.83 frames. ], batch size: 62, lr: 3.70e-02, grad_scale: 16.0 2023-04-16 13:06:55,402 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4750, 4.4154, 4.6836, 4.5190, 4.8032, 4.2923, 4.8394, 4.7879], device='cuda:1'), covar=tensor([0.0326, 0.0433, 0.0477, 0.0379, 0.0533, 0.0577, 0.0415, 0.0417], device='cuda:1'), in_proj_covar=tensor([0.0107, 0.0130, 0.0119, 0.0095, 0.0162, 0.0140, 0.0094, 0.0106], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-16 13:07:02,315 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6170.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:07:26,825 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6196.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:07:29,127 INFO [train.py:893] (1/4) Epoch 3, batch 800, loss[loss=0.3483, simple_loss=0.3632, pruned_loss=0.1667, over 13352.00 frames. ], tot_loss[loss=0.3295, simple_loss=0.3437, pruned_loss=0.1576, over 2617180.87 frames. ], batch size: 118, lr: 3.69e-02, grad_scale: 16.0 2023-04-16 13:07:31,838 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.704e+02 5.071e+02 5.835e+02 7.161e+02 1.086e+03, threshold=1.167e+03, percent-clipped=4.0 2023-04-16 13:07:32,927 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6203.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:07:40,340 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6211.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:07:42,819 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6214.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:07:59,045 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6231.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:08:11,830 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.58 vs. limit=5.0 2023-04-16 13:08:15,447 INFO [train.py:893] (1/4) Epoch 3, batch 850, loss[loss=0.319, simple_loss=0.3441, pruned_loss=0.1469, over 13513.00 frames. ], tot_loss[loss=0.3296, simple_loss=0.344, pruned_loss=0.1576, over 2628863.58 frames. ], batch size: 91, lr: 3.68e-02, grad_scale: 16.0 2023-04-16 13:08:29,070 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6264.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:08:30,628 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6266.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:08:35,655 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6272.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:09:00,755 INFO [train.py:893] (1/4) Epoch 3, batch 900, loss[loss=0.3255, simple_loss=0.3352, pruned_loss=0.1579, over 13537.00 frames. ], tot_loss[loss=0.3303, simple_loss=0.3441, pruned_loss=0.1583, over 2636918.37 frames. ], batch size: 72, lr: 3.67e-02, grad_scale: 16.0 2023-04-16 13:09:03,271 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.335e+02 4.386e+02 5.429e+02 6.427e+02 2.158e+03, threshold=1.086e+03, percent-clipped=2.0 2023-04-16 13:09:07,357 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6306.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 13:09:26,963 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6327.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:09:31,116 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-16 13:09:44,919 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6346.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:09:47,206 INFO [train.py:893] (1/4) Epoch 3, batch 950, loss[loss=0.3119, simple_loss=0.3322, pruned_loss=0.1458, over 13537.00 frames. ], tot_loss[loss=0.3302, simple_loss=0.3433, pruned_loss=0.1585, over 2645854.70 frames. ], batch size: 85, lr: 3.66e-02, grad_scale: 16.0 2023-04-16 13:09:52,156 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6354.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 13:10:25,041 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9928, 3.4682, 3.3252, 3.2691, 4.3600, 3.2311, 4.3353, 3.8443], device='cuda:1'), covar=tensor([0.0097, 0.0168, 0.0256, 0.0115, 0.0083, 0.0208, 0.0103, 0.0079], device='cuda:1'), in_proj_covar=tensor([0.0025, 0.0027, 0.0029, 0.0027, 0.0020, 0.0028, 0.0020, 0.0022], device='cuda:1'), out_proj_covar=tensor([6.7974e-05, 7.6559e-05, 7.6443e-05, 7.3084e-05, 5.4154e-05, 7.7696e-05, 5.2972e-05, 6.1670e-05], device='cuda:1') 2023-04-16 13:10:25,917 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5304, 4.1909, 3.9043, 3.9867, 4.1555, 4.4952, 4.1829, 4.1920], device='cuda:1'), covar=tensor([0.0409, 0.0194, 0.0212, 0.0463, 0.0169, 0.0196, 0.0208, 0.0264], device='cuda:1'), in_proj_covar=tensor([0.0059, 0.0055, 0.0053, 0.0078, 0.0052, 0.0067, 0.0051, 0.0063], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002, 0.0001, 0.0001], device='cuda:1') 2023-04-16 13:10:32,247 INFO [train.py:893] (1/4) Epoch 3, batch 1000, loss[loss=0.3441, simple_loss=0.3554, pruned_loss=0.1664, over 13541.00 frames. ], tot_loss[loss=0.3263, simple_loss=0.3399, pruned_loss=0.1563, over 2654461.01 frames. ], batch size: 87, lr: 3.66e-02, grad_scale: 32.0 2023-04-16 13:10:32,421 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6399.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:10:35,113 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.399e+02 4.240e+02 5.473e+02 6.639e+02 1.475e+03, threshold=1.095e+03, percent-clipped=3.0 2023-04-16 13:10:45,029 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6411.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:11:09,769 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0903, 3.8733, 4.1650, 4.3245, 4.9175, 4.3246, 4.3086, 4.7404], device='cuda:1'), covar=tensor([0.0196, 0.0160, 0.0136, 0.0307, 0.0087, 0.0111, 0.0153, 0.0087], device='cuda:1'), in_proj_covar=tensor([0.0038, 0.0022, 0.0035, 0.0047, 0.0037, 0.0034, 0.0036, 0.0027], device='cuda:1'), out_proj_covar=tensor([9.9739e-05, 7.1973e-05, 9.6793e-05, 1.2528e-04, 1.0708e-04, 9.4709e-05, 9.5753e-05, 7.8191e-05], device='cuda:1') 2023-04-16 13:11:19,997 INFO [train.py:893] (1/4) Epoch 3, batch 1050, loss[loss=0.2802, simple_loss=0.3061, pruned_loss=0.1271, over 13524.00 frames. ], tot_loss[loss=0.3222, simple_loss=0.3371, pruned_loss=0.1536, over 2658479.99 frames. ], batch size: 70, lr: 3.65e-02, grad_scale: 16.0 2023-04-16 13:11:28,608 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6459.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:11:31,202 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6462.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:11:35,860 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6467.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:11:57,828 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6491.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:12:05,655 INFO [train.py:893] (1/4) Epoch 3, batch 1100, loss[loss=0.3194, simple_loss=0.3437, pruned_loss=0.1475, over 13458.00 frames. ], tot_loss[loss=0.3218, simple_loss=0.3375, pruned_loss=0.153, over 2659397.09 frames. ], batch size: 103, lr: 3.64e-02, grad_scale: 16.0 2023-04-16 13:12:09,689 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.609e+02 4.664e+02 5.670e+02 6.887e+02 1.017e+03, threshold=1.134e+03, percent-clipped=0.0 2023-04-16 13:12:19,720 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6514.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:12:27,274 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6523.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 13:12:30,259 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6526.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:12:32,670 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6528.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:12:51,517 INFO [train.py:893] (1/4) Epoch 3, batch 1150, loss[loss=0.3059, simple_loss=0.3321, pruned_loss=0.1399, over 13465.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3359, pruned_loss=0.1511, over 2658040.26 frames. ], batch size: 103, lr: 3.63e-02, grad_scale: 16.0 2023-04-16 13:13:01,488 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6559.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:13:04,068 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6562.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:13:07,625 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.13 vs. limit=5.0 2023-04-16 13:13:08,977 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6567.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:13:38,316 INFO [train.py:893] (1/4) Epoch 3, batch 1200, loss[loss=0.3542, simple_loss=0.3598, pruned_loss=0.1743, over 13235.00 frames. ], tot_loss[loss=0.3183, simple_loss=0.3358, pruned_loss=0.1504, over 2663537.82 frames. ], batch size: 132, lr: 3.62e-02, grad_scale: 16.0 2023-04-16 13:13:42,312 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.523e+02 4.328e+02 5.221e+02 6.462e+02 1.421e+03, threshold=1.044e+03, percent-clipped=2.0 2023-04-16 13:13:59,103 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6622.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:14:01,577 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.6291, 1.7237, 3.6501, 3.1960, 3.1586, 2.8970, 2.8083, 1.9096], device='cuda:1'), covar=tensor([0.1787, 0.2084, 0.0082, 0.0282, 0.0213, 0.0295, 0.0280, 0.1310], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0143, 0.0060, 0.0065, 0.0069, 0.0079, 0.0061, 0.0102], device='cuda:1'), out_proj_covar=tensor([1.2496e-04, 1.4419e-04, 5.9817e-05, 7.3253e-05, 7.4048e-05, 7.8388e-05, 6.6357e-05, 1.0430e-04], device='cuda:1') 2023-04-16 13:14:03,390 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=6.03 vs. limit=5.0 2023-04-16 13:14:05,245 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-16 13:14:06,315 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6629.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:14:16,283 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-16 13:14:21,644 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6646.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:14:23,885 INFO [train.py:893] (1/4) Epoch 3, batch 1250, loss[loss=0.2978, simple_loss=0.3221, pruned_loss=0.1367, over 13532.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3364, pruned_loss=0.1512, over 2664460.64 frames. ], batch size: 98, lr: 3.61e-02, grad_scale: 16.0 2023-04-16 13:14:51,995 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=2.14 vs. limit=2.0 2023-04-16 13:15:02,932 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6690.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:15:06,836 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6694.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:15:10,879 INFO [train.py:893] (1/4) Epoch 3, batch 1300, loss[loss=0.3413, simple_loss=0.3578, pruned_loss=0.1624, over 13491.00 frames. ], tot_loss[loss=0.3231, simple_loss=0.3395, pruned_loss=0.1534, over 2659495.46 frames. ], batch size: 81, lr: 3.60e-02, grad_scale: 16.0 2023-04-16 13:15:11,158 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6699.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:15:14,240 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.037e+02 4.584e+02 5.620e+02 7.352e+02 1.514e+03, threshold=1.124e+03, percent-clipped=3.0 2023-04-16 13:15:45,645 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4151, 4.4517, 3.0057, 4.5072, 4.2279, 2.4428, 3.6709, 2.9448], device='cuda:1'), covar=tensor([0.0169, 0.0288, 0.1183, 0.0082, 0.0170, 0.1365, 0.0726, 0.1875], device='cuda:1'), in_proj_covar=tensor([0.0071, 0.0079, 0.0137, 0.0057, 0.0078, 0.0123, 0.0104, 0.0155], device='cuda:1'), out_proj_covar=tensor([1.1803e-04, 1.2669e-04, 1.9706e-04, 9.4566e-05, 1.2259e-04, 1.7423e-04, 1.6132e-04, 2.1308e-04], device='cuda:1') 2023-04-16 13:15:54,308 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6747.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:15:55,742 INFO [train.py:893] (1/4) Epoch 3, batch 1350, loss[loss=0.2985, simple_loss=0.3234, pruned_loss=0.1368, over 13461.00 frames. ], tot_loss[loss=0.3224, simple_loss=0.339, pruned_loss=0.1529, over 2658148.91 frames. ], batch size: 79, lr: 3.59e-02, grad_scale: 16.0 2023-04-16 13:16:12,137 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8814, 4.2766, 3.9502, 3.9067, 3.8927, 3.7784, 4.3373, 4.3218], device='cuda:1'), covar=tensor([0.0286, 0.0229, 0.0351, 0.0348, 0.0460, 0.0421, 0.0290, 0.0271], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0103, 0.0077, 0.0104, 0.0079, 0.0099, 0.0080, 0.0083], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 13:16:21,786 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9476, 1.9416, 3.8505, 3.4515, 3.3848, 3.5387, 3.1512, 2.5663], device='cuda:1'), covar=tensor([0.1907, 0.2290, 0.0117, 0.0379, 0.0280, 0.0253, 0.0217, 0.1044], device='cuda:1'), in_proj_covar=tensor([0.0134, 0.0149, 0.0066, 0.0067, 0.0068, 0.0085, 0.0061, 0.0106], device='cuda:1'), out_proj_covar=tensor([1.3394e-04, 1.5059e-04, 6.5689e-05, 7.5658e-05, 7.4548e-05, 8.4963e-05, 6.7144e-05, 1.0848e-04], device='cuda:1') 2023-04-16 13:16:34,998 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6791.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:16:41,386 INFO [train.py:893] (1/4) Epoch 3, batch 1400, loss[loss=0.3267, simple_loss=0.3447, pruned_loss=0.1543, over 13528.00 frames. ], tot_loss[loss=0.3193, simple_loss=0.3364, pruned_loss=0.1511, over 2660703.11 frames. ], batch size: 85, lr: 3.59e-02, grad_scale: 16.0 2023-04-16 13:16:42,025 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.87 vs. limit=2.0 2023-04-16 13:16:45,011 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.054e+02 4.522e+02 5.361e+02 7.195e+02 2.528e+03, threshold=1.072e+03, percent-clipped=2.0 2023-04-16 13:16:59,960 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6818.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 13:17:04,639 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6823.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:17:07,182 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6826.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:17:16,508 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9740, 3.4242, 3.6090, 3.9536, 1.8957, 3.0889, 3.4362, 1.7778], device='cuda:1'), covar=tensor([0.0069, 0.0439, 0.0274, 0.0131, 0.1549, 0.0451, 0.0521, 0.2570], device='cuda:1'), in_proj_covar=tensor([0.0057, 0.0086, 0.0093, 0.0056, 0.0135, 0.0103, 0.0102, 0.0148], device='cuda:1'), out_proj_covar=tensor([7.6830e-05, 1.2217e-04, 1.1836e-04, 8.0957e-05, 1.6536e-04, 1.2960e-04, 1.3703e-04, 1.8056e-04], device='cuda:1') 2023-04-16 13:17:18,835 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6839.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:17:27,652 INFO [train.py:893] (1/4) Epoch 3, batch 1450, loss[loss=0.3556, simple_loss=0.3594, pruned_loss=0.1758, over 13504.00 frames. ], tot_loss[loss=0.3198, simple_loss=0.3368, pruned_loss=0.1514, over 2664610.32 frames. ], batch size: 85, lr: 3.58e-02, grad_scale: 16.0 2023-04-16 13:17:37,008 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6859.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:17:43,869 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6867.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:17:50,331 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6874.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:17:53,887 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6878.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 13:18:01,223 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-04-16 13:18:12,414 INFO [train.py:893] (1/4) Epoch 3, batch 1500, loss[loss=0.2882, simple_loss=0.3053, pruned_loss=0.1355, over 13412.00 frames. ], tot_loss[loss=0.3194, simple_loss=0.3371, pruned_loss=0.1508, over 2664219.43 frames. ], batch size: 65, lr: 3.57e-02, grad_scale: 16.0 2023-04-16 13:18:16,569 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.769e+02 4.038e+02 4.963e+02 5.872e+02 1.075e+03, threshold=9.927e+02, percent-clipped=1.0 2023-04-16 13:18:20,095 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6907.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:18:21,971 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=6909.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:18:28,296 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6915.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:18:34,159 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=6922.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:18:49,028 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6939.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 13:18:58,210 INFO [train.py:893] (1/4) Epoch 3, batch 1550, loss[loss=0.3191, simple_loss=0.3365, pruned_loss=0.1508, over 13540.00 frames. ], tot_loss[loss=0.3192, simple_loss=0.3369, pruned_loss=0.1507, over 2662189.09 frames. ], batch size: 83, lr: 3.56e-02, grad_scale: 16.0 2023-04-16 13:19:16,938 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=6970.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:19:17,063 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=6970.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:19:31,004 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=6985.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:19:43,995 INFO [train.py:893] (1/4) Epoch 3, batch 1600, loss[loss=0.3215, simple_loss=0.3455, pruned_loss=0.1487, over 13559.00 frames. ], tot_loss[loss=0.3196, simple_loss=0.3376, pruned_loss=0.1508, over 2656165.56 frames. ], batch size: 89, lr: 3.55e-02, grad_scale: 16.0 2023-04-16 13:19:47,559 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.095e+02 4.094e+02 4.857e+02 5.932e+02 1.050e+03, threshold=9.715e+02, percent-clipped=2.0 2023-04-16 13:19:54,592 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 2023-04-16 13:20:13,544 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.3441, 2.3304, 2.0273, 2.3643, 2.2116, 1.2087, 2.4735, 1.4283], device='cuda:1'), covar=tensor([0.0429, 0.0667, 0.0450, 0.0340, 0.0783, 0.0742, 0.0784, 0.1044], device='cuda:1'), in_proj_covar=tensor([0.0051, 0.0041, 0.0049, 0.0046, 0.0063, 0.0056, 0.0038, 0.0067], device='cuda:1'), out_proj_covar=tensor([7.1576e-05, 6.5075e-05, 6.7410e-05, 6.3232e-05, 9.0328e-05, 7.7503e-05, 5.9477e-05, 9.2892e-05], device='cuda:1') 2023-04-16 13:20:29,965 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7048.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 13:20:30,410 INFO [train.py:893] (1/4) Epoch 3, batch 1650, loss[loss=0.3911, simple_loss=0.3852, pruned_loss=0.1985, over 13237.00 frames. ], tot_loss[loss=0.3197, simple_loss=0.3377, pruned_loss=0.1508, over 2653108.02 frames. ], batch size: 124, lr: 3.54e-02, grad_scale: 16.0 2023-04-16 13:20:32,585 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-04-16 13:21:09,272 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.90 vs. limit=2.0 2023-04-16 13:21:18,188 INFO [train.py:893] (1/4) Epoch 3, batch 1700, loss[loss=0.3484, simple_loss=0.3607, pruned_loss=0.1681, over 13339.00 frames. ], tot_loss[loss=0.319, simple_loss=0.3375, pruned_loss=0.1502, over 2649583.54 frames. ], batch size: 118, lr: 3.54e-02, grad_scale: 16.0 2023-04-16 13:21:21,488 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.842e+02 4.361e+02 5.760e+02 7.858e+02 1.557e+03, threshold=1.152e+03, percent-clipped=12.0 2023-04-16 13:21:26,773 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7109.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 13:21:34,803 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7118.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 13:21:37,106 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.3506, 4.7989, 4.6943, 4.6969, 4.6186, 4.7211, 5.2909, 4.9368], device='cuda:1'), covar=tensor([0.0751, 0.0747, 0.2072, 0.2695, 0.0598, 0.1245, 0.0690, 0.0650], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0187, 0.0255, 0.0254, 0.0132, 0.0214, 0.0229, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 13:21:38,727 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7123.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:21:54,680 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 2023-04-16 13:22:03,250 INFO [train.py:893] (1/4) Epoch 3, batch 1750, loss[loss=0.3154, simple_loss=0.3389, pruned_loss=0.146, over 13526.00 frames. ], tot_loss[loss=0.316, simple_loss=0.3352, pruned_loss=0.1484, over 2652866.92 frames. ], batch size: 98, lr: 3.53e-02, grad_scale: 16.0 2023-04-16 13:22:17,225 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7163.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:22:19,602 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7166.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:22:23,607 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7171.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:22:26,555 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.69 vs. limit=5.0 2023-04-16 13:22:49,674 INFO [train.py:893] (1/4) Epoch 3, batch 1800, loss[loss=0.2626, simple_loss=0.2893, pruned_loss=0.118, over 13411.00 frames. ], tot_loss[loss=0.3142, simple_loss=0.334, pruned_loss=0.1472, over 2657714.07 frames. ], batch size: 65, lr: 3.52e-02, grad_scale: 16.0 2023-04-16 13:22:53,260 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.362e+02 3.709e+02 4.674e+02 5.794e+02 1.455e+03, threshold=9.349e+02, percent-clipped=1.0 2023-04-16 13:22:59,878 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.35 vs. limit=5.0 2023-04-16 13:23:13,986 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7224.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 13:23:18,384 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 2023-04-16 13:23:22,103 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7234.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 13:23:34,802 INFO [train.py:893] (1/4) Epoch 3, batch 1850, loss[loss=0.2487, simple_loss=0.2831, pruned_loss=0.1071, over 13430.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3319, pruned_loss=0.146, over 2659247.04 frames. ], batch size: 65, lr: 3.51e-02, grad_scale: 16.0 2023-04-16 13:23:38,094 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-16 13:23:49,712 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7265.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:23:58,964 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7275.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:24:09,348 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7285.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:24:20,801 INFO [train.py:893] (1/4) Epoch 3, batch 1900, loss[loss=0.33, simple_loss=0.3393, pruned_loss=0.1604, over 13278.00 frames. ], tot_loss[loss=0.312, simple_loss=0.3315, pruned_loss=0.1462, over 2657369.44 frames. ], batch size: 124, lr: 3.50e-02, grad_scale: 16.0 2023-04-16 13:24:24,137 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.183e+02 4.383e+02 5.266e+02 6.271e+02 1.325e+03, threshold=1.053e+03, percent-clipped=5.0 2023-04-16 13:24:52,253 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7333.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:24:55,552 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7336.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:25:04,505 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1258, 3.6124, 3.3650, 3.5037, 3.4992, 3.8225, 3.5973, 3.5658], device='cuda:1'), covar=tensor([0.0393, 0.0234, 0.0250, 0.0549, 0.0238, 0.0225, 0.0270, 0.0588], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0059, 0.0057, 0.0090, 0.0056, 0.0070, 0.0056, 0.0061], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-16 13:25:08,205 INFO [train.py:893] (1/4) Epoch 3, batch 1950, loss[loss=0.3178, simple_loss=0.3486, pruned_loss=0.1435, over 13538.00 frames. ], tot_loss[loss=0.3108, simple_loss=0.3307, pruned_loss=0.1455, over 2653847.88 frames. ], batch size: 98, lr: 3.49e-02, grad_scale: 16.0 2023-04-16 13:25:12,675 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.9433, 4.9365, 5.1542, 4.8592, 5.3978, 4.7955, 5.3274, 5.3782], device='cuda:1'), covar=tensor([0.0326, 0.0367, 0.0433, 0.0377, 0.0396, 0.0598, 0.0450, 0.0366], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0148, 0.0133, 0.0103, 0.0179, 0.0160, 0.0110, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 13:25:53,551 INFO [train.py:893] (1/4) Epoch 3, batch 2000, loss[loss=0.3249, simple_loss=0.3282, pruned_loss=0.1608, over 13462.00 frames. ], tot_loss[loss=0.3155, simple_loss=0.3345, pruned_loss=0.1482, over 2658747.76 frames. ], batch size: 65, lr: 3.49e-02, grad_scale: 16.0 2023-04-16 13:25:57,274 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.912e+02 4.643e+02 5.664e+02 6.636e+02 1.338e+03, threshold=1.133e+03, percent-clipped=3.0 2023-04-16 13:25:58,201 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-16 13:25:58,316 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7404.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 13:26:00,780 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7407.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:26:27,897 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.71 vs. limit=2.0 2023-04-16 13:26:38,459 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-04-16 13:26:40,416 INFO [train.py:893] (1/4) Epoch 3, batch 2050, loss[loss=0.3229, simple_loss=0.3397, pruned_loss=0.1531, over 13508.00 frames. ], tot_loss[loss=0.3174, simple_loss=0.3362, pruned_loss=0.1493, over 2657215.48 frames. ], batch size: 70, lr: 3.48e-02, grad_scale: 16.0 2023-04-16 13:26:57,063 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7468.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:27:07,690 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 2023-04-16 13:27:24,592 INFO [train.py:893] (1/4) Epoch 3, batch 2100, loss[loss=0.3365, simple_loss=0.348, pruned_loss=0.1625, over 13237.00 frames. ], tot_loss[loss=0.3163, simple_loss=0.3352, pruned_loss=0.1487, over 2655171.61 frames. ], batch size: 117, lr: 3.47e-02, grad_scale: 16.0 2023-04-16 13:27:29,168 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.529e+02 4.232e+02 5.146e+02 5.882e+02 1.040e+03, threshold=1.029e+03, percent-clipped=0.0 2023-04-16 13:27:38,688 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 2023-04-16 13:27:43,598 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7519.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 13:27:43,720 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.4802, 1.3160, 1.3217, 1.1926, 0.9086, 0.6438, 1.3656, 1.6865], device='cuda:1'), covar=tensor([0.0265, 0.0317, 0.0291, 0.0233, 0.0146, 0.0303, 0.0393, 0.0303], device='cuda:1'), in_proj_covar=tensor([0.0031, 0.0030, 0.0032, 0.0025, 0.0029, 0.0028, 0.0031, 0.0033], device='cuda:1'), out_proj_covar=tensor([3.5299e-05, 3.8241e-05, 3.9041e-05, 2.9102e-05, 3.3642e-05, 3.5930e-05, 3.6454e-05, 4.0230e-05], device='cuda:1') 2023-04-16 13:27:58,405 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7534.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 13:28:10,586 INFO [train.py:893] (1/4) Epoch 3, batch 2150, loss[loss=0.2944, simple_loss=0.3313, pruned_loss=0.1288, over 13553.00 frames. ], tot_loss[loss=0.3167, simple_loss=0.336, pruned_loss=0.1487, over 2658299.67 frames. ], batch size: 83, lr: 3.46e-02, grad_scale: 16.0 2023-04-16 13:28:26,892 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7565.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:28:33,809 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.79 vs. limit=2.0 2023-04-16 13:28:40,560 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7582.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 13:28:57,078 INFO [train.py:893] (1/4) Epoch 3, batch 2200, loss[loss=0.3046, simple_loss=0.3379, pruned_loss=0.1356, over 13055.00 frames. ], tot_loss[loss=0.3143, simple_loss=0.3341, pruned_loss=0.1472, over 2656717.33 frames. ], batch size: 142, lr: 3.45e-02, grad_scale: 16.0 2023-04-16 13:29:00,652 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.779e+02 4.630e+02 5.506e+02 6.810e+02 1.281e+03, threshold=1.101e+03, percent-clipped=2.0 2023-04-16 13:29:09,244 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7613.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:29:19,157 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7624.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:29:26,925 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7631.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:29:42,336 INFO [train.py:893] (1/4) Epoch 3, batch 2250, loss[loss=0.2954, simple_loss=0.3242, pruned_loss=0.1334, over 13521.00 frames. ], tot_loss[loss=0.3114, simple_loss=0.3317, pruned_loss=0.1455, over 2651286.33 frames. ], batch size: 91, lr: 3.45e-02, grad_scale: 16.0 2023-04-16 13:30:06,799 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 2023-04-16 13:30:15,331 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7685.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:30:28,758 INFO [train.py:893] (1/4) Epoch 3, batch 2300, loss[loss=0.2912, simple_loss=0.3148, pruned_loss=0.1338, over 13481.00 frames. ], tot_loss[loss=0.3086, simple_loss=0.3297, pruned_loss=0.1437, over 2653762.30 frames. ], batch size: 81, lr: 3.44e-02, grad_scale: 16.0 2023-04-16 13:30:32,071 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.377e+02 3.990e+02 4.633e+02 6.005e+02 9.986e+02, threshold=9.265e+02, percent-clipped=0.0 2023-04-16 13:30:33,111 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7704.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 13:31:09,175 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7743.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:31:13,937 INFO [train.py:893] (1/4) Epoch 3, batch 2350, loss[loss=0.3212, simple_loss=0.3396, pruned_loss=0.1514, over 13565.00 frames. ], tot_loss[loss=0.3078, simple_loss=0.3293, pruned_loss=0.1432, over 2658109.57 frames. ], batch size: 89, lr: 3.43e-02, grad_scale: 16.0 2023-04-16 13:31:17,273 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7752.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 13:31:27,703 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7763.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:31:34,207 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-16 13:32:00,156 INFO [train.py:893] (1/4) Epoch 3, batch 2400, loss[loss=0.371, simple_loss=0.3784, pruned_loss=0.1818, over 13382.00 frames. ], tot_loss[loss=0.3065, simple_loss=0.3281, pruned_loss=0.1424, over 2661473.95 frames. ], batch size: 109, lr: 3.42e-02, grad_scale: 16.0 2023-04-16 13:32:03,381 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=7802.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:32:03,855 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.317e+02 4.190e+02 5.036e+02 6.396e+02 1.178e+03, threshold=1.007e+03, percent-clipped=5.0 2023-04-16 13:32:05,749 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7804.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:32:10,067 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 2023-04-16 13:32:20,273 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7819.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 13:32:46,644 INFO [train.py:893] (1/4) Epoch 3, batch 2450, loss[loss=0.2763, simple_loss=0.3078, pruned_loss=0.1224, over 13504.00 frames. ], tot_loss[loss=0.3081, simple_loss=0.329, pruned_loss=0.1436, over 2664032.70 frames. ], batch size: 81, lr: 3.41e-02, grad_scale: 16.0 2023-04-16 13:32:53,683 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-04-16 13:32:59,374 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=7863.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:33:02,560 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7867.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:33:17,186 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.4543, 1.5145, 1.3849, 1.4630, 0.8666, 0.8393, 1.3481, 1.6875], device='cuda:1'), covar=tensor([0.0182, 0.0215, 0.0209, 0.0163, 0.0201, 0.0252, 0.0317, 0.0187], device='cuda:1'), in_proj_covar=tensor([0.0033, 0.0030, 0.0035, 0.0028, 0.0033, 0.0030, 0.0034, 0.0035], device='cuda:1'), out_proj_covar=tensor([3.7765e-05, 3.7981e-05, 4.2864e-05, 3.2509e-05, 3.7853e-05, 3.7640e-05, 4.0865e-05, 4.2805e-05], device='cuda:1') 2023-04-16 13:33:31,683 INFO [train.py:893] (1/4) Epoch 3, batch 2500, loss[loss=0.2772, simple_loss=0.2877, pruned_loss=0.1333, over 13173.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3283, pruned_loss=0.1431, over 2661082.14 frames. ], batch size: 58, lr: 3.41e-02, grad_scale: 16.0 2023-04-16 13:33:35,662 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.517e+02 4.233e+02 5.484e+02 6.792e+02 1.086e+03, threshold=1.097e+03, percent-clipped=1.0 2023-04-16 13:33:43,706 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-04-16 13:34:00,749 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=7931.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:34:17,472 INFO [train.py:893] (1/4) Epoch 3, batch 2550, loss[loss=0.345, simple_loss=0.3536, pruned_loss=0.1682, over 13387.00 frames. ], tot_loss[loss=0.3059, simple_loss=0.3273, pruned_loss=0.1423, over 2663272.94 frames. ], batch size: 113, lr: 3.40e-02, grad_scale: 16.0 2023-04-16 13:34:38,364 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-16 13:34:44,988 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=7979.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:34:45,834 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=7980.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:35:01,882 INFO [train.py:893] (1/4) Epoch 3, batch 2600, loss[loss=0.3168, simple_loss=0.3229, pruned_loss=0.1553, over 13357.00 frames. ], tot_loss[loss=0.3073, simple_loss=0.3282, pruned_loss=0.1432, over 2665636.97 frames. ], batch size: 62, lr: 3.39e-02, grad_scale: 16.0 2023-04-16 13:35:09,081 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.046e+02 4.385e+02 5.353e+02 6.768e+02 2.035e+03, threshold=1.071e+03, percent-clipped=3.0 2023-04-16 13:35:45,053 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0 2023-04-16 13:35:46,770 INFO [train.py:893] (1/4) Epoch 3, batch 2650, loss[loss=0.332, simple_loss=0.3469, pruned_loss=0.1585, over 13137.00 frames. ], tot_loss[loss=0.3104, simple_loss=0.3301, pruned_loss=0.1454, over 2650805.67 frames. ], batch size: 142, lr: 3.38e-02, grad_scale: 16.0 2023-04-16 13:35:57,889 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8063.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:36:42,752 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 13:36:53,356 INFO [train.py:893] (1/4) Epoch 4, batch 0, loss[loss=0.2893, simple_loss=0.3186, pruned_loss=0.13, over 13485.00 frames. ], tot_loss[loss=0.2893, simple_loss=0.3186, pruned_loss=0.13, over 13485.00 frames. ], batch size: 81, lr: 3.16e-02, grad_scale: 16.0 2023-04-16 13:36:53,356 INFO [train.py:918] (1/4) Computing validation loss 2023-04-16 13:37:12,269 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.6372, 1.5718, 1.3776, 1.3494, 1.1970, 1.3622, 1.6705, 1.9808], device='cuda:1'), covar=tensor([0.0231, 0.0324, 0.0395, 0.0404, 0.0194, 0.0267, 0.0488, 0.0310], device='cuda:1'), in_proj_covar=tensor([0.0033, 0.0029, 0.0035, 0.0029, 0.0032, 0.0029, 0.0033, 0.0036], device='cuda:1'), out_proj_covar=tensor([3.8529e-05, 3.7272e-05, 4.3199e-05, 3.3824e-05, 3.6622e-05, 3.5343e-05, 3.9286e-05, 4.3801e-05], device='cuda:1') 2023-04-16 13:37:16,201 INFO [train.py:927] (1/4) Epoch 4, validation: loss=0.2213, simple_loss=0.2628, pruned_loss=0.08994, over 2446609.00 frames. 2023-04-16 13:37:16,203 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12761MB 2023-04-16 13:37:17,250 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8099.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:37:20,319 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.901e+02 4.145e+02 5.208e+02 6.723e+02 1.537e+03, threshold=1.042e+03, percent-clipped=4.0 2023-04-16 13:37:27,227 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8111.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:37:31,247 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.9990, 4.5406, 4.5167, 4.4454, 4.2309, 4.3652, 4.9497, 4.4657], device='cuda:1'), covar=tensor([0.0853, 0.0740, 0.2204, 0.2598, 0.0717, 0.1172, 0.0836, 0.0840], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0181, 0.0254, 0.0255, 0.0131, 0.0207, 0.0227, 0.0164], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 13:38:02,420 INFO [train.py:893] (1/4) Epoch 4, batch 50, loss[loss=0.3066, simple_loss=0.3183, pruned_loss=0.1474, over 11789.00 frames. ], tot_loss[loss=0.2939, simple_loss=0.3145, pruned_loss=0.1367, over 596029.90 frames. ], batch size: 157, lr: 3.15e-02, grad_scale: 16.0 2023-04-16 13:38:06,715 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.8057, 2.2123, 1.7080, 1.4022, 1.5532, 2.2310, 1.8259, 2.1091], device='cuda:1'), covar=tensor([0.0406, 0.0385, 0.1271, 0.0982, 0.0534, 0.0203, 0.0582, 0.0309], device='cuda:1'), in_proj_covar=tensor([0.0057, 0.0049, 0.0051, 0.0072, 0.0046, 0.0049, 0.0064, 0.0050], device='cuda:1'), out_proj_covar=tensor([5.1475e-05, 4.1152e-05, 5.0725e-05, 6.6636e-05, 5.1568e-05, 4.2866e-05, 5.9724e-05, 4.0768e-05], device='cuda:1') 2023-04-16 13:38:07,491 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8620, 4.5283, 4.1804, 4.0001, 4.1191, 3.9813, 4.4685, 4.5697], device='cuda:1'), covar=tensor([0.0314, 0.0206, 0.0299, 0.0390, 0.0317, 0.0365, 0.0327, 0.0199], device='cuda:1'), in_proj_covar=tensor([0.0089, 0.0115, 0.0084, 0.0111, 0.0079, 0.0103, 0.0085, 0.0085], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 13:38:12,397 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8158.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:38:26,303 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-16 13:38:26,304 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-16 13:38:26,304 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-16 13:38:26,319 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-16 13:38:27,842 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-16 13:38:27,856 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-16 13:38:27,873 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-16 13:38:48,456 INFO [train.py:893] (1/4) Epoch 4, batch 100, loss[loss=0.344, simple_loss=0.3555, pruned_loss=0.1663, over 13071.00 frames. ], tot_loss[loss=0.3031, simple_loss=0.3216, pruned_loss=0.1424, over 1053268.59 frames. ], batch size: 142, lr: 3.14e-02, grad_scale: 16.0 2023-04-16 13:38:52,771 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-04-16 13:38:53,071 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.887e+02 4.151e+02 4.936e+02 5.923e+02 1.224e+03, threshold=9.871e+02, percent-clipped=1.0 2023-04-16 13:39:35,969 INFO [train.py:893] (1/4) Epoch 4, batch 150, loss[loss=0.2987, simple_loss=0.3248, pruned_loss=0.1363, over 13533.00 frames. ], tot_loss[loss=0.3044, simple_loss=0.3228, pruned_loss=0.143, over 1391048.82 frames. ], batch size: 98, lr: 3.14e-02, grad_scale: 8.0 2023-04-16 13:39:50,130 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.72 vs. limit=5.0 2023-04-16 13:39:51,415 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8265.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:40:06,105 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8280.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:40:16,033 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.2495, 4.3104, 4.1422, 3.4820, 3.6834, 2.4892, 4.4468, 2.9082], device='cuda:1'), covar=tensor([0.0998, 0.0217, 0.0230, 0.0579, 0.0359, 0.1930, 0.0153, 0.2133], device='cuda:1'), in_proj_covar=tensor([0.0201, 0.0141, 0.0156, 0.0186, 0.0152, 0.0198, 0.0113, 0.0245], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0003], device='cuda:1') 2023-04-16 13:40:21,458 INFO [train.py:893] (1/4) Epoch 4, batch 200, loss[loss=0.3041, simple_loss=0.316, pruned_loss=0.1461, over 13429.00 frames. ], tot_loss[loss=0.3068, simple_loss=0.3262, pruned_loss=0.1437, over 1668299.93 frames. ], batch size: 65, lr: 3.13e-02, grad_scale: 8.0 2023-04-16 13:40:26,453 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.066e+02 4.274e+02 5.241e+02 6.308e+02 1.197e+03, threshold=1.048e+03, percent-clipped=2.0 2023-04-16 13:40:47,683 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8326.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:40:49,217 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8328.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:41:08,862 INFO [train.py:893] (1/4) Epoch 4, batch 250, loss[loss=0.2712, simple_loss=0.2993, pruned_loss=0.1215, over 13115.00 frames. ], tot_loss[loss=0.3072, simple_loss=0.3275, pruned_loss=0.1435, over 1890701.84 frames. ], batch size: 142, lr: 3.12e-02, grad_scale: 8.0 2023-04-16 13:41:54,052 INFO [train.py:893] (1/4) Epoch 4, batch 300, loss[loss=0.2716, simple_loss=0.2942, pruned_loss=0.1245, over 13373.00 frames. ], tot_loss[loss=0.3058, simple_loss=0.327, pruned_loss=0.1423, over 2053315.18 frames. ], batch size: 67, lr: 3.12e-02, grad_scale: 8.0 2023-04-16 13:41:55,783 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8399.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:42:01,461 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.173e+02 4.573e+02 5.288e+02 6.511e+02 1.234e+03, threshold=1.058e+03, percent-clipped=2.0 2023-04-16 13:42:41,082 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8447.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:42:41,753 INFO [train.py:893] (1/4) Epoch 4, batch 350, loss[loss=0.285, simple_loss=0.2984, pruned_loss=0.1358, over 13214.00 frames. ], tot_loss[loss=0.3087, simple_loss=0.3291, pruned_loss=0.1441, over 2188058.05 frames. ], batch size: 58, lr: 3.11e-02, grad_scale: 8.0 2023-04-16 13:42:47,979 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7402, 2.2446, 2.1168, 4.0715, 3.4989, 3.8602, 2.8369, 2.0909], device='cuda:1'), covar=tensor([0.0177, 0.1372, 0.1132, 0.0049, 0.0246, 0.0127, 0.0554, 0.1387], device='cuda:1'), in_proj_covar=tensor([0.0058, 0.0114, 0.0100, 0.0058, 0.0058, 0.0054, 0.0090, 0.0104], device='cuda:1'), out_proj_covar=tensor([9.3246e-05, 1.5931e-04, 1.4204e-04, 8.3896e-05, 9.4057e-05, 8.0040e-05, 1.2731e-04, 1.4498e-04], device='cuda:1') 2023-04-16 13:42:49,574 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3455, 4.2432, 4.3975, 3.9376, 4.8935, 4.4157, 4.4510, 4.7654], device='cuda:1'), covar=tensor([0.0178, 0.0129, 0.0139, 0.0451, 0.0127, 0.0122, 0.0116, 0.0076], device='cuda:1'), in_proj_covar=tensor([0.0042, 0.0025, 0.0037, 0.0054, 0.0043, 0.0036, 0.0038, 0.0030], device='cuda:1'), out_proj_covar=tensor([1.3157e-04, 9.2460e-05, 1.2087e-04, 1.7494e-04, 1.5678e-04, 1.2227e-04, 1.2528e-04, 1.0335e-04], device='cuda:1') 2023-04-16 13:42:50,422 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8458.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:43:27,837 INFO [train.py:893] (1/4) Epoch 4, batch 400, loss[loss=0.2525, simple_loss=0.2829, pruned_loss=0.111, over 13342.00 frames. ], tot_loss[loss=0.3074, simple_loss=0.3288, pruned_loss=0.143, over 2290287.05 frames. ], batch size: 67, lr: 3.10e-02, grad_scale: 8.0 2023-04-16 13:43:32,889 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.623e+02 4.353e+02 5.245e+02 6.268e+02 1.002e+03, threshold=1.049e+03, percent-clipped=0.0 2023-04-16 13:43:35,408 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8506.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:43:43,086 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.4874, 1.6219, 1.1827, 1.4815, 1.0831, 1.1764, 1.4667, 1.2242], device='cuda:1'), covar=tensor([0.0213, 0.0270, 0.0325, 0.0331, 0.0208, 0.0191, 0.0408, 0.0450], device='cuda:1'), in_proj_covar=tensor([0.0034, 0.0032, 0.0036, 0.0030, 0.0032, 0.0031, 0.0033, 0.0036], device='cuda:1'), out_proj_covar=tensor([3.9634e-05, 3.9536e-05, 4.4336e-05, 3.6538e-05, 3.6092e-05, 3.8133e-05, 3.9951e-05, 4.3996e-05], device='cuda:1') 2023-04-16 13:44:03,571 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-16 13:44:12,225 INFO [train.py:893] (1/4) Epoch 4, batch 450, loss[loss=0.2607, simple_loss=0.2898, pruned_loss=0.1158, over 13508.00 frames. ], tot_loss[loss=0.3084, simple_loss=0.33, pruned_loss=0.1434, over 2375632.87 frames. ], batch size: 70, lr: 3.10e-02, grad_scale: 8.0 2023-04-16 13:44:37,116 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-16 13:44:58,354 INFO [train.py:893] (1/4) Epoch 4, batch 500, loss[loss=0.2948, simple_loss=0.3251, pruned_loss=0.1323, over 13438.00 frames. ], tot_loss[loss=0.3066, simple_loss=0.329, pruned_loss=0.1421, over 2443229.86 frames. ], batch size: 95, lr: 3.09e-02, grad_scale: 8.0 2023-04-16 13:45:04,385 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.634e+02 4.142e+02 5.081e+02 6.274e+02 1.184e+03, threshold=1.016e+03, percent-clipped=3.0 2023-04-16 13:45:14,755 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.2145, 4.6316, 4.4691, 4.4871, 4.5749, 4.5300, 5.0775, 4.5475], device='cuda:1'), covar=tensor([0.0766, 0.0889, 0.2472, 0.2847, 0.0702, 0.1312, 0.0969, 0.0985], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0189, 0.0269, 0.0270, 0.0142, 0.0219, 0.0242, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 13:45:20,080 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=8621.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:45:43,355 INFO [train.py:893] (1/4) Epoch 4, batch 550, loss[loss=0.2971, simple_loss=0.3163, pruned_loss=0.1389, over 13536.00 frames. ], tot_loss[loss=0.3064, simple_loss=0.3293, pruned_loss=0.1417, over 2491755.08 frames. ], batch size: 76, lr: 3.08e-02, grad_scale: 8.0 2023-04-16 13:45:47,787 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3272, 4.2145, 3.8492, 3.4510, 3.1897, 2.4531, 4.2902, 2.7225], device='cuda:1'), covar=tensor([0.0881, 0.0202, 0.0316, 0.0689, 0.0545, 0.2043, 0.0143, 0.2425], device='cuda:1'), in_proj_covar=tensor([0.0206, 0.0144, 0.0160, 0.0190, 0.0156, 0.0200, 0.0115, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0003], device='cuda:1') 2023-04-16 13:45:52,598 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2023-04-16 13:46:00,551 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3232, 3.4468, 3.0374, 3.0673, 2.7879, 1.5218, 3.2884, 1.8877], device='cuda:1'), covar=tensor([0.0292, 0.0274, 0.0245, 0.0267, 0.0597, 0.0950, 0.0547, 0.1050], device='cuda:1'), in_proj_covar=tensor([0.0066, 0.0050, 0.0061, 0.0057, 0.0080, 0.0083, 0.0049, 0.0083], device='cuda:1'), out_proj_covar=tensor([9.9685e-05, 8.2925e-05, 9.0499e-05, 8.5474e-05, 1.2392e-04, 1.2104e-04, 8.3464e-05, 1.2071e-04], device='cuda:1') 2023-04-16 13:46:04,641 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.4961, 1.7110, 3.6476, 3.3340, 3.4833, 3.0638, 3.5042, 2.2513], device='cuda:1'), covar=tensor([0.2361, 0.2210, 0.0098, 0.0399, 0.0323, 0.0425, 0.0142, 0.1277], device='cuda:1'), in_proj_covar=tensor([0.0157, 0.0161, 0.0071, 0.0078, 0.0082, 0.0106, 0.0076, 0.0125], device='cuda:1'), out_proj_covar=tensor([1.6209e-04, 1.7120e-04, 7.8983e-05, 9.4421e-05, 9.6253e-05, 1.1611e-04, 8.9789e-05, 1.3473e-04], device='cuda:1') 2023-04-16 13:46:29,271 INFO [train.py:893] (1/4) Epoch 4, batch 600, loss[loss=0.2803, simple_loss=0.3122, pruned_loss=0.1242, over 13523.00 frames. ], tot_loss[loss=0.3043, simple_loss=0.3268, pruned_loss=0.1409, over 2531066.69 frames. ], batch size: 91, lr: 3.08e-02, grad_scale: 8.0 2023-04-16 13:46:33,702 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7602, 2.9420, 2.7172, 3.4050, 4.3070, 3.4098, 4.1867, 3.6662], device='cuda:1'), covar=tensor([0.0126, 0.0310, 0.0536, 0.0123, 0.0097, 0.0223, 0.0127, 0.0104], device='cuda:1'), in_proj_covar=tensor([0.0032, 0.0038, 0.0045, 0.0035, 0.0023, 0.0036, 0.0023, 0.0029], device='cuda:1'), out_proj_covar=tensor([1.1198e-04, 1.3423e-04, 1.5004e-04, 1.2307e-04, 8.1902e-05, 1.2830e-04, 8.0684e-05, 1.0444e-04], device='cuda:1') 2023-04-16 13:46:34,198 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.491e+02 4.071e+02 5.189e+02 6.724e+02 1.349e+03, threshold=1.038e+03, percent-clipped=5.0 2023-04-16 13:47:09,428 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3436, 3.8194, 3.8097, 4.2473, 1.8479, 3.2015, 3.7935, 1.9412], device='cuda:1'), covar=tensor([0.0089, 0.0368, 0.0318, 0.0092, 0.2100, 0.0588, 0.0469, 0.2416], device='cuda:1'), in_proj_covar=tensor([0.0062, 0.0104, 0.0116, 0.0066, 0.0151, 0.0123, 0.0118, 0.0152], device='cuda:1'), out_proj_covar=tensor([8.9320e-05, 1.5155e-04, 1.6088e-04, 1.0009e-04, 1.9675e-04, 1.6718e-04, 1.6727e-04, 2.0159e-04], device='cuda:1') 2023-04-16 13:47:14,843 INFO [train.py:893] (1/4) Epoch 4, batch 650, loss[loss=0.3087, simple_loss=0.3333, pruned_loss=0.1421, over 13492.00 frames. ], tot_loss[loss=0.3029, simple_loss=0.3258, pruned_loss=0.14, over 2564545.25 frames. ], batch size: 79, lr: 3.07e-02, grad_scale: 8.0 2023-04-16 13:47:24,145 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4119, 3.3143, 3.4263, 2.9378, 3.7524, 3.2799, 3.4274, 3.6541], device='cuda:1'), covar=tensor([0.0186, 0.0112, 0.0134, 0.0577, 0.0114, 0.0162, 0.0127, 0.0107], device='cuda:1'), in_proj_covar=tensor([0.0044, 0.0024, 0.0039, 0.0055, 0.0045, 0.0039, 0.0039, 0.0029], device='cuda:1'), out_proj_covar=tensor([1.4276e-04, 9.2251e-05, 1.3035e-04, 1.8193e-04, 1.6566e-04, 1.3431e-04, 1.2812e-04, 1.0424e-04], device='cuda:1') 2023-04-16 13:47:24,267 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1064, 3.7459, 3.4753, 3.2196, 3.1212, 2.0322, 3.8770, 2.3166], device='cuda:1'), covar=tensor([0.0619, 0.0179, 0.0256, 0.0460, 0.0318, 0.1981, 0.0107, 0.2066], device='cuda:1'), in_proj_covar=tensor([0.0208, 0.0145, 0.0162, 0.0192, 0.0153, 0.0199, 0.0114, 0.0249], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0003], device='cuda:1') 2023-04-16 13:47:25,903 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8760.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:48:00,253 INFO [train.py:893] (1/4) Epoch 4, batch 700, loss[loss=0.303, simple_loss=0.334, pruned_loss=0.136, over 13377.00 frames. ], tot_loss[loss=0.3003, simple_loss=0.3239, pruned_loss=0.1384, over 2584578.65 frames. ], batch size: 113, lr: 3.06e-02, grad_scale: 8.0 2023-04-16 13:48:02,570 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7819, 3.0457, 4.2317, 4.0390, 4.3757, 3.9451, 4.1992, 3.3990], device='cuda:1'), covar=tensor([0.1426, 0.1245, 0.0077, 0.0170, 0.0113, 0.0201, 0.0107, 0.0595], device='cuda:1'), in_proj_covar=tensor([0.0156, 0.0156, 0.0070, 0.0078, 0.0080, 0.0104, 0.0074, 0.0121], device='cuda:1'), out_proj_covar=tensor([1.6172e-04, 1.6605e-04, 7.6627e-05, 9.4157e-05, 9.4688e-05, 1.1428e-04, 8.8284e-05, 1.3050e-04], device='cuda:1') 2023-04-16 13:48:05,598 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.391e+02 3.986e+02 4.492e+02 5.620e+02 1.146e+03, threshold=8.983e+02, percent-clipped=1.0 2023-04-16 13:48:21,792 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8821.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:48:29,293 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 2023-04-16 13:48:46,631 INFO [train.py:893] (1/4) Epoch 4, batch 750, loss[loss=0.2892, simple_loss=0.3174, pruned_loss=0.1305, over 13567.00 frames. ], tot_loss[loss=0.3015, simple_loss=0.3242, pruned_loss=0.1394, over 2601677.46 frames. ], batch size: 89, lr: 3.06e-02, grad_scale: 8.0 2023-04-16 13:49:30,922 INFO [train.py:893] (1/4) Epoch 4, batch 800, loss[loss=0.3083, simple_loss=0.3176, pruned_loss=0.1495, over 13180.00 frames. ], tot_loss[loss=0.3037, simple_loss=0.3267, pruned_loss=0.1404, over 2619410.72 frames. ], batch size: 58, lr: 3.05e-02, grad_scale: 8.0 2023-04-16 13:49:36,354 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.977e+02 4.254e+02 5.779e+02 7.290e+02 1.699e+03, threshold=1.156e+03, percent-clipped=10.0 2023-04-16 13:49:52,863 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=8921.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:49:57,963 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8927.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:50:17,241 INFO [train.py:893] (1/4) Epoch 4, batch 850, loss[loss=0.2787, simple_loss=0.3155, pruned_loss=0.121, over 13510.00 frames. ], tot_loss[loss=0.3042, simple_loss=0.3275, pruned_loss=0.1405, over 2626684.61 frames. ], batch size: 91, lr: 3.04e-02, grad_scale: 8.0 2023-04-16 13:50:37,350 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=8969.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:50:52,267 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=8986.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:50:53,929 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=8988.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:51:03,290 INFO [train.py:893] (1/4) Epoch 4, batch 900, loss[loss=0.311, simple_loss=0.3329, pruned_loss=0.1446, over 13523.00 frames. ], tot_loss[loss=0.3039, simple_loss=0.3271, pruned_loss=0.1404, over 2638450.30 frames. ], batch size: 76, lr: 3.04e-02, grad_scale: 8.0 2023-04-16 13:51:09,133 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.569e+02 4.214e+02 4.860e+02 6.080e+02 1.245e+03, threshold=9.720e+02, percent-clipped=1.0 2023-04-16 13:51:16,754 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9013.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:51:34,948 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-16 13:51:47,531 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9047.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:51:48,068 INFO [train.py:893] (1/4) Epoch 4, batch 950, loss[loss=0.2828, simple_loss=0.3062, pruned_loss=0.1296, over 13534.00 frames. ], tot_loss[loss=0.3034, simple_loss=0.326, pruned_loss=0.1404, over 2647464.57 frames. ], batch size: 98, lr: 3.03e-02, grad_scale: 8.0 2023-04-16 13:51:54,989 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.93 vs. limit=2.0 2023-04-16 13:52:06,973 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9853, 2.3681, 2.0283, 1.2913, 1.7176, 1.7090, 1.7564, 2.1996], device='cuda:1'), covar=tensor([0.0489, 0.0730, 0.0859, 0.0985, 0.0483, 0.0154, 0.0530, 0.0450], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0054, 0.0055, 0.0077, 0.0049, 0.0053, 0.0066, 0.0056], device='cuda:1'), out_proj_covar=tensor([5.8656e-05, 4.6360e-05, 5.2859e-05, 7.1783e-05, 5.3775e-05, 4.4936e-05, 6.1635e-05, 4.5162e-05], device='cuda:1') 2023-04-16 13:52:12,709 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9074.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:52:34,808 INFO [train.py:893] (1/4) Epoch 4, batch 1000, loss[loss=0.3226, simple_loss=0.3405, pruned_loss=0.1524, over 13520.00 frames. ], tot_loss[loss=0.3009, simple_loss=0.3233, pruned_loss=0.1393, over 2648242.03 frames. ], batch size: 76, lr: 3.02e-02, grad_scale: 8.0 2023-04-16 13:52:39,604 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.434e+02 3.856e+02 4.710e+02 6.550e+02 2.342e+03, threshold=9.421e+02, percent-clipped=6.0 2023-04-16 13:52:50,582 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9116.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:52:56,337 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4658, 4.2958, 4.4222, 4.0614, 4.9090, 4.5177, 4.4840, 4.8130], device='cuda:1'), covar=tensor([0.0138, 0.0126, 0.0114, 0.0400, 0.0116, 0.0093, 0.0119, 0.0088], device='cuda:1'), in_proj_covar=tensor([0.0044, 0.0024, 0.0039, 0.0055, 0.0045, 0.0037, 0.0038, 0.0030], device='cuda:1'), out_proj_covar=tensor([1.4291e-04, 9.4806e-05, 1.3286e-04, 1.8775e-04, 1.7563e-04, 1.3163e-04, 1.3120e-04, 1.0899e-04], device='cuda:1') 2023-04-16 13:53:19,570 INFO [train.py:893] (1/4) Epoch 4, batch 1050, loss[loss=0.3029, simple_loss=0.3309, pruned_loss=0.1375, over 13364.00 frames. ], tot_loss[loss=0.2967, simple_loss=0.3203, pruned_loss=0.1365, over 2654248.81 frames. ], batch size: 113, lr: 3.02e-02, grad_scale: 8.0 2023-04-16 13:53:39,142 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.4299, 1.4991, 3.2699, 3.3339, 3.3209, 2.6176, 2.9894, 2.2774], device='cuda:1'), covar=tensor([0.2626, 0.2686, 0.0162, 0.0386, 0.0246, 0.0760, 0.0191, 0.1337], device='cuda:1'), in_proj_covar=tensor([0.0171, 0.0171, 0.0076, 0.0082, 0.0085, 0.0115, 0.0080, 0.0133], device='cuda:1'), out_proj_covar=tensor([1.7840e-04, 1.8228e-04, 8.4286e-05, 1.0067e-04, 1.0164e-04, 1.2692e-04, 9.5636e-05, 1.4367e-04], device='cuda:1') 2023-04-16 13:54:04,169 INFO [train.py:893] (1/4) Epoch 4, batch 1100, loss[loss=0.3163, simple_loss=0.3417, pruned_loss=0.1454, over 13045.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.322, pruned_loss=0.1366, over 2660553.46 frames. ], batch size: 142, lr: 3.01e-02, grad_scale: 8.0 2023-04-16 13:54:09,399 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.293e+02 4.489e+02 5.675e+02 6.902e+02 1.290e+03, threshold=1.135e+03, percent-clipped=12.0 2023-04-16 13:54:22,660 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.42 vs. limit=2.0 2023-04-16 13:54:49,878 INFO [train.py:893] (1/4) Epoch 4, batch 1150, loss[loss=0.2806, simple_loss=0.2968, pruned_loss=0.1322, over 13376.00 frames. ], tot_loss[loss=0.2956, simple_loss=0.3209, pruned_loss=0.1351, over 2655066.31 frames. ], batch size: 62, lr: 3.00e-02, grad_scale: 8.0 2023-04-16 13:55:22,780 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9283.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:55:35,808 INFO [train.py:893] (1/4) Epoch 4, batch 1200, loss[loss=0.3148, simple_loss=0.3339, pruned_loss=0.1478, over 11686.00 frames. ], tot_loss[loss=0.295, simple_loss=0.3209, pruned_loss=0.1346, over 2654059.51 frames. ], batch size: 158, lr: 3.00e-02, grad_scale: 8.0 2023-04-16 13:55:41,982 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.867e+02 4.129e+02 4.847e+02 6.110e+02 1.320e+03, threshold=9.693e+02, percent-clipped=2.0 2023-04-16 13:56:02,691 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-16 13:56:14,817 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-16 13:56:17,322 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9342.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:56:22,747 INFO [train.py:893] (1/4) Epoch 4, batch 1250, loss[loss=0.3431, simple_loss=0.3549, pruned_loss=0.1656, over 13061.00 frames. ], tot_loss[loss=0.2968, simple_loss=0.322, pruned_loss=0.1358, over 2656209.24 frames. ], batch size: 142, lr: 2.99e-02, grad_scale: 8.0 2023-04-16 13:56:40,946 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3980, 3.9139, 3.6769, 4.4532, 2.0336, 3.2118, 3.8097, 1.9447], device='cuda:1'), covar=tensor([0.0079, 0.0308, 0.0486, 0.0095, 0.1825, 0.0632, 0.0375, 0.2458], device='cuda:1'), in_proj_covar=tensor([0.0069, 0.0111, 0.0128, 0.0070, 0.0158, 0.0133, 0.0128, 0.0164], device='cuda:1'), out_proj_covar=tensor([9.9195e-05, 1.6689e-04, 1.7860e-04, 1.0834e-04, 2.0944e-04, 1.8360e-04, 1.8223e-04, 2.2073e-04], device='cuda:1') 2023-04-16 13:56:41,606 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9369.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:56:50,396 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4878, 3.4659, 3.5975, 3.1161, 3.9837, 3.5669, 3.7858, 4.0039], device='cuda:1'), covar=tensor([0.0245, 0.0127, 0.0145, 0.0591, 0.0140, 0.0163, 0.0118, 0.0102], device='cuda:1'), in_proj_covar=tensor([0.0045, 0.0025, 0.0039, 0.0057, 0.0047, 0.0039, 0.0039, 0.0031], device='cuda:1'), out_proj_covar=tensor([1.4916e-04, 9.9241e-05, 1.3556e-04, 1.9474e-04, 1.8489e-04, 1.4111e-04, 1.3481e-04, 1.1247e-04], device='cuda:1') 2023-04-16 13:57:08,259 INFO [train.py:893] (1/4) Epoch 4, batch 1300, loss[loss=0.2929, simple_loss=0.3185, pruned_loss=0.1336, over 13524.00 frames. ], tot_loss[loss=0.2976, simple_loss=0.3228, pruned_loss=0.1362, over 2658636.20 frames. ], batch size: 72, lr: 2.99e-02, grad_scale: 8.0 2023-04-16 13:57:14,238 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.514e+02 4.181e+02 5.160e+02 6.106e+02 1.309e+03, threshold=1.032e+03, percent-clipped=5.0 2023-04-16 13:57:23,478 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9414.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:57:25,172 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9416.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:57:33,597 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.68 vs. limit=2.0 2023-04-16 13:57:48,025 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-16 13:57:54,157 INFO [train.py:893] (1/4) Epoch 4, batch 1350, loss[loss=0.3107, simple_loss=0.34, pruned_loss=0.1407, over 13448.00 frames. ], tot_loss[loss=0.2989, simple_loss=0.324, pruned_loss=0.1369, over 2659794.03 frames. ], batch size: 103, lr: 2.98e-02, grad_scale: 8.0 2023-04-16 13:58:08,933 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9464.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:58:18,779 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9475.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:58:26,073 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6340, 2.7975, 2.2631, 2.4150, 2.4301, 1.5281, 2.6799, 1.5489], device='cuda:1'), covar=tensor([0.0391, 0.0479, 0.0409, 0.0390, 0.0585, 0.1239, 0.0685, 0.1137], device='cuda:1'), in_proj_covar=tensor([0.0071, 0.0053, 0.0069, 0.0059, 0.0082, 0.0092, 0.0054, 0.0084], device='cuda:1'), out_proj_covar=tensor([1.0921e-04, 8.9303e-05, 1.0425e-04, 9.0439e-05, 1.2812e-04, 1.3710e-04, 9.1413e-05, 1.2465e-04], device='cuda:1') 2023-04-16 13:58:27,257 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.75 vs. limit=2.0 2023-04-16 13:58:39,244 INFO [train.py:893] (1/4) Epoch 4, batch 1400, loss[loss=0.2862, simple_loss=0.3081, pruned_loss=0.1321, over 12039.00 frames. ], tot_loss[loss=0.297, simple_loss=0.3222, pruned_loss=0.1359, over 2656590.42 frames. ], batch size: 157, lr: 2.97e-02, grad_scale: 8.0 2023-04-16 13:58:40,364 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1859, 3.2976, 3.1682, 3.8917, 4.6954, 3.6070, 4.4487, 4.0559], device='cuda:1'), covar=tensor([0.0104, 0.0308, 0.0521, 0.0120, 0.0079, 0.0222, 0.0133, 0.0084], device='cuda:1'), in_proj_covar=tensor([0.0033, 0.0042, 0.0051, 0.0038, 0.0026, 0.0041, 0.0025, 0.0031], device='cuda:1'), out_proj_covar=tensor([1.2144e-04, 1.5660e-04, 1.7766e-04, 1.4075e-04, 9.7770e-05, 1.5148e-04, 9.1731e-05, 1.1459e-04], device='cuda:1') 2023-04-16 13:58:41,906 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.5217, 1.5061, 3.2762, 3.3541, 3.2953, 3.0314, 3.0963, 2.2909], device='cuda:1'), covar=tensor([0.2464, 0.2408, 0.0158, 0.0323, 0.0359, 0.0462, 0.0231, 0.1168], device='cuda:1'), in_proj_covar=tensor([0.0164, 0.0167, 0.0074, 0.0083, 0.0086, 0.0114, 0.0079, 0.0126], device='cuda:1'), out_proj_covar=tensor([1.7304e-04, 1.7813e-04, 8.4499e-05, 1.0172e-04, 1.0288e-04, 1.2816e-04, 9.4561e-05, 1.3848e-04], device='cuda:1') 2023-04-16 13:58:44,926 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.732e+02 4.170e+02 5.056e+02 6.250e+02 1.519e+03, threshold=1.011e+03, percent-clipped=3.0 2023-04-16 13:59:23,884 INFO [train.py:893] (1/4) Epoch 4, batch 1450, loss[loss=0.3304, simple_loss=0.3503, pruned_loss=0.1553, over 13508.00 frames. ], tot_loss[loss=0.2966, simple_loss=0.3216, pruned_loss=0.1358, over 2658030.16 frames. ], batch size: 93, lr: 2.97e-02, grad_scale: 8.0 2023-04-16 13:59:50,524 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1991, 3.9210, 4.0059, 3.5844, 4.7027, 4.1541, 4.3838, 4.7598], device='cuda:1'), covar=tensor([0.0184, 0.0132, 0.0162, 0.0565, 0.0114, 0.0115, 0.0120, 0.0073], device='cuda:1'), in_proj_covar=tensor([0.0045, 0.0025, 0.0040, 0.0059, 0.0047, 0.0040, 0.0040, 0.0032], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-16 13:59:56,216 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9583.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 13:59:57,153 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9584.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:00:10,764 INFO [train.py:893] (1/4) Epoch 4, batch 1500, loss[loss=0.2888, simple_loss=0.3166, pruned_loss=0.1305, over 13029.00 frames. ], tot_loss[loss=0.294, simple_loss=0.3201, pruned_loss=0.134, over 2662299.65 frames. ], batch size: 142, lr: 2.96e-02, grad_scale: 8.0 2023-04-16 14:00:16,014 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.528e+02 4.187e+02 4.847e+02 6.018e+02 1.274e+03, threshold=9.694e+02, percent-clipped=5.0 2023-04-16 14:00:39,217 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.69 vs. limit=2.0 2023-04-16 14:00:41,431 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9631.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:00:50,622 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9642.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:00:53,973 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9645.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:00:56,102 INFO [train.py:893] (1/4) Epoch 4, batch 1550, loss[loss=0.2745, simple_loss=0.2994, pruned_loss=0.1248, over 13548.00 frames. ], tot_loss[loss=0.293, simple_loss=0.3195, pruned_loss=0.1333, over 2660757.49 frames. ], batch size: 76, lr: 2.96e-02, grad_scale: 8.0 2023-04-16 14:01:06,394 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9658.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:01:11,662 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.98 vs. limit=5.0 2023-04-16 14:01:15,503 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=9669.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:01:35,752 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9690.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:01:42,231 INFO [train.py:893] (1/4) Epoch 4, batch 1600, loss[loss=0.2943, simple_loss=0.3237, pruned_loss=0.1325, over 13413.00 frames. ], tot_loss[loss=0.2945, simple_loss=0.3211, pruned_loss=0.1339, over 2661523.05 frames. ], batch size: 95, lr: 2.95e-02, grad_scale: 8.0 2023-04-16 14:01:46,954 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.855e+02 4.347e+02 5.620e+02 6.749e+02 1.079e+03, threshold=1.124e+03, percent-clipped=2.0 2023-04-16 14:01:49,243 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0 2023-04-16 14:01:59,697 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=9717.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:02:02,246 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9719.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:02:12,246 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0304, 2.0248, 2.0167, 3.4677, 3.1699, 3.3779, 2.3274, 1.9698], device='cuda:1'), covar=tensor([0.0355, 0.1221, 0.1189, 0.0094, 0.0181, 0.0108, 0.0668, 0.1344], device='cuda:1'), in_proj_covar=tensor([0.0067, 0.0120, 0.0116, 0.0064, 0.0060, 0.0056, 0.0101, 0.0114], device='cuda:1'), out_proj_covar=tensor([1.1141e-04, 1.7399e-04, 1.6933e-04, 9.7321e-05, 1.0416e-04, 8.7531e-05, 1.4680e-04, 1.6558e-04], device='cuda:1') 2023-04-16 14:02:23,518 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=9743.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:02:27,977 INFO [train.py:893] (1/4) Epoch 4, batch 1650, loss[loss=0.3059, simple_loss=0.332, pruned_loss=0.1399, over 13082.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3206, pruned_loss=0.1328, over 2664947.69 frames. ], batch size: 142, lr: 2.94e-02, grad_scale: 8.0 2023-04-16 14:02:47,863 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9770.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:02:51,599 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.9341, 4.3370, 4.4026, 4.3235, 4.1918, 4.2427, 4.8374, 4.3906], device='cuda:1'), covar=tensor([0.0806, 0.1020, 0.1928, 0.2696, 0.0678, 0.1486, 0.0999, 0.0864], device='cuda:1'), in_proj_covar=tensor([0.0194, 0.0203, 0.0278, 0.0277, 0.0150, 0.0223, 0.0260, 0.0183], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 14:03:12,543 INFO [train.py:893] (1/4) Epoch 4, batch 1700, loss[loss=0.2785, simple_loss=0.3115, pruned_loss=0.1227, over 11986.00 frames. ], tot_loss[loss=0.2931, simple_loss=0.3209, pruned_loss=0.1326, over 2664612.17 frames. ], batch size: 158, lr: 2.94e-02, grad_scale: 8.0 2023-04-16 14:03:18,390 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.356e+02 4.078e+02 4.799e+02 5.634e+02 1.078e+03, threshold=9.597e+02, percent-clipped=0.0 2023-04-16 14:03:18,742 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=9804.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:03:31,892 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-16 14:03:58,747 INFO [train.py:893] (1/4) Epoch 4, batch 1750, loss[loss=0.2829, simple_loss=0.3211, pruned_loss=0.1224, over 13446.00 frames. ], tot_loss[loss=0.2919, simple_loss=0.3198, pruned_loss=0.132, over 2659968.53 frames. ], batch size: 103, lr: 2.93e-02, grad_scale: 8.0 2023-04-16 14:04:14,709 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8274, 3.5480, 3.8294, 3.1484, 4.3026, 3.5969, 3.8161, 4.2615], device='cuda:1'), covar=tensor([0.0206, 0.0137, 0.0144, 0.0629, 0.0134, 0.0195, 0.0156, 0.0109], device='cuda:1'), in_proj_covar=tensor([0.0045, 0.0025, 0.0040, 0.0060, 0.0048, 0.0040, 0.0040, 0.0032], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-16 14:04:43,291 INFO [train.py:893] (1/4) Epoch 4, batch 1800, loss[loss=0.2826, simple_loss=0.3157, pruned_loss=0.1248, over 13070.00 frames. ], tot_loss[loss=0.29, simple_loss=0.318, pruned_loss=0.131, over 2659039.79 frames. ], batch size: 142, lr: 2.93e-02, grad_scale: 8.0 2023-04-16 14:04:49,601 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.550e+02 3.859e+02 4.667e+02 6.022e+02 1.586e+03, threshold=9.334e+02, percent-clipped=8.0 2023-04-16 14:05:22,831 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=9940.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:05:29,076 INFO [train.py:893] (1/4) Epoch 4, batch 1850, loss[loss=0.2859, simple_loss=0.3076, pruned_loss=0.1321, over 13524.00 frames. ], tot_loss[loss=0.2897, simple_loss=0.3182, pruned_loss=0.1306, over 2660568.92 frames. ], batch size: 91, lr: 2.92e-02, grad_scale: 8.0 2023-04-16 14:05:31,583 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-16 14:06:14,474 INFO [train.py:893] (1/4) Epoch 4, batch 1900, loss[loss=0.2656, simple_loss=0.3067, pruned_loss=0.1123, over 13334.00 frames. ], tot_loss[loss=0.2891, simple_loss=0.3169, pruned_loss=0.1307, over 2657371.29 frames. ], batch size: 118, lr: 2.91e-02, grad_scale: 8.0 2023-04-16 14:06:23,225 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.631e+02 4.099e+02 5.413e+02 6.648e+02 1.210e+03, threshold=1.083e+03, percent-clipped=7.0 2023-04-16 14:06:31,756 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10014.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:06:44,813 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.3721, 2.5300, 1.9800, 1.2226, 1.1382, 1.8148, 1.5280, 2.3534], device='cuda:1'), covar=tensor([0.0620, 0.0238, 0.0683, 0.1118, 0.0287, 0.0224, 0.0583, 0.0238], device='cuda:1'), in_proj_covar=tensor([0.0068, 0.0055, 0.0055, 0.0091, 0.0052, 0.0059, 0.0073, 0.0057], device='cuda:1'), out_proj_covar=tensor([5.8766e-05, 4.6598e-05, 5.2363e-05, 8.3772e-05, 5.6284e-05, 4.7582e-05, 6.5467e-05, 4.5327e-05], device='cuda:1') 2023-04-16 14:06:54,567 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10039.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:07:01,575 INFO [train.py:893] (1/4) Epoch 4, batch 1950, loss[loss=0.2969, simple_loss=0.3315, pruned_loss=0.1312, over 13431.00 frames. ], tot_loss[loss=0.2882, simple_loss=0.3162, pruned_loss=0.1302, over 2661163.35 frames. ], batch size: 95, lr: 2.91e-02, grad_scale: 8.0 2023-04-16 14:07:12,187 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9958, 3.6102, 3.8818, 3.3003, 4.5356, 3.9777, 3.9865, 4.4582], device='cuda:1'), covar=tensor([0.0172, 0.0149, 0.0165, 0.0652, 0.0114, 0.0156, 0.0158, 0.0080], device='cuda:1'), in_proj_covar=tensor([0.0046, 0.0025, 0.0041, 0.0061, 0.0048, 0.0039, 0.0040, 0.0032], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-16 14:07:14,743 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10062.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 14:07:21,965 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10070.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:07:47,209 INFO [train.py:893] (1/4) Epoch 4, batch 2000, loss[loss=0.2658, simple_loss=0.3016, pruned_loss=0.115, over 13552.00 frames. ], tot_loss[loss=0.2902, simple_loss=0.3176, pruned_loss=0.1314, over 2660218.59 frames. ], batch size: 72, lr: 2.90e-02, grad_scale: 8.0 2023-04-16 14:07:48,262 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10099.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:07:49,205 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10100.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 14:07:52,258 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.683e+02 4.046e+02 5.081e+02 6.604e+02 1.044e+03, threshold=1.016e+03, percent-clipped=0.0 2023-04-16 14:07:53,096 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-16 14:08:05,281 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10118.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:08:05,422 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5642, 2.5527, 2.7618, 4.1237, 3.6381, 3.9938, 3.1206, 2.6045], device='cuda:1'), covar=tensor([0.0276, 0.1427, 0.0923, 0.0081, 0.0312, 0.0076, 0.0804, 0.1212], device='cuda:1'), in_proj_covar=tensor([0.0071, 0.0127, 0.0116, 0.0064, 0.0064, 0.0054, 0.0106, 0.0117], device='cuda:1'), out_proj_covar=tensor([1.1864e-04, 1.8610e-04, 1.6952e-04, 9.7627e-05, 1.0848e-04, 8.6332e-05, 1.5738e-04, 1.7170e-04], device='cuda:1') 2023-04-16 14:08:10,107 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10123.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 14:08:26,093 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.75 vs. limit=2.0 2023-04-16 14:08:31,933 INFO [train.py:893] (1/4) Epoch 4, batch 2050, loss[loss=0.3074, simple_loss=0.3337, pruned_loss=0.1406, over 13526.00 frames. ], tot_loss[loss=0.2937, simple_loss=0.3202, pruned_loss=0.1336, over 2655042.20 frames. ], batch size: 91, lr: 2.90e-02, grad_scale: 8.0 2023-04-16 14:08:43,455 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4390, 2.2986, 2.6051, 3.8911, 3.4827, 3.8197, 2.7080, 2.2277], device='cuda:1'), covar=tensor([0.0247, 0.1272, 0.0929, 0.0072, 0.0196, 0.0049, 0.0789, 0.1360], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0127, 0.0118, 0.0065, 0.0064, 0.0054, 0.0108, 0.0119], device='cuda:1'), out_proj_covar=tensor([1.1979e-04, 1.8650e-04, 1.7231e-04, 9.8968e-05, 1.0928e-04, 8.5833e-05, 1.6083e-04, 1.7450e-04], device='cuda:1') 2023-04-16 14:08:52,747 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9230, 3.7152, 3.9185, 3.1894, 4.4404, 3.9544, 4.1493, 4.3627], device='cuda:1'), covar=tensor([0.0184, 0.0116, 0.0134, 0.0624, 0.0120, 0.0131, 0.0097, 0.0082], device='cuda:1'), in_proj_covar=tensor([0.0046, 0.0025, 0.0041, 0.0062, 0.0048, 0.0041, 0.0041, 0.0033], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001], device='cuda:1') 2023-04-16 14:09:18,008 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 2023-04-16 14:09:18,321 INFO [train.py:893] (1/4) Epoch 4, batch 2100, loss[loss=0.271, simple_loss=0.3013, pruned_loss=0.1204, over 13342.00 frames. ], tot_loss[loss=0.2928, simple_loss=0.3198, pruned_loss=0.1328, over 2652395.71 frames. ], batch size: 67, lr: 2.89e-02, grad_scale: 8.0 2023-04-16 14:09:23,522 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.173e+02 4.294e+02 5.277e+02 6.464e+02 1.434e+03, threshold=1.055e+03, percent-clipped=4.0 2023-04-16 14:09:52,700 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-04-16 14:09:57,786 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10240.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:10:05,655 INFO [train.py:893] (1/4) Epoch 4, batch 2150, loss[loss=0.2856, simple_loss=0.3197, pruned_loss=0.1258, over 13437.00 frames. ], tot_loss[loss=0.2915, simple_loss=0.3194, pruned_loss=0.1318, over 2653901.29 frames. ], batch size: 95, lr: 2.88e-02, grad_scale: 16.0 2023-04-16 14:10:41,504 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10288.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:10:47,418 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4318, 4.0644, 3.8560, 2.9479, 2.1945, 2.8642, 4.0993, 4.1704], device='cuda:1'), covar=tensor([0.0644, 0.0495, 0.0317, 0.1173, 0.1736, 0.0947, 0.0131, 0.0072], device='cuda:1'), in_proj_covar=tensor([0.0154, 0.0108, 0.0120, 0.0177, 0.0176, 0.0144, 0.0088, 0.0075], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001], device='cuda:1') 2023-04-16 14:10:49,434 INFO [train.py:893] (1/4) Epoch 4, batch 2200, loss[loss=0.2893, simple_loss=0.3205, pruned_loss=0.129, over 13491.00 frames. ], tot_loss[loss=0.2894, simple_loss=0.3178, pruned_loss=0.1305, over 2649218.98 frames. ], batch size: 93, lr: 2.88e-02, grad_scale: 16.0 2023-04-16 14:10:55,069 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.608e+02 3.593e+02 4.262e+02 5.500e+02 1.224e+03, threshold=8.524e+02, percent-clipped=2.0 2023-04-16 14:11:05,670 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10314.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:11:07,316 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.2193, 5.0063, 5.2989, 5.0000, 5.5167, 4.9845, 5.4963, 5.5215], device='cuda:1'), covar=tensor([0.0275, 0.0446, 0.0433, 0.0467, 0.0490, 0.0676, 0.0418, 0.0359], device='cuda:1'), in_proj_covar=tensor([0.0130, 0.0158, 0.0145, 0.0109, 0.0201, 0.0175, 0.0126, 0.0127], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 14:11:35,831 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.02 vs. limit=2.0 2023-04-16 14:11:36,202 INFO [train.py:893] (1/4) Epoch 4, batch 2250, loss[loss=0.3008, simple_loss=0.3233, pruned_loss=0.1391, over 13524.00 frames. ], tot_loss[loss=0.2866, simple_loss=0.3155, pruned_loss=0.1289, over 2653039.28 frames. ], batch size: 98, lr: 2.87e-02, grad_scale: 16.0 2023-04-16 14:11:38,234 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7098, 4.4387, 4.2652, 3.2248, 2.5281, 3.1778, 4.4087, 4.5217], device='cuda:1'), covar=tensor([0.0664, 0.0407, 0.0321, 0.1267, 0.1606, 0.0991, 0.0146, 0.0075], device='cuda:1'), in_proj_covar=tensor([0.0156, 0.0107, 0.0119, 0.0175, 0.0175, 0.0143, 0.0089, 0.0074], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0001], device='cuda:1') 2023-04-16 14:11:39,857 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.5318, 2.1716, 1.7223, 1.0959, 1.3108, 1.3527, 1.1525, 1.8809], device='cuda:1'), covar=tensor([0.0567, 0.0323, 0.0951, 0.1217, 0.0210, 0.0189, 0.0674, 0.0321], device='cuda:1'), in_proj_covar=tensor([0.0073, 0.0060, 0.0061, 0.0094, 0.0059, 0.0062, 0.0079, 0.0063], device='cuda:1'), out_proj_covar=tensor([6.3590e-05, 4.9541e-05, 5.6745e-05, 8.7411e-05, 6.2215e-05, 5.0202e-05, 6.9341e-05, 4.9334e-05], device='cuda:1') 2023-04-16 14:11:47,690 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.90 vs. limit=5.0 2023-04-16 14:11:48,028 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10362.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:12:18,718 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10395.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 14:12:21,698 INFO [train.py:893] (1/4) Epoch 4, batch 2300, loss[loss=0.2806, simple_loss=0.3085, pruned_loss=0.1263, over 11778.00 frames. ], tot_loss[loss=0.2842, simple_loss=0.3139, pruned_loss=0.1272, over 2657103.67 frames. ], batch size: 157, lr: 2.87e-02, grad_scale: 16.0 2023-04-16 14:12:22,893 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3908, 4.7554, 4.5838, 4.3863, 4.5086, 4.3065, 4.8215, 4.8189], device='cuda:1'), covar=tensor([0.0198, 0.0210, 0.0176, 0.0313, 0.0233, 0.0307, 0.0216, 0.0175], device='cuda:1'), in_proj_covar=tensor([0.0087, 0.0122, 0.0085, 0.0114, 0.0080, 0.0112, 0.0088, 0.0091], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 14:12:22,931 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10399.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:12:27,575 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.018e+02 3.904e+02 5.068e+02 6.089e+02 1.155e+03, threshold=1.014e+03, percent-clipped=7.0 2023-04-16 14:12:32,704 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10409.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:12:40,242 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10418.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 14:13:06,929 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10447.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:13:07,564 INFO [train.py:893] (1/4) Epoch 4, batch 2350, loss[loss=0.3313, simple_loss=0.3401, pruned_loss=0.1613, over 13203.00 frames. ], tot_loss[loss=0.2841, simple_loss=0.3134, pruned_loss=0.1274, over 2655108.79 frames. ], batch size: 132, lr: 2.86e-02, grad_scale: 16.0 2023-04-16 14:13:12,910 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10454.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:13:28,021 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10470.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:13:29,461 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-16 14:13:52,776 INFO [train.py:893] (1/4) Epoch 4, batch 2400, loss[loss=0.3289, simple_loss=0.3336, pruned_loss=0.1621, over 11934.00 frames. ], tot_loss[loss=0.2845, simple_loss=0.3132, pruned_loss=0.1279, over 2657854.63 frames. ], batch size: 157, lr: 2.86e-02, grad_scale: 16.0 2023-04-16 14:13:56,601 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.68 vs. limit=2.0 2023-04-16 14:13:58,105 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.68 vs. limit=5.0 2023-04-16 14:13:58,432 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.647e+02 3.691e+02 4.650e+02 5.820e+02 1.816e+03, threshold=9.300e+02, percent-clipped=3.0 2023-04-16 14:14:08,803 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10515.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 14:14:12,033 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8077, 4.2097, 3.9248, 3.8445, 3.8975, 3.7762, 4.2907, 4.2541], device='cuda:1'), covar=tensor([0.0266, 0.0260, 0.0243, 0.0342, 0.0345, 0.0361, 0.0279, 0.0240], device='cuda:1'), in_proj_covar=tensor([0.0090, 0.0123, 0.0086, 0.0116, 0.0082, 0.0114, 0.0090, 0.0093], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 14:14:38,403 INFO [train.py:893] (1/4) Epoch 4, batch 2450, loss[loss=0.322, simple_loss=0.3454, pruned_loss=0.1493, over 13258.00 frames. ], tot_loss[loss=0.2857, simple_loss=0.3144, pruned_loss=0.1285, over 2657204.18 frames. ], batch size: 117, lr: 2.85e-02, grad_scale: 16.0 2023-04-16 14:14:51,534 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9512, 3.8751, 3.9372, 3.2135, 4.5276, 4.1191, 4.1619, 4.5117], device='cuda:1'), covar=tensor([0.0193, 0.0111, 0.0123, 0.0682, 0.0116, 0.0124, 0.0130, 0.0086], device='cuda:1'), in_proj_covar=tensor([0.0045, 0.0025, 0.0039, 0.0060, 0.0048, 0.0039, 0.0041, 0.0031], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-16 14:14:51,725 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.75 vs. limit=5.0 2023-04-16 14:15:24,371 INFO [train.py:893] (1/4) Epoch 4, batch 2500, loss[loss=0.255, simple_loss=0.2952, pruned_loss=0.1073, over 13459.00 frames. ], tot_loss[loss=0.2828, simple_loss=0.3123, pruned_loss=0.1267, over 2659114.34 frames. ], batch size: 79, lr: 2.85e-02, grad_scale: 16.0 2023-04-16 14:15:29,857 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0219, 4.0049, 3.9898, 3.1710, 4.4965, 4.1740, 4.1511, 4.4679], device='cuda:1'), covar=tensor([0.0185, 0.0091, 0.0114, 0.0678, 0.0113, 0.0107, 0.0117, 0.0066], device='cuda:1'), in_proj_covar=tensor([0.0046, 0.0025, 0.0039, 0.0061, 0.0048, 0.0039, 0.0042, 0.0031], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001], device='cuda:1') 2023-04-16 14:15:30,368 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.433e+02 3.816e+02 4.383e+02 5.440e+02 1.104e+03, threshold=8.765e+02, percent-clipped=1.0 2023-04-16 14:16:10,380 INFO [train.py:893] (1/4) Epoch 4, batch 2550, loss[loss=0.3028, simple_loss=0.3277, pruned_loss=0.1389, over 13492.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.313, pruned_loss=0.1273, over 2663280.50 frames. ], batch size: 93, lr: 2.84e-02, grad_scale: 16.0 2023-04-16 14:16:18,422 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.79 vs. limit=2.0 2023-04-16 14:16:34,254 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-16 14:16:37,626 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.5909, 5.0171, 5.1014, 5.0200, 4.8752, 4.9829, 5.4990, 5.0515], device='cuda:1'), covar=tensor([0.0816, 0.0944, 0.2023, 0.2814, 0.0626, 0.1336, 0.0830, 0.0714], device='cuda:1'), in_proj_covar=tensor([0.0200, 0.0210, 0.0281, 0.0289, 0.0149, 0.0229, 0.0259, 0.0189], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 14:16:53,628 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10695.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:16:55,856 INFO [train.py:893] (1/4) Epoch 4, batch 2600, loss[loss=0.2657, simple_loss=0.2996, pruned_loss=0.1159, over 13546.00 frames. ], tot_loss[loss=0.2833, simple_loss=0.3124, pruned_loss=0.1271, over 2664928.00 frames. ], batch size: 78, lr: 2.83e-02, grad_scale: 16.0 2023-04-16 14:17:01,527 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.544e+02 3.996e+02 5.390e+02 7.018e+02 1.415e+03, threshold=1.078e+03, percent-clipped=12.0 2023-04-16 14:17:15,043 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=10718.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 14:17:18,314 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 2023-04-16 14:17:34,604 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10743.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:17:35,577 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10744.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:17:38,419 INFO [train.py:893] (1/4) Epoch 4, batch 2650, loss[loss=0.3026, simple_loss=0.3313, pruned_loss=0.1369, over 13518.00 frames. ], tot_loss[loss=0.2844, simple_loss=0.3132, pruned_loss=0.1279, over 2661582.88 frames. ], batch size: 85, lr: 2.83e-02, grad_scale: 16.0 2023-04-16 14:17:46,828 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0568, 4.4873, 2.8645, 4.5957, 4.4729, 2.1226, 3.7080, 2.9098], device='cuda:1'), covar=tensor([0.0314, 0.0389, 0.1501, 0.0065, 0.0211, 0.1827, 0.0619, 0.1965], device='cuda:1'), in_proj_covar=tensor([0.0099, 0.0100, 0.0161, 0.0068, 0.0097, 0.0144, 0.0124, 0.0158], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-16 14:17:51,219 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10765.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:17:51,938 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=10766.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 14:17:55,689 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.7389, 4.5711, 4.9719, 4.7207, 5.0312, 4.5593, 5.0471, 5.0906], device='cuda:1'), covar=tensor([0.0294, 0.0457, 0.0438, 0.0341, 0.0552, 0.0621, 0.0593, 0.0428], device='cuda:1'), in_proj_covar=tensor([0.0135, 0.0165, 0.0149, 0.0109, 0.0212, 0.0184, 0.0129, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 14:17:58,052 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10774.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:18:08,666 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.74 vs. limit=5.0 2023-04-16 14:18:34,663 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 14:18:44,291 INFO [train.py:893] (1/4) Epoch 5, batch 0, loss[loss=0.3071, simple_loss=0.3226, pruned_loss=0.1458, over 13460.00 frames. ], tot_loss[loss=0.3071, simple_loss=0.3226, pruned_loss=0.1458, over 13460.00 frames. ], batch size: 100, lr: 2.63e-02, grad_scale: 16.0 2023-04-16 14:18:44,291 INFO [train.py:918] (1/4) Computing validation loss 2023-04-16 14:19:06,508 INFO [train.py:927] (1/4) Epoch 5, validation: loss=0.2077, simple_loss=0.2523, pruned_loss=0.08155, over 2446609.00 frames. 2023-04-16 14:19:06,509 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12761MB 2023-04-16 14:19:13,548 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.868e+02 4.381e+02 5.128e+02 6.177e+02 2.040e+03, threshold=1.026e+03, percent-clipped=5.0 2023-04-16 14:19:13,789 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6948, 4.5843, 4.9810, 4.6737, 5.0747, 4.4997, 5.1421, 5.0885], device='cuda:1'), covar=tensor([0.0303, 0.0399, 0.0357, 0.0431, 0.0441, 0.0594, 0.0324, 0.0386], device='cuda:1'), in_proj_covar=tensor([0.0135, 0.0164, 0.0149, 0.0110, 0.0212, 0.0182, 0.0127, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 14:19:14,752 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10805.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:19:19,627 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=10810.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 14:19:43,018 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10835.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:19:53,451 INFO [train.py:893] (1/4) Epoch 5, batch 50, loss[loss=0.2428, simple_loss=0.2721, pruned_loss=0.1068, over 13347.00 frames. ], tot_loss[loss=0.281, simple_loss=0.3073, pruned_loss=0.1274, over 600709.14 frames. ], batch size: 62, lr: 2.63e-02, grad_scale: 16.0 2023-04-16 14:20:18,220 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-16 14:20:18,220 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-16 14:20:18,221 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-16 14:20:18,233 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-16 14:20:18,241 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-16 14:20:19,208 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-16 14:20:19,227 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-16 14:20:28,680 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=10885.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 14:20:40,859 INFO [train.py:893] (1/4) Epoch 5, batch 100, loss[loss=0.2745, simple_loss=0.3101, pruned_loss=0.1194, over 13472.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3085, pruned_loss=0.1272, over 1055478.94 frames. ], batch size: 81, lr: 2.62e-02, grad_scale: 16.0 2023-04-16 14:20:46,854 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.725e+02 3.963e+02 4.715e+02 5.825e+02 1.168e+03, threshold=9.430e+02, percent-clipped=1.0 2023-04-16 14:21:27,210 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=10946.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 14:21:27,774 INFO [train.py:893] (1/4) Epoch 5, batch 150, loss[loss=0.2955, simple_loss=0.3226, pruned_loss=0.1342, over 13543.00 frames. ], tot_loss[loss=0.2867, simple_loss=0.3131, pruned_loss=0.1302, over 1411889.89 frames. ], batch size: 78, lr: 2.62e-02, grad_scale: 16.0 2023-04-16 14:21:54,095 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.83 vs. limit=5.0 2023-04-16 14:21:57,253 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.45 vs. limit=5.0 2023-04-16 14:22:15,272 INFO [train.py:893] (1/4) Epoch 5, batch 200, loss[loss=0.3085, simple_loss=0.3389, pruned_loss=0.139, over 13458.00 frames. ], tot_loss[loss=0.288, simple_loss=0.3144, pruned_loss=0.1309, over 1673324.37 frames. ], batch size: 103, lr: 2.61e-02, grad_scale: 16.0 2023-04-16 14:22:22,138 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 3.004e+02 4.283e+02 5.211e+02 6.279e+02 9.649e+02, threshold=1.042e+03, percent-clipped=2.0 2023-04-16 14:23:02,406 INFO [train.py:893] (1/4) Epoch 5, batch 250, loss[loss=0.2702, simple_loss=0.3099, pruned_loss=0.1153, over 13458.00 frames. ], tot_loss[loss=0.2874, simple_loss=0.3142, pruned_loss=0.1303, over 1895520.92 frames. ], batch size: 79, lr: 2.61e-02, grad_scale: 16.0 2023-04-16 14:23:20,276 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11065.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:23:50,387 INFO [train.py:893] (1/4) Epoch 5, batch 300, loss[loss=0.3005, simple_loss=0.3317, pruned_loss=0.1346, over 13389.00 frames. ], tot_loss[loss=0.2868, simple_loss=0.3142, pruned_loss=0.1297, over 2055968.40 frames. ], batch size: 109, lr: 2.60e-02, grad_scale: 16.0 2023-04-16 14:23:53,141 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11100.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:23:56,164 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.454e+02 3.650e+02 4.425e+02 5.854e+02 1.368e+03, threshold=8.851e+02, percent-clipped=2.0 2023-04-16 14:24:02,136 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11110.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 14:24:04,043 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 2023-04-16 14:24:04,576 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11113.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:24:20,958 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11130.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:24:36,580 INFO [train.py:893] (1/4) Epoch 5, batch 350, loss[loss=0.2836, simple_loss=0.3137, pruned_loss=0.1268, over 13549.00 frames. ], tot_loss[loss=0.2873, simple_loss=0.3144, pruned_loss=0.13, over 2184678.37 frames. ], batch size: 83, lr: 2.60e-02, grad_scale: 16.0 2023-04-16 14:24:36,974 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5204, 4.3982, 4.0220, 3.6276, 3.7398, 2.3786, 4.6147, 2.8887], device='cuda:1'), covar=tensor([0.0966, 0.0194, 0.0325, 0.0634, 0.0374, 0.2259, 0.0106, 0.2308], device='cuda:1'), in_proj_covar=tensor([0.0226, 0.0176, 0.0184, 0.0210, 0.0165, 0.0219, 0.0130, 0.0269], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 14:24:46,751 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11158.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:25:01,543 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11174.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:25:22,607 INFO [train.py:893] (1/4) Epoch 5, batch 400, loss[loss=0.3115, simple_loss=0.3415, pruned_loss=0.1408, over 13427.00 frames. ], tot_loss[loss=0.2854, simple_loss=0.3137, pruned_loss=0.1285, over 2293778.28 frames. ], batch size: 95, lr: 2.59e-02, grad_scale: 16.0 2023-04-16 14:25:29,547 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.497e+02 3.829e+02 4.791e+02 5.945e+02 1.287e+03, threshold=9.581e+02, percent-clipped=2.0 2023-04-16 14:25:42,637 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.00 vs. limit=5.0 2023-04-16 14:25:58,573 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11235.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:26:04,302 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11241.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 14:26:09,284 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6035, 3.6554, 3.5519, 3.4888, 3.7053, 3.4553, 3.6993, 3.6773], device='cuda:1'), covar=tensor([0.0316, 0.0569, 0.0375, 0.0600, 0.0305, 0.0534, 0.0540, 0.0564], device='cuda:1'), in_proj_covar=tensor([0.0090, 0.0126, 0.0089, 0.0119, 0.0084, 0.0118, 0.0089, 0.0097], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 14:26:09,822 INFO [train.py:893] (1/4) Epoch 5, batch 450, loss[loss=0.3793, simple_loss=0.3766, pruned_loss=0.1911, over 11784.00 frames. ], tot_loss[loss=0.2849, simple_loss=0.3138, pruned_loss=0.1281, over 2374841.74 frames. ], batch size: 157, lr: 2.59e-02, grad_scale: 16.0 2023-04-16 14:26:38,562 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-16 14:26:57,073 INFO [train.py:893] (1/4) Epoch 5, batch 500, loss[loss=0.2861, simple_loss=0.3219, pruned_loss=0.1252, over 13584.00 frames. ], tot_loss[loss=0.2836, simple_loss=0.3132, pruned_loss=0.127, over 2440218.95 frames. ], batch size: 89, lr: 2.58e-02, grad_scale: 16.0 2023-04-16 14:27:01,615 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1252, 2.1320, 2.1164, 3.6156, 3.3749, 3.7387, 2.8765, 1.9656], device='cuda:1'), covar=tensor([0.0332, 0.1381, 0.1140, 0.0085, 0.0191, 0.0045, 0.0587, 0.1479], device='cuda:1'), in_proj_covar=tensor([0.0073, 0.0133, 0.0119, 0.0068, 0.0066, 0.0056, 0.0113, 0.0122], device='cuda:1'), out_proj_covar=tensor([1.2569e-04, 1.9644e-04, 1.7799e-04, 1.0477e-04, 1.1393e-04, 8.8203e-05, 1.6999e-04, 1.8258e-04], device='cuda:1') 2023-04-16 14:27:03,860 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.346e+02 3.644e+02 4.498e+02 5.635e+02 7.796e+02, threshold=8.995e+02, percent-clipped=0.0 2023-04-16 14:27:07,510 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4516, 3.8813, 3.5229, 4.2064, 2.0253, 2.8967, 3.7335, 1.8955], device='cuda:1'), covar=tensor([0.0091, 0.0325, 0.0533, 0.0138, 0.1784, 0.0790, 0.0511, 0.2375], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0122, 0.0138, 0.0084, 0.0161, 0.0146, 0.0132, 0.0166], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 14:27:44,998 INFO [train.py:893] (1/4) Epoch 5, batch 550, loss[loss=0.2671, simple_loss=0.2969, pruned_loss=0.1186, over 13541.00 frames. ], tot_loss[loss=0.2838, simple_loss=0.3135, pruned_loss=0.127, over 2490476.93 frames. ], batch size: 72, lr: 2.58e-02, grad_scale: 16.0 2023-04-16 14:28:24,955 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=2.09 vs. limit=2.0 2023-04-16 14:28:31,861 INFO [train.py:893] (1/4) Epoch 5, batch 600, loss[loss=0.291, simple_loss=0.3198, pruned_loss=0.1311, over 13219.00 frames. ], tot_loss[loss=0.2827, simple_loss=0.3123, pruned_loss=0.1266, over 2528546.55 frames. ], batch size: 132, lr: 2.57e-02, grad_scale: 16.0 2023-04-16 14:28:35,643 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11400.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:28:38,565 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.783e+02 4.017e+02 4.794e+02 6.254e+02 1.645e+03, threshold=9.587e+02, percent-clipped=3.0 2023-04-16 14:29:03,580 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11430.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:29:19,323 INFO [train.py:893] (1/4) Epoch 5, batch 650, loss[loss=0.2349, simple_loss=0.2695, pruned_loss=0.1002, over 13350.00 frames. ], tot_loss[loss=0.2821, simple_loss=0.3113, pruned_loss=0.1264, over 2555645.08 frames. ], batch size: 67, lr: 2.57e-02, grad_scale: 16.0 2023-04-16 14:29:20,303 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11448.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:29:48,616 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11478.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:30:05,796 INFO [train.py:893] (1/4) Epoch 5, batch 700, loss[loss=0.2789, simple_loss=0.3016, pruned_loss=0.1281, over 11878.00 frames. ], tot_loss[loss=0.2798, simple_loss=0.3097, pruned_loss=0.1249, over 2581156.41 frames. ], batch size: 157, lr: 2.56e-02, grad_scale: 16.0 2023-04-16 14:30:12,399 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.679e+02 3.843e+02 4.625e+02 6.070e+02 1.081e+03, threshold=9.250e+02, percent-clipped=1.0 2023-04-16 14:30:36,446 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11530.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:30:39,053 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11532.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:30:47,084 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11541.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 14:30:49,199 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.75 vs. limit=2.0 2023-04-16 14:30:51,862 INFO [train.py:893] (1/4) Epoch 5, batch 750, loss[loss=0.2873, simple_loss=0.3181, pruned_loss=0.1282, over 13531.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3101, pruned_loss=0.1264, over 2597631.38 frames. ], batch size: 85, lr: 2.56e-02, grad_scale: 16.0 2023-04-16 14:31:06,408 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4982, 4.1365, 3.5359, 3.7835, 3.9855, 4.3650, 4.1403, 3.9452], device='cuda:1'), covar=tensor([0.0428, 0.0289, 0.0380, 0.1243, 0.0240, 0.0270, 0.0244, 0.0464], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0086, 0.0087, 0.0159, 0.0089, 0.0100, 0.0086, 0.0080], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0004, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 14:31:25,616 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.04 vs. limit=2.0 2023-04-16 14:31:31,759 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11589.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 14:31:35,150 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11593.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:31:39,415 INFO [train.py:893] (1/4) Epoch 5, batch 800, loss[loss=0.2784, simple_loss=0.3058, pruned_loss=0.1255, over 13238.00 frames. ], tot_loss[loss=0.2815, simple_loss=0.3106, pruned_loss=0.1262, over 2613873.64 frames. ], batch size: 124, lr: 2.55e-02, grad_scale: 16.0 2023-04-16 14:31:46,339 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.973e+02 3.812e+02 4.919e+02 6.723e+02 1.204e+03, threshold=9.839e+02, percent-clipped=3.0 2023-04-16 14:32:00,633 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3568, 4.7270, 3.1267, 4.8690, 4.5179, 2.8416, 3.8773, 2.8932], device='cuda:1'), covar=tensor([0.0206, 0.0231, 0.0993, 0.0043, 0.0150, 0.1013, 0.0538, 0.1597], device='cuda:1'), in_proj_covar=tensor([0.0103, 0.0100, 0.0155, 0.0068, 0.0095, 0.0144, 0.0124, 0.0154], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0001, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-16 14:32:15,393 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11635.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:32:26,045 INFO [train.py:893] (1/4) Epoch 5, batch 850, loss[loss=0.272, simple_loss=0.2971, pruned_loss=0.1235, over 13526.00 frames. ], tot_loss[loss=0.2817, simple_loss=0.3108, pruned_loss=0.1263, over 2625628.56 frames. ], batch size: 72, lr: 2.55e-02, grad_scale: 16.0 2023-04-16 14:33:12,646 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11696.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:33:13,141 INFO [train.py:893] (1/4) Epoch 5, batch 900, loss[loss=0.2616, simple_loss=0.2941, pruned_loss=0.1145, over 13507.00 frames. ], tot_loss[loss=0.2814, simple_loss=0.3106, pruned_loss=0.1261, over 2631418.59 frames. ], batch size: 70, lr: 2.54e-02, grad_scale: 16.0 2023-04-16 14:33:15,774 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11699.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:33:19,657 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.389e+02 3.741e+02 4.900e+02 6.173e+02 9.470e+02, threshold=9.800e+02, percent-clipped=0.0 2023-04-16 14:33:45,851 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-16 14:33:59,674 INFO [train.py:893] (1/4) Epoch 5, batch 950, loss[loss=0.2357, simple_loss=0.251, pruned_loss=0.1102, over 6870.00 frames. ], tot_loss[loss=0.2813, simple_loss=0.3098, pruned_loss=0.1264, over 2633317.13 frames. ], batch size: 27, lr: 2.54e-02, grad_scale: 16.0 2023-04-16 14:34:09,316 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7128, 4.2298, 3.8038, 3.8358, 4.0184, 4.4672, 4.2054, 4.0982], device='cuda:1'), covar=tensor([0.0374, 0.0252, 0.0297, 0.1298, 0.0273, 0.0249, 0.0280, 0.0378], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0087, 0.0089, 0.0160, 0.0091, 0.0101, 0.0088, 0.0080], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 14:34:11,862 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11760.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:34:33,190 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.7740, 4.7106, 4.9665, 4.7294, 5.1336, 4.5387, 5.1292, 5.1851], device='cuda:1'), covar=tensor([0.0312, 0.0411, 0.0423, 0.0390, 0.0469, 0.0596, 0.0473, 0.0320], device='cuda:1'), in_proj_covar=tensor([0.0141, 0.0171, 0.0156, 0.0116, 0.0225, 0.0189, 0.0140, 0.0141], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 14:34:46,151 INFO [train.py:893] (1/4) Epoch 5, batch 1000, loss[loss=0.2939, simple_loss=0.3196, pruned_loss=0.1341, over 13320.00 frames. ], tot_loss[loss=0.2778, simple_loss=0.307, pruned_loss=0.1243, over 2639199.54 frames. ], batch size: 118, lr: 2.53e-02, grad_scale: 16.0 2023-04-16 14:34:53,074 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.519e+02 3.813e+02 4.790e+02 5.853e+02 1.056e+03, threshold=9.580e+02, percent-clipped=1.0 2023-04-16 14:34:54,304 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6356, 3.8251, 3.6181, 3.5641, 3.7357, 3.5513, 3.7514, 3.9084], device='cuda:1'), covar=tensor([0.0328, 0.0574, 0.0404, 0.0557, 0.0346, 0.0431, 0.0594, 0.0459], device='cuda:1'), in_proj_covar=tensor([0.0089, 0.0130, 0.0093, 0.0118, 0.0087, 0.0118, 0.0092, 0.0096], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 14:35:05,269 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11816.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:35:17,681 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=11830.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:35:24,564 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0 2023-04-16 14:35:33,471 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11846.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:35:33,988 INFO [train.py:893] (1/4) Epoch 5, batch 1050, loss[loss=0.3011, simple_loss=0.3237, pruned_loss=0.1392, over 11837.00 frames. ], tot_loss[loss=0.2747, simple_loss=0.3049, pruned_loss=0.1222, over 2644639.57 frames. ], batch size: 157, lr: 2.53e-02, grad_scale: 16.0 2023-04-16 14:35:34,179 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6437, 4.2593, 4.2463, 4.0940, 3.8919, 4.0417, 4.5965, 4.1830], device='cuda:1'), covar=tensor([0.0796, 0.0898, 0.1996, 0.2783, 0.1034, 0.1411, 0.0978, 0.0978], device='cuda:1'), in_proj_covar=tensor([0.0201, 0.0224, 0.0293, 0.0298, 0.0165, 0.0235, 0.0276, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 14:36:02,940 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11877.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:36:03,628 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=11878.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:36:05,582 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.29 vs. limit=5.0 2023-04-16 14:36:09,534 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.4632, 1.8546, 3.9512, 3.6295, 3.8478, 3.1811, 3.8100, 2.8503], device='cuda:1'), covar=tensor([0.2360, 0.2117, 0.0052, 0.0237, 0.0127, 0.0535, 0.0134, 0.0798], device='cuda:1'), in_proj_covar=tensor([0.0177, 0.0173, 0.0074, 0.0090, 0.0087, 0.0133, 0.0081, 0.0144], device='cuda:1'), out_proj_covar=tensor([1.9370e-04, 1.9299e-04, 8.7608e-05, 1.1178e-04, 1.0805e-04, 1.5358e-04, 1.0132e-04, 1.6407e-04], device='cuda:1') 2023-04-16 14:36:12,620 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11888.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:36:20,692 INFO [train.py:893] (1/4) Epoch 5, batch 1100, loss[loss=0.2423, simple_loss=0.2823, pruned_loss=0.1011, over 13540.00 frames. ], tot_loss[loss=0.2741, simple_loss=0.3051, pruned_loss=0.1216, over 2647860.77 frames. ], batch size: 72, lr: 2.53e-02, grad_scale: 16.0 2023-04-16 14:36:27,224 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.277e+02 3.881e+02 4.569e+02 5.813e+02 1.527e+03, threshold=9.139e+02, percent-clipped=3.0 2023-04-16 14:36:30,930 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11907.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:36:42,601 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11920.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:37:07,303 INFO [train.py:893] (1/4) Epoch 5, batch 1150, loss[loss=0.2818, simple_loss=0.3147, pruned_loss=0.1244, over 13574.00 frames. ], tot_loss[loss=0.2719, simple_loss=0.3039, pruned_loss=0.1199, over 2653863.11 frames. ], batch size: 89, lr: 2.52e-02, grad_scale: 16.0 2023-04-16 14:37:08,501 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11948.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:37:16,967 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.04 vs. limit=5.0 2023-04-16 14:37:38,815 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=11981.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:37:41,875 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=11984.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:37:48,203 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=11991.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:37:53,631 INFO [train.py:893] (1/4) Epoch 5, batch 1200, loss[loss=0.2273, simple_loss=0.2599, pruned_loss=0.09728, over 13395.00 frames. ], tot_loss[loss=0.2718, simple_loss=0.3045, pruned_loss=0.1196, over 2657473.39 frames. ], batch size: 62, lr: 2.52e-02, grad_scale: 16.0 2023-04-16 14:38:03,290 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.645e+02 3.919e+02 5.054e+02 6.149e+02 1.256e+03, threshold=1.011e+03, percent-clipped=6.0 2023-04-16 14:38:07,824 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12009.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:38:21,018 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9652, 4.0762, 3.5364, 3.1201, 2.9656, 2.5792, 4.2348, 2.8643], device='cuda:1'), covar=tensor([0.1168, 0.0249, 0.0466, 0.0835, 0.0638, 0.2051, 0.0131, 0.2217], device='cuda:1'), in_proj_covar=tensor([0.0233, 0.0189, 0.0191, 0.0216, 0.0170, 0.0220, 0.0135, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 14:38:22,612 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.4511, 1.9194, 1.7444, 2.4790, 1.8267, 2.0530, 2.1611, 2.1438], device='cuda:1'), covar=tensor([0.0148, 0.0271, 0.0175, 0.0129, 0.0153, 0.0091, 0.0255, 0.0218], device='cuda:1'), in_proj_covar=tensor([0.0043, 0.0042, 0.0045, 0.0036, 0.0045, 0.0038, 0.0046, 0.0044], device='cuda:1'), out_proj_covar=tensor([5.1906e-05, 5.1543e-05, 5.8088e-05, 4.4315e-05, 5.3795e-05, 4.5509e-05, 5.8110e-05, 5.5336e-05], device='cuda:1') 2023-04-16 14:38:25,525 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-16 14:38:37,233 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-16 14:38:40,741 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12045.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:38:42,800 INFO [train.py:893] (1/4) Epoch 5, batch 1250, loss[loss=0.3034, simple_loss=0.3301, pruned_loss=0.1383, over 13523.00 frames. ], tot_loss[loss=0.2739, simple_loss=0.3057, pruned_loss=0.121, over 2658983.84 frames. ], batch size: 98, lr: 2.51e-02, grad_scale: 16.0 2023-04-16 14:38:50,522 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12055.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:39:09,578 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12076.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:39:28,281 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.76 vs. limit=5.0 2023-04-16 14:39:29,492 INFO [train.py:893] (1/4) Epoch 5, batch 1300, loss[loss=0.2875, simple_loss=0.3175, pruned_loss=0.1288, over 13388.00 frames. ], tot_loss[loss=0.2762, simple_loss=0.3077, pruned_loss=0.1223, over 2658752.34 frames. ], batch size: 113, lr: 2.51e-02, grad_scale: 16.0 2023-04-16 14:39:35,346 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.667e+02 3.833e+02 4.783e+02 5.972e+02 1.404e+03, threshold=9.567e+02, percent-clipped=5.0 2023-04-16 14:40:00,856 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2023-04-16 14:40:06,436 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12137.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:40:15,948 INFO [train.py:893] (1/4) Epoch 5, batch 1350, loss[loss=0.276, simple_loss=0.3133, pruned_loss=0.1194, over 13468.00 frames. ], tot_loss[loss=0.277, simple_loss=0.3083, pruned_loss=0.1228, over 2655882.56 frames. ], batch size: 103, lr: 2.50e-02, grad_scale: 16.0 2023-04-16 14:40:24,334 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.6566, 1.7242, 4.0062, 3.5396, 3.7442, 3.2488, 3.6587, 2.6592], device='cuda:1'), covar=tensor([0.2796, 0.2378, 0.0087, 0.0346, 0.0247, 0.0483, 0.0173, 0.1046], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0175, 0.0078, 0.0092, 0.0090, 0.0133, 0.0084, 0.0142], device='cuda:1'), out_proj_covar=tensor([2.0091e-04, 1.9603e-04, 9.2408e-05, 1.1411e-04, 1.1269e-04, 1.5431e-04, 1.0437e-04, 1.6284e-04], device='cuda:1') 2023-04-16 14:40:39,048 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12172.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:40:52,296 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9549, 2.5475, 1.8741, 1.4963, 1.1414, 1.8752, 1.8308, 2.5578], device='cuda:1'), covar=tensor([0.0665, 0.0493, 0.1045, 0.1270, 0.0385, 0.0306, 0.0544, 0.0247], device='cuda:1'), in_proj_covar=tensor([0.0081, 0.0070, 0.0063, 0.0103, 0.0059, 0.0069, 0.0079, 0.0064], device='cuda:1'), out_proj_covar=tensor([6.8392e-05, 5.6533e-05, 5.7393e-05, 9.3121e-05, 5.7513e-05, 5.6042e-05, 6.6725e-05, 5.1351e-05], device='cuda:1') 2023-04-16 14:40:54,595 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12188.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:41:01,902 INFO [train.py:893] (1/4) Epoch 5, batch 1400, loss[loss=0.2825, simple_loss=0.3115, pruned_loss=0.1267, over 13304.00 frames. ], tot_loss[loss=0.2771, simple_loss=0.3083, pruned_loss=0.1229, over 2658996.57 frames. ], batch size: 124, lr: 2.50e-02, grad_scale: 16.0 2023-04-16 14:41:06,614 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12202.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:41:08,834 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.680e+02 3.684e+02 4.695e+02 6.104e+02 1.016e+03, threshold=9.390e+02, percent-clipped=1.0 2023-04-16 14:41:34,877 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.1022, 1.8216, 3.9962, 3.6890, 3.8657, 3.1630, 3.7039, 2.6723], device='cuda:1'), covar=tensor([0.2132, 0.2417, 0.0083, 0.0265, 0.0177, 0.0614, 0.0168, 0.0956], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0175, 0.0077, 0.0090, 0.0089, 0.0134, 0.0084, 0.0141], device='cuda:1'), out_proj_covar=tensor([1.9619e-04, 1.9571e-04, 9.1682e-05, 1.1277e-04, 1.1119e-04, 1.5538e-04, 1.0413e-04, 1.6179e-04], device='cuda:1') 2023-04-16 14:41:39,432 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12236.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:41:49,230 INFO [train.py:893] (1/4) Epoch 5, batch 1450, loss[loss=0.2595, simple_loss=0.2952, pruned_loss=0.1118, over 13552.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3083, pruned_loss=0.1228, over 2659613.83 frames. ], batch size: 76, lr: 2.49e-02, grad_scale: 32.0 2023-04-16 14:42:17,029 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12276.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:42:30,884 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12291.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:42:36,230 INFO [train.py:893] (1/4) Epoch 5, batch 1500, loss[loss=0.2922, simple_loss=0.3245, pruned_loss=0.1299, over 13370.00 frames. ], tot_loss[loss=0.2751, simple_loss=0.3067, pruned_loss=0.1217, over 2660857.17 frames. ], batch size: 109, lr: 2.49e-02, grad_scale: 32.0 2023-04-16 14:42:42,849 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12304.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:42:43,471 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.442e+02 3.651e+02 4.300e+02 5.550e+02 1.141e+03, threshold=8.601e+02, percent-clipped=1.0 2023-04-16 14:43:16,140 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12339.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:43:16,984 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12340.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:43:23,417 INFO [train.py:893] (1/4) Epoch 5, batch 1550, loss[loss=0.2712, simple_loss=0.305, pruned_loss=0.1187, over 13471.00 frames. ], tot_loss[loss=0.2748, simple_loss=0.3066, pruned_loss=0.1215, over 2664132.44 frames. ], batch size: 100, lr: 2.49e-02, grad_scale: 16.0 2023-04-16 14:43:30,223 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12355.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:44:03,441 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6568, 5.0072, 3.6678, 4.9943, 4.7260, 3.1544, 4.0618, 3.4355], device='cuda:1'), covar=tensor([0.0168, 0.0231, 0.0910, 0.0059, 0.0163, 0.1075, 0.0462, 0.1479], device='cuda:1'), in_proj_covar=tensor([0.0109, 0.0108, 0.0164, 0.0074, 0.0100, 0.0149, 0.0129, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 14:44:09,637 INFO [train.py:893] (1/4) Epoch 5, batch 1600, loss[loss=0.2934, simple_loss=0.3183, pruned_loss=0.1342, over 13527.00 frames. ], tot_loss[loss=0.2755, simple_loss=0.3072, pruned_loss=0.1219, over 2660296.17 frames. ], batch size: 85, lr: 2.48e-02, grad_scale: 16.0 2023-04-16 14:44:16,082 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12403.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:44:17,654 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.771e+02 3.931e+02 5.417e+02 6.159e+02 1.328e+03, threshold=1.083e+03, percent-clipped=6.0 2023-04-16 14:44:25,198 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-04-16 14:44:43,295 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12432.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:44:52,925 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.59 vs. limit=5.0 2023-04-16 14:44:57,076 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12446.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:44:57,539 INFO [train.py:893] (1/4) Epoch 5, batch 1650, loss[loss=0.296, simple_loss=0.3253, pruned_loss=0.1333, over 13531.00 frames. ], tot_loss[loss=0.2759, simple_loss=0.3087, pruned_loss=0.1215, over 2665558.61 frames. ], batch size: 91, lr: 2.48e-02, grad_scale: 16.0 2023-04-16 14:44:58,483 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.0357, 4.5537, 4.4546, 4.4657, 4.2523, 4.4509, 4.9410, 4.4774], device='cuda:1'), covar=tensor([0.0797, 0.0934, 0.2284, 0.3000, 0.0814, 0.1196, 0.0967, 0.0856], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0237, 0.0318, 0.0318, 0.0167, 0.0247, 0.0288, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 14:45:21,480 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12472.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:45:44,635 INFO [train.py:893] (1/4) Epoch 5, batch 1700, loss[loss=0.2471, simple_loss=0.2922, pruned_loss=0.101, over 13567.00 frames. ], tot_loss[loss=0.2769, simple_loss=0.3097, pruned_loss=0.1221, over 2663784.01 frames. ], batch size: 89, lr: 2.47e-02, grad_scale: 16.0 2023-04-16 14:45:49,669 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12502.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:45:51,910 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.574e+02 4.089e+02 5.133e+02 6.937e+02 1.880e+03, threshold=1.027e+03, percent-clipped=5.0 2023-04-16 14:45:53,865 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12507.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:46:06,221 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12520.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:46:12,057 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6095, 4.5655, 4.0363, 3.7303, 3.9836, 2.5746, 4.8101, 2.9922], device='cuda:1'), covar=tensor([0.0882, 0.0192, 0.0377, 0.0681, 0.0291, 0.1971, 0.0128, 0.2264], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0195, 0.0201, 0.0223, 0.0176, 0.0229, 0.0143, 0.0278], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 14:46:31,418 INFO [train.py:893] (1/4) Epoch 5, batch 1750, loss[loss=0.2883, simple_loss=0.3245, pruned_loss=0.1261, over 13439.00 frames. ], tot_loss[loss=0.2743, simple_loss=0.3079, pruned_loss=0.1204, over 2667203.62 frames. ], batch size: 103, lr: 2.47e-02, grad_scale: 16.0 2023-04-16 14:46:34,155 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12550.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:46:58,505 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12576.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:47:13,738 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-16 14:47:18,892 INFO [train.py:893] (1/4) Epoch 5, batch 1800, loss[loss=0.268, simple_loss=0.3093, pruned_loss=0.1134, over 13519.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.306, pruned_loss=0.1185, over 2666989.93 frames. ], batch size: 85, lr: 2.46e-02, grad_scale: 16.0 2023-04-16 14:47:25,400 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12604.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:47:25,957 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.081e+02 3.697e+02 4.449e+02 5.454e+02 1.218e+03, threshold=8.898e+02, percent-clipped=2.0 2023-04-16 14:47:44,590 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12624.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:47:58,719 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12640.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:48:05,087 INFO [train.py:893] (1/4) Epoch 5, batch 1850, loss[loss=0.2598, simple_loss=0.2985, pruned_loss=0.1105, over 13470.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3055, pruned_loss=0.1183, over 2664444.91 frames. ], batch size: 103, lr: 2.46e-02, grad_scale: 16.0 2023-04-16 14:48:05,593 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2023-04-16 14:48:09,213 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-16 14:48:11,155 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12652.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:48:44,545 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12688.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:48:45,546 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.6693, 2.2643, 1.9162, 1.2994, 0.9352, 1.7761, 1.5117, 2.3393], device='cuda:1'), covar=tensor([0.0607, 0.0375, 0.0744, 0.1187, 0.0276, 0.0257, 0.0565, 0.0254], device='cuda:1'), in_proj_covar=tensor([0.0087, 0.0075, 0.0064, 0.0107, 0.0061, 0.0073, 0.0086, 0.0068], device='cuda:1'), out_proj_covar=tensor([7.2474e-05, 6.0052e-05, 5.7798e-05, 9.5969e-05, 5.9438e-05, 5.7967e-05, 7.1951e-05, 5.4095e-05], device='cuda:1') 2023-04-16 14:48:52,003 INFO [train.py:893] (1/4) Epoch 5, batch 1900, loss[loss=0.2912, simple_loss=0.3195, pruned_loss=0.1314, over 13370.00 frames. ], tot_loss[loss=0.2706, simple_loss=0.3046, pruned_loss=0.1183, over 2661740.65 frames. ], batch size: 109, lr: 2.46e-02, grad_scale: 16.0 2023-04-16 14:48:59,086 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4274, 4.3658, 4.3615, 3.4954, 4.9323, 4.6244, 4.5475, 4.9360], device='cuda:1'), covar=tensor([0.0179, 0.0081, 0.0126, 0.0667, 0.0090, 0.0107, 0.0095, 0.0068], device='cuda:1'), in_proj_covar=tensor([0.0053, 0.0028, 0.0045, 0.0068, 0.0056, 0.0047, 0.0046, 0.0035], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0001], device='cuda:1') 2023-04-16 14:48:59,623 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.759e+02 3.839e+02 4.372e+02 5.320e+02 1.307e+03, threshold=8.743e+02, percent-clipped=2.0 2023-04-16 14:49:24,689 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=12732.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:49:40,115 INFO [train.py:893] (1/4) Epoch 5, batch 1950, loss[loss=0.2547, simple_loss=0.294, pruned_loss=0.1077, over 13347.00 frames. ], tot_loss[loss=0.2698, simple_loss=0.3041, pruned_loss=0.1178, over 2660050.26 frames. ], batch size: 118, lr: 2.45e-02, grad_scale: 16.0 2023-04-16 14:49:53,815 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=12763.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:50:10,899 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=12780.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:50:25,727 INFO [train.py:893] (1/4) Epoch 5, batch 2000, loss[loss=0.2496, simple_loss=0.2854, pruned_loss=0.1069, over 13372.00 frames. ], tot_loss[loss=0.2731, simple_loss=0.3065, pruned_loss=0.1198, over 2660049.30 frames. ], batch size: 67, lr: 2.45e-02, grad_scale: 16.0 2023-04-16 14:50:30,276 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=12802.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:50:32,522 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-16 14:50:33,266 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.676e+02 3.883e+02 4.614e+02 5.841e+02 1.269e+03, threshold=9.228e+02, percent-clipped=6.0 2023-04-16 14:50:51,115 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=12824.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 14:50:54,395 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.23 vs. limit=5.0 2023-04-16 14:51:02,100 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.8204, 5.3413, 5.2637, 5.1328, 4.7377, 5.1190, 5.6905, 5.1360], device='cuda:1'), covar=tensor([0.0538, 0.0652, 0.1743, 0.2451, 0.0716, 0.1383, 0.0812, 0.0856], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0226, 0.0307, 0.0309, 0.0162, 0.0235, 0.0277, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 14:51:12,429 INFO [train.py:893] (1/4) Epoch 5, batch 2050, loss[loss=0.2667, simple_loss=0.3066, pruned_loss=0.1134, over 13532.00 frames. ], tot_loss[loss=0.2746, simple_loss=0.3075, pruned_loss=0.1209, over 2658435.58 frames. ], batch size: 91, lr: 2.44e-02, grad_scale: 16.0 2023-04-16 14:51:16,023 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4983, 2.0968, 2.4700, 3.9114, 3.5960, 3.8127, 3.0545, 2.2976], device='cuda:1'), covar=tensor([0.0375, 0.1401, 0.1103, 0.0063, 0.0222, 0.0061, 0.0799, 0.1308], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0135, 0.0127, 0.0071, 0.0071, 0.0062, 0.0123, 0.0123], device='cuda:1'), out_proj_covar=tensor([1.3755e-04, 2.0703e-04, 1.9513e-04, 1.1048e-04, 1.2508e-04, 9.8970e-05, 1.8930e-04, 1.8872e-04], device='cuda:1') 2023-04-16 14:51:41,524 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0713, 3.0799, 2.3252, 4.0179, 4.5509, 3.3040, 4.3020, 4.1459], device='cuda:1'), covar=tensor([0.0117, 0.0429, 0.0865, 0.0096, 0.0104, 0.0338, 0.0111, 0.0065], device='cuda:1'), in_proj_covar=tensor([0.0043, 0.0055, 0.0068, 0.0048, 0.0033, 0.0052, 0.0030, 0.0039], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-16 14:51:57,936 INFO [train.py:893] (1/4) Epoch 5, batch 2100, loss[loss=0.2728, simple_loss=0.3145, pruned_loss=0.1155, over 13557.00 frames. ], tot_loss[loss=0.2723, simple_loss=0.3057, pruned_loss=0.1194, over 2657802.39 frames. ], batch size: 89, lr: 2.44e-02, grad_scale: 16.0 2023-04-16 14:52:05,977 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.944e+02 3.926e+02 4.592e+02 5.677e+02 1.574e+03, threshold=9.184e+02, percent-clipped=7.0 2023-04-16 14:52:07,426 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=2.06 vs. limit=2.0 2023-04-16 14:52:42,727 INFO [train.py:893] (1/4) Epoch 5, batch 2150, loss[loss=0.2498, simple_loss=0.2963, pruned_loss=0.1017, over 13452.00 frames. ], tot_loss[loss=0.2715, simple_loss=0.3054, pruned_loss=0.1188, over 2657790.90 frames. ], batch size: 95, lr: 2.44e-02, grad_scale: 16.0 2023-04-16 14:53:19,195 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0883, 1.9257, 1.2623, 2.1905, 1.3981, 1.5905, 1.9199, 1.7376], device='cuda:1'), covar=tensor([0.0101, 0.0135, 0.0157, 0.0107, 0.0137, 0.0142, 0.0202, 0.0191], device='cuda:1'), in_proj_covar=tensor([0.0044, 0.0043, 0.0047, 0.0037, 0.0049, 0.0040, 0.0048, 0.0045], device='cuda:1'), out_proj_covar=tensor([5.2129e-05, 5.3421e-05, 6.0178e-05, 4.6049e-05, 6.0309e-05, 4.8325e-05, 6.0651e-05, 5.5807e-05], device='cuda:1') 2023-04-16 14:53:29,480 INFO [train.py:893] (1/4) Epoch 5, batch 2200, loss[loss=0.2635, simple_loss=0.3037, pruned_loss=0.1117, over 13388.00 frames. ], tot_loss[loss=0.2683, simple_loss=0.3031, pruned_loss=0.1168, over 2657083.66 frames. ], batch size: 113, lr: 2.43e-02, grad_scale: 16.0 2023-04-16 14:53:36,950 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.620e+02 3.386e+02 4.268e+02 5.257e+02 8.997e+02, threshold=8.536e+02, percent-clipped=0.0 2023-04-16 14:53:47,302 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0 2023-04-16 14:53:58,833 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-16 14:54:08,515 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.8910, 4.7824, 4.6766, 4.5865, 5.1601, 4.5027, 5.0216, 5.1494], device='cuda:1'), covar=tensor([0.0472, 0.0823, 0.1123, 0.0674, 0.1166, 0.1452, 0.0948, 0.0715], device='cuda:1'), in_proj_covar=tensor([0.0144, 0.0177, 0.0163, 0.0119, 0.0235, 0.0197, 0.0145, 0.0151], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0004, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 14:54:15,906 INFO [train.py:893] (1/4) Epoch 5, batch 2250, loss[loss=0.268, simple_loss=0.3022, pruned_loss=0.1169, over 12050.00 frames. ], tot_loss[loss=0.2689, simple_loss=0.3031, pruned_loss=0.1173, over 2653695.83 frames. ], batch size: 157, lr: 2.43e-02, grad_scale: 16.0 2023-04-16 14:54:17,927 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7725, 1.8599, 3.5742, 3.5431, 3.3653, 2.8122, 3.4643, 2.4375], device='cuda:1'), covar=tensor([0.2147, 0.1939, 0.0070, 0.0151, 0.0210, 0.0601, 0.0118, 0.1040], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0181, 0.0080, 0.0093, 0.0093, 0.0139, 0.0090, 0.0149], device='cuda:1'), out_proj_covar=tensor([2.0323e-04, 2.0556e-04, 9.6334e-05, 1.1539e-04, 1.1758e-04, 1.6347e-04, 1.1321e-04, 1.7298e-04], device='cuda:1') 2023-04-16 14:54:26,476 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-04-16 14:54:32,483 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.80 vs. limit=2.0 2023-04-16 14:54:45,396 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5494, 3.2049, 3.4997, 2.7904, 3.6521, 3.5112, 3.4568, 3.7532], device='cuda:1'), covar=tensor([0.0165, 0.0124, 0.0143, 0.0642, 0.0117, 0.0124, 0.0134, 0.0078], device='cuda:1'), in_proj_covar=tensor([0.0052, 0.0029, 0.0045, 0.0070, 0.0056, 0.0047, 0.0046, 0.0036], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0001], device='cuda:1') 2023-04-16 14:55:03,665 INFO [train.py:893] (1/4) Epoch 5, batch 2300, loss[loss=0.2807, simple_loss=0.3154, pruned_loss=0.123, over 13449.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.3022, pruned_loss=0.1164, over 2657145.21 frames. ], batch size: 100, lr: 2.42e-02, grad_scale: 16.0 2023-04-16 14:55:08,193 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=13102.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:55:11,127 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.210e+02 3.671e+02 4.489e+02 5.410e+02 8.963e+02, threshold=8.978e+02, percent-clipped=2.0 2023-04-16 14:55:24,571 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=13119.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 14:55:49,228 INFO [train.py:893] (1/4) Epoch 5, batch 2350, loss[loss=0.2739, simple_loss=0.3026, pruned_loss=0.1226, over 13516.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3006, pruned_loss=0.1155, over 2658411.78 frames. ], batch size: 85, lr: 2.42e-02, grad_scale: 16.0 2023-04-16 14:55:52,646 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=13150.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 14:56:04,413 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2483, 2.8752, 2.3232, 4.1237, 4.6445, 3.3310, 4.5043, 4.1708], device='cuda:1'), covar=tensor([0.0085, 0.0467, 0.0799, 0.0082, 0.0080, 0.0339, 0.0097, 0.0069], device='cuda:1'), in_proj_covar=tensor([0.0041, 0.0055, 0.0068, 0.0048, 0.0034, 0.0052, 0.0030, 0.0040], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-16 14:56:05,223 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4398, 1.9644, 2.4090, 3.7910, 3.5732, 3.6629, 2.8810, 2.0045], device='cuda:1'), covar=tensor([0.0234, 0.1385, 0.1003, 0.0045, 0.0174, 0.0059, 0.0684, 0.1341], device='cuda:1'), in_proj_covar=tensor([0.0081, 0.0138, 0.0131, 0.0071, 0.0073, 0.0061, 0.0123, 0.0126], device='cuda:1'), out_proj_covar=tensor([1.4092e-04, 2.1210e-04, 2.0048e-04, 1.1138e-04, 1.2754e-04, 9.7977e-05, 1.8967e-04, 1.9342e-04], device='cuda:1') 2023-04-16 14:56:13,235 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-16 14:56:18,559 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.44 vs. limit=5.0 2023-04-16 14:56:22,687 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.09 vs. limit=2.0 2023-04-16 14:56:36,309 INFO [train.py:893] (1/4) Epoch 5, batch 2400, loss[loss=0.31, simple_loss=0.3262, pruned_loss=0.1469, over 11673.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3001, pruned_loss=0.1158, over 2659370.30 frames. ], batch size: 157, lr: 2.42e-02, grad_scale: 16.0 2023-04-16 14:56:43,918 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.422e+02 3.773e+02 4.446e+02 5.529e+02 1.176e+03, threshold=8.892e+02, percent-clipped=3.0 2023-04-16 14:56:45,986 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6339, 3.2013, 2.8582, 3.2199, 2.8193, 1.6255, 3.1134, 1.7180], device='cuda:1'), covar=tensor([0.0649, 0.0809, 0.0441, 0.0335, 0.0916, 0.1986, 0.0857, 0.1623], device='cuda:1'), in_proj_covar=tensor([0.0091, 0.0076, 0.0085, 0.0075, 0.0106, 0.0128, 0.0078, 0.0104], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-16 14:57:22,837 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-04-16 14:57:23,065 INFO [train.py:893] (1/4) Epoch 5, batch 2450, loss[loss=0.2294, simple_loss=0.2732, pruned_loss=0.09282, over 13414.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3, pruned_loss=0.1158, over 2657256.09 frames. ], batch size: 65, lr: 2.41e-02, grad_scale: 16.0 2023-04-16 14:57:27,562 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1887, 1.9162, 2.3209, 3.4409, 3.1786, 3.4774, 2.9006, 2.0044], device='cuda:1'), covar=tensor([0.0241, 0.1300, 0.0993, 0.0069, 0.0263, 0.0051, 0.0531, 0.1234], device='cuda:1'), in_proj_covar=tensor([0.0081, 0.0138, 0.0132, 0.0070, 0.0074, 0.0064, 0.0123, 0.0127], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-16 14:57:29,445 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.99 vs. limit=2.0 2023-04-16 14:58:02,938 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.73 vs. limit=2.0 2023-04-16 14:58:09,761 INFO [train.py:893] (1/4) Epoch 5, batch 2500, loss[loss=0.2654, simple_loss=0.2998, pruned_loss=0.1155, over 13208.00 frames. ], tot_loss[loss=0.2661, simple_loss=0.3003, pruned_loss=0.1159, over 2654492.24 frames. ], batch size: 132, lr: 2.41e-02, grad_scale: 16.0 2023-04-16 14:58:16,735 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.4623, 1.8494, 3.6840, 3.7516, 3.4594, 2.9393, 3.4448, 2.7575], device='cuda:1'), covar=tensor([0.2485, 0.1800, 0.0084, 0.0121, 0.0230, 0.0491, 0.0164, 0.0890], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0178, 0.0078, 0.0089, 0.0091, 0.0134, 0.0090, 0.0146], device='cuda:1'), out_proj_covar=tensor([2.0186e-04, 2.0249e-04, 9.3967e-05, 1.1019e-04, 1.1422e-04, 1.5788e-04, 1.1225e-04, 1.7107e-04], device='cuda:1') 2023-04-16 14:58:17,221 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.475e+02 3.839e+02 4.684e+02 5.527e+02 8.923e+02, threshold=9.368e+02, percent-clipped=1.0 2023-04-16 14:58:57,627 INFO [train.py:893] (1/4) Epoch 5, batch 2550, loss[loss=0.2904, simple_loss=0.3268, pruned_loss=0.127, over 13461.00 frames. ], tot_loss[loss=0.2658, simple_loss=0.3001, pruned_loss=0.1158, over 2649869.19 frames. ], batch size: 103, lr: 2.40e-02, grad_scale: 16.0 2023-04-16 14:58:58,866 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0414, 2.5662, 2.2698, 1.6124, 1.3370, 2.3292, 1.7317, 2.6659], device='cuda:1'), covar=tensor([0.0669, 0.0304, 0.0748, 0.1305, 0.0520, 0.0319, 0.0753, 0.0404], device='cuda:1'), in_proj_covar=tensor([0.0090, 0.0080, 0.0071, 0.0115, 0.0065, 0.0078, 0.0090, 0.0071], device='cuda:1'), out_proj_covar=tensor([7.5183e-05, 6.3653e-05, 6.3407e-05, 1.0248e-04, 6.1965e-05, 6.1410e-05, 7.4690e-05, 5.5517e-05], device='cuda:1') 2023-04-16 14:59:13,800 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7156, 4.0895, 4.3741, 3.3470, 2.7531, 3.2317, 4.5132, 4.5783], device='cuda:1'), covar=tensor([0.0593, 0.0596, 0.0192, 0.0950, 0.1241, 0.0748, 0.0121, 0.0097], device='cuda:1'), in_proj_covar=tensor([0.0179, 0.0132, 0.0133, 0.0181, 0.0188, 0.0152, 0.0107, 0.0090], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0001], device='cuda:1') 2023-04-16 14:59:22,415 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-16 14:59:43,136 INFO [train.py:893] (1/4) Epoch 5, batch 2600, loss[loss=0.2462, simple_loss=0.2836, pruned_loss=0.1044, over 13521.00 frames. ], tot_loss[loss=0.2674, simple_loss=0.3009, pruned_loss=0.117, over 2644264.70 frames. ], batch size: 83, lr: 2.40e-02, grad_scale: 16.0 2023-04-16 14:59:50,063 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.342e+02 4.063e+02 4.946e+02 6.470e+02 1.543e+03, threshold=9.893e+02, percent-clipped=8.0 2023-04-16 15:00:00,966 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4159, 4.2483, 3.8315, 3.3391, 3.5092, 2.5524, 4.4692, 2.7421], device='cuda:1'), covar=tensor([0.0855, 0.0197, 0.0294, 0.0657, 0.0338, 0.1726, 0.0126, 0.2037], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0202, 0.0206, 0.0228, 0.0180, 0.0234, 0.0142, 0.0283], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 15:00:04,160 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=13419.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:00:25,853 INFO [train.py:893] (1/4) Epoch 5, batch 2650, loss[loss=0.2683, simple_loss=0.2979, pruned_loss=0.1194, over 13344.00 frames. ], tot_loss[loss=0.2693, simple_loss=0.3017, pruned_loss=0.1184, over 2635913.74 frames. ], batch size: 67, lr: 2.40e-02, grad_scale: 16.0 2023-04-16 15:00:26,804 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0633, 4.5168, 4.2891, 4.1646, 4.0692, 4.0578, 4.5062, 4.5622], device='cuda:1'), covar=tensor([0.0216, 0.0197, 0.0168, 0.0265, 0.0400, 0.0232, 0.0241, 0.0187], device='cuda:1'), in_proj_covar=tensor([0.0093, 0.0137, 0.0100, 0.0129, 0.0095, 0.0124, 0.0094, 0.0104], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:00:39,055 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-04-16 15:00:40,995 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=13467.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:01:22,498 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 15:01:32,174 INFO [train.py:893] (1/4) Epoch 6, batch 0, loss[loss=0.2907, simple_loss=0.3097, pruned_loss=0.1359, over 13449.00 frames. ], tot_loss[loss=0.2907, simple_loss=0.3097, pruned_loss=0.1359, over 13449.00 frames. ], batch size: 103, lr: 2.23e-02, grad_scale: 16.0 2023-04-16 15:01:32,174 INFO [train.py:918] (1/4) Computing validation loss 2023-04-16 15:01:54,542 INFO [train.py:927] (1/4) Epoch 6, validation: loss=0.195, simple_loss=0.2424, pruned_loss=0.07384, over 2446609.00 frames. 2023-04-16 15:01:54,543 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12761MB 2023-04-16 15:02:02,878 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.472e+02 3.959e+02 4.690e+02 5.704e+02 9.254e+02, threshold=9.380e+02, percent-clipped=0.0 2023-04-16 15:02:34,281 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9951, 1.8889, 1.2183, 2.0635, 1.3523, 1.8024, 1.7645, 1.9610], device='cuda:1'), covar=tensor([0.0077, 0.0173, 0.0170, 0.0100, 0.0145, 0.0130, 0.0239, 0.0137], device='cuda:1'), in_proj_covar=tensor([0.0045, 0.0045, 0.0050, 0.0038, 0.0051, 0.0040, 0.0049, 0.0045], device='cuda:1'), out_proj_covar=tensor([5.3830e-05, 5.6514e-05, 6.4026e-05, 4.7954e-05, 6.1618e-05, 4.7756e-05, 6.2917e-05, 5.6301e-05], device='cuda:1') 2023-04-16 15:02:41,152 INFO [train.py:893] (1/4) Epoch 6, batch 50, loss[loss=0.2397, simple_loss=0.2789, pruned_loss=0.1003, over 13547.00 frames. ], tot_loss[loss=0.2595, simple_loss=0.2929, pruned_loss=0.1131, over 604136.50 frames. ], batch size: 76, lr: 2.23e-02, grad_scale: 16.0 2023-04-16 15:03:05,486 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-16 15:03:05,486 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-16 15:03:05,486 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-16 15:03:05,501 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-16 15:03:05,508 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-16 15:03:05,529 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-16 15:03:05,538 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-16 15:03:26,048 INFO [train.py:893] (1/4) Epoch 6, batch 100, loss[loss=0.2864, simple_loss=0.3138, pruned_loss=0.1295, over 13519.00 frames. ], tot_loss[loss=0.2636, simple_loss=0.2963, pruned_loss=0.1154, over 1060198.82 frames. ], batch size: 91, lr: 2.22e-02, grad_scale: 16.0 2023-04-16 15:03:35,310 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.691e+02 3.972e+02 4.871e+02 6.380e+02 1.409e+03, threshold=9.742e+02, percent-clipped=4.0 2023-04-16 15:03:54,670 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.72 vs. limit=2.0 2023-04-16 15:04:12,844 INFO [train.py:893] (1/4) Epoch 6, batch 150, loss[loss=0.2865, simple_loss=0.3228, pruned_loss=0.1251, over 13281.00 frames. ], tot_loss[loss=0.2675, simple_loss=0.2996, pruned_loss=0.1177, over 1415992.27 frames. ], batch size: 124, lr: 2.22e-02, grad_scale: 16.0 2023-04-16 15:04:29,411 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6221, 3.8305, 3.6463, 3.5765, 3.4620, 3.5543, 3.8730, 3.8957], device='cuda:1'), covar=tensor([0.0269, 0.0269, 0.0302, 0.0340, 0.0420, 0.0319, 0.0295, 0.0251], device='cuda:1'), in_proj_covar=tensor([0.0094, 0.0134, 0.0100, 0.0129, 0.0094, 0.0127, 0.0094, 0.0104], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:04:58,315 INFO [train.py:893] (1/4) Epoch 6, batch 200, loss[loss=0.2986, simple_loss=0.3256, pruned_loss=0.1358, over 13503.00 frames. ], tot_loss[loss=0.2696, simple_loss=0.3018, pruned_loss=0.1187, over 1687355.82 frames. ], batch size: 93, lr: 2.22e-02, grad_scale: 16.0 2023-04-16 15:05:06,624 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.332e+02 3.880e+02 4.687e+02 5.534e+02 1.302e+03, threshold=9.374e+02, percent-clipped=1.0 2023-04-16 15:05:15,294 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.9191, 4.2627, 3.7058, 4.5861, 2.5398, 3.2451, 4.0819, 2.1043], device='cuda:1'), covar=tensor([0.0062, 0.0430, 0.0542, 0.0265, 0.1790, 0.0884, 0.0603, 0.2807], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0140, 0.0152, 0.0110, 0.0169, 0.0161, 0.0148, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-16 15:05:36,672 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=13737.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:05:44,470 INFO [train.py:893] (1/4) Epoch 6, batch 250, loss[loss=0.2385, simple_loss=0.263, pruned_loss=0.107, over 12922.00 frames. ], tot_loss[loss=0.2685, simple_loss=0.3009, pruned_loss=0.1181, over 1891146.71 frames. ], batch size: 52, lr: 2.21e-02, grad_scale: 16.0 2023-04-16 15:06:04,927 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0308, 4.4758, 4.1953, 4.0506, 3.9170, 3.9391, 4.5034, 4.4790], device='cuda:1'), covar=tensor([0.0239, 0.0227, 0.0226, 0.0433, 0.0395, 0.0322, 0.0241, 0.0232], device='cuda:1'), in_proj_covar=tensor([0.0097, 0.0140, 0.0105, 0.0135, 0.0098, 0.0134, 0.0098, 0.0110], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:06:22,127 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.4603, 2.6931, 2.4577, 1.9802, 1.5510, 2.1014, 2.3043, 2.7557], device='cuda:1'), covar=tensor([0.0486, 0.0318, 0.0691, 0.0953, 0.0578, 0.0233, 0.0402, 0.0270], device='cuda:1'), in_proj_covar=tensor([0.0089, 0.0081, 0.0071, 0.0116, 0.0064, 0.0076, 0.0089, 0.0070], device='cuda:1'), out_proj_covar=tensor([7.4258e-05, 6.4071e-05, 6.3054e-05, 1.0245e-04, 6.2003e-05, 6.1658e-05, 7.3939e-05, 5.4379e-05], device='cuda:1') 2023-04-16 15:06:30,650 INFO [train.py:893] (1/4) Epoch 6, batch 300, loss[loss=0.2669, simple_loss=0.2989, pruned_loss=0.1174, over 13476.00 frames. ], tot_loss[loss=0.2697, simple_loss=0.3022, pruned_loss=0.1186, over 2059347.99 frames. ], batch size: 79, lr: 2.21e-02, grad_scale: 16.0 2023-04-16 15:06:32,534 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=13798.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 15:06:38,900 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.402e+02 3.920e+02 4.915e+02 6.095e+02 1.429e+03, threshold=9.831e+02, percent-clipped=4.0 2023-04-16 15:06:49,786 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.5964, 5.1631, 5.0868, 4.9748, 4.8224, 4.9661, 5.5584, 5.2251], device='cuda:1'), covar=tensor([0.0746, 0.0903, 0.2124, 0.2369, 0.0633, 0.1443, 0.0907, 0.0733], device='cuda:1'), in_proj_covar=tensor([0.0220, 0.0239, 0.0313, 0.0315, 0.0168, 0.0245, 0.0284, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 15:06:57,262 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 2023-04-16 15:07:16,567 INFO [train.py:893] (1/4) Epoch 6, batch 350, loss[loss=0.2463, simple_loss=0.2892, pruned_loss=0.1017, over 13240.00 frames. ], tot_loss[loss=0.271, simple_loss=0.3038, pruned_loss=0.1192, over 2192830.89 frames. ], batch size: 132, lr: 2.21e-02, grad_scale: 16.0 2023-04-16 15:08:02,430 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2023-04-16 15:08:02,689 INFO [train.py:893] (1/4) Epoch 6, batch 400, loss[loss=0.2662, simple_loss=0.3061, pruned_loss=0.1132, over 13518.00 frames. ], tot_loss[loss=0.2725, simple_loss=0.3053, pruned_loss=0.1199, over 2299603.89 frames. ], batch size: 98, lr: 2.20e-02, grad_scale: 16.0 2023-04-16 15:08:10,846 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.241e+02 3.793e+02 4.578e+02 5.807e+02 9.391e+02, threshold=9.156e+02, percent-clipped=0.0 2023-04-16 15:08:49,261 INFO [train.py:893] (1/4) Epoch 6, batch 450, loss[loss=0.2775, simple_loss=0.3129, pruned_loss=0.1211, over 13460.00 frames. ], tot_loss[loss=0.2728, simple_loss=0.3057, pruned_loss=0.1199, over 2376432.81 frames. ], batch size: 79, lr: 2.20e-02, grad_scale: 16.0 2023-04-16 15:09:12,308 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-16 15:09:20,483 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3001, 4.4732, 4.5150, 4.2538, 4.3384, 4.1181, 4.5805, 4.5833], device='cuda:1'), covar=tensor([0.0268, 0.0497, 0.0268, 0.0467, 0.0350, 0.0448, 0.0469, 0.0359], device='cuda:1'), in_proj_covar=tensor([0.0095, 0.0139, 0.0100, 0.0129, 0.0095, 0.0130, 0.0096, 0.0107], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:09:34,373 INFO [train.py:893] (1/4) Epoch 6, batch 500, loss[loss=0.2485, simple_loss=0.2967, pruned_loss=0.1002, over 13438.00 frames. ], tot_loss[loss=0.2713, simple_loss=0.305, pruned_loss=0.1187, over 2441822.07 frames. ], batch size: 103, lr: 2.20e-02, grad_scale: 16.0 2023-04-16 15:09:46,560 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.786e+02 3.676e+02 4.489e+02 5.724e+02 1.174e+03, threshold=8.978e+02, percent-clipped=2.0 2023-04-16 15:10:02,568 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.5019, 2.0000, 1.7840, 2.3943, 1.6132, 2.1117, 2.1513, 2.3980], device='cuda:1'), covar=tensor([0.0093, 0.0191, 0.0165, 0.0142, 0.0149, 0.0143, 0.0229, 0.0126], device='cuda:1'), in_proj_covar=tensor([0.0045, 0.0046, 0.0052, 0.0039, 0.0052, 0.0041, 0.0049, 0.0049], device='cuda:1'), out_proj_covar=tensor([5.4639e-05, 5.6882e-05, 6.6220e-05, 4.8653e-05, 6.3638e-05, 4.9349e-05, 6.3206e-05, 6.0859e-05], device='cuda:1') 2023-04-16 15:10:21,063 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6307, 4.0852, 4.2510, 3.0855, 2.5776, 3.0515, 4.4216, 4.5477], device='cuda:1'), covar=tensor([0.0801, 0.0509, 0.0282, 0.1219, 0.1540, 0.0994, 0.0140, 0.0076], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0136, 0.0136, 0.0182, 0.0185, 0.0149, 0.0110, 0.0091], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0001], device='cuda:1') 2023-04-16 15:10:25,590 INFO [train.py:893] (1/4) Epoch 6, batch 550, loss[loss=0.2526, simple_loss=0.2816, pruned_loss=0.1118, over 13214.00 frames. ], tot_loss[loss=0.2691, simple_loss=0.3033, pruned_loss=0.1174, over 2490709.24 frames. ], batch size: 58, lr: 2.19e-02, grad_scale: 16.0 2023-04-16 15:10:50,890 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14072.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:10:57,557 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3333, 3.1437, 2.8143, 4.0904, 4.7388, 3.6727, 4.7446, 4.3093], device='cuda:1'), covar=tensor([0.0077, 0.0428, 0.0657, 0.0101, 0.0081, 0.0258, 0.0075, 0.0064], device='cuda:1'), in_proj_covar=tensor([0.0043, 0.0058, 0.0072, 0.0050, 0.0036, 0.0054, 0.0032, 0.0042], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-16 15:11:09,785 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14093.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 15:11:12,003 INFO [train.py:893] (1/4) Epoch 6, batch 600, loss[loss=0.2469, simple_loss=0.2782, pruned_loss=0.1077, over 13340.00 frames. ], tot_loss[loss=0.267, simple_loss=0.3015, pruned_loss=0.1163, over 2529773.69 frames. ], batch size: 73, lr: 2.19e-02, grad_scale: 16.0 2023-04-16 15:11:19,386 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.9309, 4.8031, 5.1022, 4.7795, 5.3090, 4.7487, 5.3352, 5.3284], device='cuda:1'), covar=tensor([0.0299, 0.0389, 0.0445, 0.0458, 0.0443, 0.0684, 0.0384, 0.0327], device='cuda:1'), in_proj_covar=tensor([0.0155, 0.0188, 0.0170, 0.0133, 0.0252, 0.0210, 0.0150, 0.0158], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0004, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:11:19,446 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14103.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:11:21,642 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.367e+02 3.788e+02 4.827e+02 5.908e+02 1.295e+03, threshold=9.655e+02, percent-clipped=4.0 2023-04-16 15:11:22,230 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 2023-04-16 15:11:34,459 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14120.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 15:11:46,713 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14133.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:11:52,470 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8755, 3.3558, 3.1109, 3.7202, 2.0062, 2.6825, 3.2836, 1.7909], device='cuda:1'), covar=tensor([0.0126, 0.0545, 0.0628, 0.0313, 0.1490, 0.0831, 0.0672, 0.2408], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0144, 0.0156, 0.0114, 0.0170, 0.0162, 0.0146, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-16 15:11:57,816 INFO [train.py:893] (1/4) Epoch 6, batch 650, loss[loss=0.2593, simple_loss=0.2932, pruned_loss=0.1127, over 13011.00 frames. ], tot_loss[loss=0.266, simple_loss=0.3004, pruned_loss=0.1158, over 2561201.17 frames. ], batch size: 142, lr: 2.19e-02, grad_scale: 8.0 2023-04-16 15:12:14,521 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14164.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:12:30,217 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14181.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 15:12:44,462 INFO [train.py:893] (1/4) Epoch 6, batch 700, loss[loss=0.2099, simple_loss=0.2597, pruned_loss=0.08006, over 13551.00 frames. ], tot_loss[loss=0.2639, simple_loss=0.2986, pruned_loss=0.1146, over 2582922.12 frames. ], batch size: 72, lr: 2.18e-02, grad_scale: 8.0 2023-04-16 15:12:53,792 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.573e+02 3.430e+02 4.164e+02 5.021e+02 1.126e+03, threshold=8.328e+02, percent-clipped=1.0 2023-04-16 15:13:13,619 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6316, 2.2671, 2.3016, 3.9637, 3.6921, 3.8201, 3.0386, 2.1454], device='cuda:1'), covar=tensor([0.0174, 0.1287, 0.1008, 0.0051, 0.0191, 0.0059, 0.0589, 0.1222], device='cuda:1'), in_proj_covar=tensor([0.0084, 0.0145, 0.0138, 0.0074, 0.0079, 0.0066, 0.0126, 0.0132], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:13:30,093 INFO [train.py:893] (1/4) Epoch 6, batch 750, loss[loss=0.2767, simple_loss=0.3144, pruned_loss=0.1195, over 13457.00 frames. ], tot_loss[loss=0.2644, simple_loss=0.2994, pruned_loss=0.1147, over 2604881.74 frames. ], batch size: 79, lr: 2.18e-02, grad_scale: 8.0 2023-04-16 15:13:54,231 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2023-04-16 15:14:17,549 INFO [train.py:893] (1/4) Epoch 6, batch 800, loss[loss=0.2394, simple_loss=0.2735, pruned_loss=0.1027, over 13438.00 frames. ], tot_loss[loss=0.2646, simple_loss=0.3, pruned_loss=0.1146, over 2620098.99 frames. ], batch size: 65, lr: 2.18e-02, grad_scale: 8.0 2023-04-16 15:14:25,820 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.150e+02 4.040e+02 4.852e+02 5.703e+02 9.161e+02, threshold=9.703e+02, percent-clipped=4.0 2023-04-16 15:14:27,768 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7524, 3.4049, 3.7313, 2.6242, 4.1696, 3.7606, 3.7880, 4.1983], device='cuda:1'), covar=tensor([0.0206, 0.0146, 0.0165, 0.0850, 0.0128, 0.0160, 0.0135, 0.0081], device='cuda:1'), in_proj_covar=tensor([0.0056, 0.0031, 0.0049, 0.0072, 0.0060, 0.0052, 0.0049, 0.0039], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:15:01,606 INFO [train.py:893] (1/4) Epoch 6, batch 850, loss[loss=0.2657, simple_loss=0.2949, pruned_loss=0.1183, over 13233.00 frames. ], tot_loss[loss=0.2657, simple_loss=0.3008, pruned_loss=0.1153, over 2631336.95 frames. ], batch size: 132, lr: 2.17e-02, grad_scale: 8.0 2023-04-16 15:15:11,478 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7663, 2.0934, 4.0092, 3.7750, 3.9232, 3.4601, 3.8695, 2.9104], device='cuda:1'), covar=tensor([0.2061, 0.1772, 0.0061, 0.0219, 0.0154, 0.0374, 0.0103, 0.0861], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0178, 0.0086, 0.0092, 0.0094, 0.0139, 0.0093, 0.0149], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-16 15:15:44,438 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14393.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:15:46,734 INFO [train.py:893] (1/4) Epoch 6, batch 900, loss[loss=0.257, simple_loss=0.2829, pruned_loss=0.1156, over 13453.00 frames. ], tot_loss[loss=0.2662, simple_loss=0.3003, pruned_loss=0.1161, over 2638720.19 frames. ], batch size: 65, lr: 2.17e-02, grad_scale: 8.0 2023-04-16 15:15:56,737 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.530e+02 4.113e+02 4.965e+02 5.706e+02 1.317e+03, threshold=9.930e+02, percent-clipped=5.0 2023-04-16 15:16:03,004 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14413.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:16:17,025 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14428.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:16:17,764 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-16 15:16:29,138 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14441.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:16:33,030 INFO [train.py:893] (1/4) Epoch 6, batch 950, loss[loss=0.2585, simple_loss=0.2892, pruned_loss=0.1139, over 13531.00 frames. ], tot_loss[loss=0.2653, simple_loss=0.2992, pruned_loss=0.1157, over 2647051.63 frames. ], batch size: 83, lr: 2.17e-02, grad_scale: 8.0 2023-04-16 15:16:39,751 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6023, 4.8956, 3.3693, 4.9426, 4.6309, 2.6738, 3.9524, 3.0266], device='cuda:1'), covar=tensor([0.0230, 0.0270, 0.1149, 0.0078, 0.0170, 0.1271, 0.0463, 0.1509], device='cuda:1'), in_proj_covar=tensor([0.0116, 0.0117, 0.0168, 0.0080, 0.0108, 0.0154, 0.0136, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 15:16:45,379 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14459.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:16:59,201 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14474.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:17:00,596 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14476.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 15:17:11,126 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14487.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:17:18,268 INFO [train.py:893] (1/4) Epoch 6, batch 1000, loss[loss=0.2336, simple_loss=0.2765, pruned_loss=0.09539, over 13533.00 frames. ], tot_loss[loss=0.2621, simple_loss=0.296, pruned_loss=0.114, over 2645564.29 frames. ], batch size: 91, lr: 2.16e-02, grad_scale: 8.0 2023-04-16 15:17:28,002 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.605e+02 3.533e+02 4.135e+02 5.114e+02 9.887e+02, threshold=8.269e+02, percent-clipped=0.0 2023-04-16 15:18:01,638 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 2023-04-16 15:18:05,130 INFO [train.py:893] (1/4) Epoch 6, batch 1050, loss[loss=0.2785, simple_loss=0.3105, pruned_loss=0.1233, over 13528.00 frames. ], tot_loss[loss=0.2611, simple_loss=0.2954, pruned_loss=0.1135, over 2648504.64 frames. ], batch size: 91, lr: 2.16e-02, grad_scale: 8.0 2023-04-16 15:18:07,794 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14548.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:18:50,782 INFO [train.py:893] (1/4) Epoch 6, batch 1100, loss[loss=0.2514, simple_loss=0.2973, pruned_loss=0.1028, over 13406.00 frames. ], tot_loss[loss=0.2603, simple_loss=0.2956, pruned_loss=0.1125, over 2651693.84 frames. ], batch size: 113, lr: 2.16e-02, grad_scale: 8.0 2023-04-16 15:19:00,911 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.517e+02 3.733e+02 4.712e+02 5.511e+02 1.792e+03, threshold=9.424e+02, percent-clipped=5.0 2023-04-16 15:19:37,949 INFO [train.py:893] (1/4) Epoch 6, batch 1150, loss[loss=0.2253, simple_loss=0.2791, pruned_loss=0.08579, over 12992.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.2953, pruned_loss=0.1114, over 2650089.04 frames. ], batch size: 142, lr: 2.15e-02, grad_scale: 8.0 2023-04-16 15:19:47,891 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0645, 3.9093, 4.0367, 2.5889, 4.5509, 4.2102, 4.2062, 4.5552], device='cuda:1'), covar=tensor([0.0186, 0.0111, 0.0141, 0.0898, 0.0118, 0.0126, 0.0119, 0.0080], device='cuda:1'), in_proj_covar=tensor([0.0057, 0.0030, 0.0050, 0.0074, 0.0061, 0.0052, 0.0049, 0.0039], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:19:57,816 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8401, 3.7207, 3.8529, 2.7097, 4.4318, 4.0523, 4.0731, 4.5422], device='cuda:1'), covar=tensor([0.0242, 0.0117, 0.0151, 0.0888, 0.0148, 0.0154, 0.0139, 0.0067], device='cuda:1'), in_proj_covar=tensor([0.0056, 0.0030, 0.0050, 0.0073, 0.0061, 0.0051, 0.0049, 0.0039], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:20:03,104 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2094, 4.8406, 4.5052, 4.4771, 4.3543, 4.2830, 4.8633, 4.8596], device='cuda:1'), covar=tensor([0.0212, 0.0205, 0.0185, 0.0313, 0.0319, 0.0414, 0.0295, 0.0184], device='cuda:1'), in_proj_covar=tensor([0.0093, 0.0134, 0.0097, 0.0126, 0.0092, 0.0130, 0.0095, 0.0102], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:20:23,469 INFO [train.py:893] (1/4) Epoch 6, batch 1200, loss[loss=0.2548, simple_loss=0.3006, pruned_loss=0.1045, over 13439.00 frames. ], tot_loss[loss=0.2583, simple_loss=0.2952, pruned_loss=0.1107, over 2654403.39 frames. ], batch size: 103, lr: 2.15e-02, grad_scale: 8.0 2023-04-16 15:20:32,943 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.652e+02 3.608e+02 4.164e+02 5.469e+02 1.026e+03, threshold=8.328e+02, percent-clipped=1.0 2023-04-16 15:20:50,203 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-16 15:20:52,842 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14728.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:21:03,278 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-16 15:21:04,313 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=14740.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:21:09,003 INFO [train.py:893] (1/4) Epoch 6, batch 1250, loss[loss=0.2669, simple_loss=0.2923, pruned_loss=0.1207, over 13554.00 frames. ], tot_loss[loss=0.2584, simple_loss=0.295, pruned_loss=0.1109, over 2657371.90 frames. ], batch size: 72, lr: 2.15e-02, grad_scale: 8.0 2023-04-16 15:21:21,646 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14759.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:21:31,330 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14769.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:21:34,826 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6781, 3.0539, 2.5521, 3.1289, 2.6366, 1.5098, 3.1352, 1.7282], device='cuda:1'), covar=tensor([0.0601, 0.0617, 0.0540, 0.0313, 0.1060, 0.1866, 0.0788, 0.1521], device='cuda:1'), in_proj_covar=tensor([0.0096, 0.0079, 0.0089, 0.0078, 0.0110, 0.0132, 0.0085, 0.0108], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-16 15:21:37,134 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14776.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:21:37,258 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=14776.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 15:21:56,342 INFO [train.py:893] (1/4) Epoch 6, batch 1300, loss[loss=0.26, simple_loss=0.2978, pruned_loss=0.1111, over 13526.00 frames. ], tot_loss[loss=0.2608, simple_loss=0.2972, pruned_loss=0.1122, over 2651932.62 frames. ], batch size: 83, lr: 2.14e-02, grad_scale: 8.0 2023-04-16 15:22:01,110 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=14801.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:22:05,026 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.718e+02 3.545e+02 4.251e+02 5.698e+02 1.080e+03, threshold=8.501e+02, percent-clipped=3.0 2023-04-16 15:22:06,045 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14807.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:22:18,392 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4339, 4.7778, 2.8242, 4.7900, 4.4935, 2.7365, 3.9926, 2.9353], device='cuda:1'), covar=tensor([0.0205, 0.0182, 0.1520, 0.0078, 0.0158, 0.1166, 0.0458, 0.1649], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0119, 0.0171, 0.0084, 0.0107, 0.0155, 0.0139, 0.0164], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 15:22:22,108 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=14824.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 15:22:22,197 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8461, 2.4570, 1.9584, 3.6806, 4.2640, 3.1375, 4.2673, 3.8004], device='cuda:1'), covar=tensor([0.0083, 0.0590, 0.0927, 0.0103, 0.0077, 0.0370, 0.0061, 0.0074], device='cuda:1'), in_proj_covar=tensor([0.0046, 0.0064, 0.0075, 0.0054, 0.0037, 0.0057, 0.0033, 0.0044], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-16 15:22:39,252 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=14843.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:22:41,603 INFO [train.py:893] (1/4) Epoch 6, batch 1350, loss[loss=0.2568, simple_loss=0.3035, pruned_loss=0.1051, over 13567.00 frames. ], tot_loss[loss=0.2618, simple_loss=0.2979, pruned_loss=0.1129, over 2655291.96 frames. ], batch size: 89, lr: 2.14e-02, grad_scale: 8.0 2023-04-16 15:23:24,748 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4525, 2.3877, 2.1991, 4.0149, 3.6691, 3.9518, 3.1965, 2.3823], device='cuda:1'), covar=tensor([0.0296, 0.1201, 0.1134, 0.0064, 0.0266, 0.0060, 0.0583, 0.1153], device='cuda:1'), in_proj_covar=tensor([0.0089, 0.0142, 0.0139, 0.0077, 0.0083, 0.0066, 0.0133, 0.0132], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:23:27,734 INFO [train.py:893] (1/4) Epoch 6, batch 1400, loss[loss=0.246, simple_loss=0.2841, pruned_loss=0.1039, over 13470.00 frames. ], tot_loss[loss=0.2606, simple_loss=0.2967, pruned_loss=0.1123, over 2653790.58 frames. ], batch size: 81, lr: 2.14e-02, grad_scale: 8.0 2023-04-16 15:23:36,719 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.557e+02 3.909e+02 4.450e+02 5.303e+02 8.590e+02, threshold=8.900e+02, percent-clipped=1.0 2023-04-16 15:24:01,835 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5885, 4.5770, 4.0255, 3.5320, 3.6650, 2.4774, 4.6214, 2.6615], device='cuda:1'), covar=tensor([0.0840, 0.0167, 0.0273, 0.0657, 0.0340, 0.1951, 0.0106, 0.2478], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0216, 0.0217, 0.0239, 0.0189, 0.0244, 0.0148, 0.0294], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 15:24:12,713 INFO [train.py:893] (1/4) Epoch 6, batch 1450, loss[loss=0.2225, simple_loss=0.2741, pruned_loss=0.08543, over 13538.00 frames. ], tot_loss[loss=0.26, simple_loss=0.296, pruned_loss=0.112, over 2658822.84 frames. ], batch size: 78, lr: 2.13e-02, grad_scale: 8.0 2023-04-16 15:24:17,824 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2526, 3.5890, 3.3327, 4.1551, 1.8539, 2.6460, 3.3988, 1.9532], device='cuda:1'), covar=tensor([0.0095, 0.0470, 0.0642, 0.0276, 0.1721, 0.0948, 0.0801, 0.1974], device='cuda:1'), in_proj_covar=tensor([0.0081, 0.0145, 0.0159, 0.0126, 0.0172, 0.0169, 0.0151, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-16 15:24:23,459 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4733, 3.8010, 3.4870, 4.4276, 2.0269, 2.8382, 3.7021, 2.0119], device='cuda:1'), covar=tensor([0.0100, 0.0474, 0.0681, 0.0243, 0.1756, 0.0984, 0.0658, 0.2313], device='cuda:1'), in_proj_covar=tensor([0.0081, 0.0145, 0.0159, 0.0125, 0.0171, 0.0168, 0.0150, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-16 15:24:57,577 INFO [train.py:893] (1/4) Epoch 6, batch 1500, loss[loss=0.2544, simple_loss=0.2875, pruned_loss=0.1107, over 13353.00 frames. ], tot_loss[loss=0.258, simple_loss=0.2947, pruned_loss=0.1107, over 2661861.32 frames. ], batch size: 73, lr: 2.13e-02, grad_scale: 8.0 2023-04-16 15:25:06,394 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.363e+02 3.329e+02 4.229e+02 4.892e+02 9.506e+02, threshold=8.458e+02, percent-clipped=2.0 2023-04-16 15:25:43,716 INFO [train.py:893] (1/4) Epoch 6, batch 1550, loss[loss=0.3031, simple_loss=0.3378, pruned_loss=0.1342, over 13357.00 frames. ], tot_loss[loss=0.2592, simple_loss=0.2958, pruned_loss=0.1113, over 2658440.02 frames. ], batch size: 118, lr: 2.13e-02, grad_scale: 8.0 2023-04-16 15:26:04,620 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15069.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:26:29,861 INFO [train.py:893] (1/4) Epoch 6, batch 1600, loss[loss=0.2344, simple_loss=0.2833, pruned_loss=0.09281, over 13546.00 frames. ], tot_loss[loss=0.2591, simple_loss=0.2958, pruned_loss=0.1112, over 2655945.22 frames. ], batch size: 78, lr: 2.12e-02, grad_scale: 8.0 2023-04-16 15:26:30,022 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15096.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:26:30,292 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.31 vs. limit=5.0 2023-04-16 15:26:39,280 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.239e+02 3.868e+02 4.961e+02 5.958e+02 8.730e+02, threshold=9.921e+02, percent-clipped=2.0 2023-04-16 15:26:49,405 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15117.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:27:13,661 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15143.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:27:15,906 INFO [train.py:893] (1/4) Epoch 6, batch 1650, loss[loss=0.2437, simple_loss=0.2704, pruned_loss=0.1085, over 12752.00 frames. ], tot_loss[loss=0.2586, simple_loss=0.2962, pruned_loss=0.1105, over 2653722.14 frames. ], batch size: 52, lr: 2.12e-02, grad_scale: 8.0 2023-04-16 15:27:37,870 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.11 vs. limit=2.0 2023-04-16 15:27:56,318 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15191.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:27:59,771 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15195.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:28:00,329 INFO [train.py:893] (1/4) Epoch 6, batch 1700, loss[loss=0.2595, simple_loss=0.2979, pruned_loss=0.1106, over 13177.00 frames. ], tot_loss[loss=0.2585, simple_loss=0.2964, pruned_loss=0.1103, over 2653574.98 frames. ], batch size: 132, lr: 2.12e-02, grad_scale: 8.0 2023-04-16 15:28:10,720 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.413e+02 3.567e+02 4.505e+02 5.770e+02 1.056e+03, threshold=9.010e+02, percent-clipped=1.0 2023-04-16 15:28:26,295 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.4225, 2.3841, 1.9684, 1.2171, 1.0825, 1.6715, 1.6685, 2.3249], device='cuda:1'), covar=tensor([0.0754, 0.0280, 0.0723, 0.1363, 0.0220, 0.0414, 0.0761, 0.0220], device='cuda:1'), in_proj_covar=tensor([0.0093, 0.0078, 0.0073, 0.0118, 0.0066, 0.0079, 0.0094, 0.0073], device='cuda:1'), out_proj_covar=tensor([7.5417e-05, 6.2187e-05, 6.2236e-05, 1.0240e-04, 6.1696e-05, 6.2769e-05, 7.6549e-05, 5.5803e-05], device='cuda:1') 2023-04-16 15:28:46,129 INFO [train.py:893] (1/4) Epoch 6, batch 1750, loss[loss=0.2443, simple_loss=0.2892, pruned_loss=0.0997, over 13472.00 frames. ], tot_loss[loss=0.2577, simple_loss=0.2959, pruned_loss=0.1098, over 2652926.27 frames. ], batch size: 79, lr: 2.11e-02, grad_scale: 8.0 2023-04-16 15:28:53,351 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-04-16 15:28:55,620 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15256.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 15:29:04,230 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7238, 2.4199, 2.6554, 4.0997, 3.8261, 4.0863, 3.3303, 2.4743], device='cuda:1'), covar=tensor([0.0221, 0.1243, 0.0989, 0.0060, 0.0172, 0.0045, 0.0465, 0.1046], device='cuda:1'), in_proj_covar=tensor([0.0086, 0.0144, 0.0142, 0.0077, 0.0083, 0.0067, 0.0134, 0.0132], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:29:32,560 INFO [train.py:893] (1/4) Epoch 6, batch 1800, loss[loss=0.2357, simple_loss=0.2856, pruned_loss=0.09289, over 13217.00 frames. ], tot_loss[loss=0.2569, simple_loss=0.2953, pruned_loss=0.1092, over 2657443.34 frames. ], batch size: 124, lr: 2.11e-02, grad_scale: 8.0 2023-04-16 15:29:40,588 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.268e+02 3.529e+02 4.378e+02 5.414e+02 8.128e+02, threshold=8.756e+02, percent-clipped=0.0 2023-04-16 15:30:17,107 INFO [train.py:893] (1/4) Epoch 6, batch 1850, loss[loss=0.257, simple_loss=0.2983, pruned_loss=0.1078, over 13526.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.2945, pruned_loss=0.1089, over 2658141.45 frames. ], batch size: 98, lr: 2.11e-02, grad_scale: 8.0 2023-04-16 15:30:18,841 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-16 15:30:46,932 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1851, 4.1093, 3.5874, 3.2646, 3.1519, 2.2929, 4.2420, 2.5709], device='cuda:1'), covar=tensor([0.0867, 0.0226, 0.0353, 0.0712, 0.0430, 0.2267, 0.0113, 0.2451], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0217, 0.0221, 0.0245, 0.0191, 0.0242, 0.0149, 0.0297], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 15:30:58,997 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.86 vs. limit=5.0 2023-04-16 15:31:03,341 INFO [train.py:893] (1/4) Epoch 6, batch 1900, loss[loss=0.2863, simple_loss=0.3128, pruned_loss=0.1299, over 13548.00 frames. ], tot_loss[loss=0.2553, simple_loss=0.2932, pruned_loss=0.1087, over 2655950.73 frames. ], batch size: 87, lr: 2.10e-02, grad_scale: 8.0 2023-04-16 15:31:03,632 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15396.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:31:12,711 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.431e+02 3.722e+02 4.317e+02 5.433e+02 1.645e+03, threshold=8.635e+02, percent-clipped=4.0 2023-04-16 15:31:17,444 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2023-04-16 15:31:46,995 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15444.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:31:48,464 INFO [train.py:893] (1/4) Epoch 6, batch 1950, loss[loss=0.2567, simple_loss=0.2936, pruned_loss=0.1099, over 13540.00 frames. ], tot_loss[loss=0.2549, simple_loss=0.2927, pruned_loss=0.1086, over 2658344.66 frames. ], batch size: 83, lr: 2.10e-02, grad_scale: 8.0 2023-04-16 15:32:22,614 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.54 vs. limit=5.0 2023-04-16 15:32:27,444 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7226, 2.5579, 2.6279, 4.0306, 3.7296, 3.9189, 3.1550, 2.5122], device='cuda:1'), covar=tensor([0.0181, 0.1220, 0.0975, 0.0043, 0.0210, 0.0054, 0.0618, 0.1006], device='cuda:1'), in_proj_covar=tensor([0.0088, 0.0149, 0.0145, 0.0077, 0.0085, 0.0068, 0.0140, 0.0133], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:32:28,342 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6325, 4.1455, 4.3602, 3.1085, 2.7302, 3.1955, 4.5070, 4.5777], device='cuda:1'), covar=tensor([0.0738, 0.0518, 0.0283, 0.1360, 0.1521, 0.1033, 0.0143, 0.0079], device='cuda:1'), in_proj_covar=tensor([0.0188, 0.0150, 0.0139, 0.0188, 0.0189, 0.0152, 0.0116, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0001], device='cuda:1') 2023-04-16 15:32:34,553 INFO [train.py:893] (1/4) Epoch 6, batch 2000, loss[loss=0.2603, simple_loss=0.3019, pruned_loss=0.1094, over 13336.00 frames. ], tot_loss[loss=0.257, simple_loss=0.2945, pruned_loss=0.1098, over 2658464.70 frames. ], batch size: 118, lr: 2.10e-02, grad_scale: 8.0 2023-04-16 15:32:40,170 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-16 15:32:43,470 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.298e+02 3.534e+02 4.151e+02 5.068e+02 1.089e+03, threshold=8.302e+02, percent-clipped=2.0 2023-04-16 15:33:20,626 INFO [train.py:893] (1/4) Epoch 6, batch 2050, loss[loss=0.222, simple_loss=0.2606, pruned_loss=0.09166, over 13491.00 frames. ], tot_loss[loss=0.2597, simple_loss=0.2968, pruned_loss=0.1113, over 2654741.28 frames. ], batch size: 70, lr: 2.10e-02, grad_scale: 8.0 2023-04-16 15:33:24,861 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=15551.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 15:34:05,169 INFO [train.py:893] (1/4) Epoch 6, batch 2100, loss[loss=0.2335, simple_loss=0.278, pruned_loss=0.09453, over 13482.00 frames. ], tot_loss[loss=0.2589, simple_loss=0.2964, pruned_loss=0.1107, over 2654333.51 frames. ], batch size: 70, lr: 2.09e-02, grad_scale: 8.0 2023-04-16 15:34:15,328 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.728e+02 3.686e+02 4.329e+02 5.255e+02 8.476e+02, threshold=8.658e+02, percent-clipped=1.0 2023-04-16 15:34:51,927 INFO [train.py:893] (1/4) Epoch 6, batch 2150, loss[loss=0.2681, simple_loss=0.2946, pruned_loss=0.1207, over 13566.00 frames. ], tot_loss[loss=0.2572, simple_loss=0.2952, pruned_loss=0.1096, over 2653016.27 frames. ], batch size: 72, lr: 2.09e-02, grad_scale: 8.0 2023-04-16 15:35:26,597 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 2023-04-16 15:35:36,596 INFO [train.py:893] (1/4) Epoch 6, batch 2200, loss[loss=0.2427, simple_loss=0.2849, pruned_loss=0.1003, over 13524.00 frames. ], tot_loss[loss=0.256, simple_loss=0.2944, pruned_loss=0.1089, over 2653171.04 frames. ], batch size: 85, lr: 2.09e-02, grad_scale: 8.0 2023-04-16 15:35:46,069 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.020e+02 3.661e+02 4.082e+02 5.276e+02 1.095e+03, threshold=8.164e+02, percent-clipped=2.0 2023-04-16 15:35:53,444 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2023-04-16 15:35:56,403 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.0428, 4.6141, 4.5872, 4.5059, 4.1155, 4.3302, 4.9790, 4.4871], device='cuda:1'), covar=tensor([0.0803, 0.0898, 0.2040, 0.2631, 0.0716, 0.1344, 0.0870, 0.1013], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0256, 0.0327, 0.0338, 0.0177, 0.0255, 0.0307, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 15:36:13,164 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2207, 4.6522, 4.3380, 4.3304, 4.3119, 4.2580, 4.7187, 4.6339], device='cuda:1'), covar=tensor([0.0181, 0.0201, 0.0180, 0.0263, 0.0205, 0.0259, 0.0185, 0.0199], device='cuda:1'), in_proj_covar=tensor([0.0098, 0.0141, 0.0105, 0.0129, 0.0097, 0.0135, 0.0099, 0.0111], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:36:19,102 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-04-16 15:36:21,885 INFO [train.py:893] (1/4) Epoch 6, batch 2250, loss[loss=0.25, simple_loss=0.2897, pruned_loss=0.1052, over 13378.00 frames. ], tot_loss[loss=0.2554, simple_loss=0.2938, pruned_loss=0.1085, over 2654946.80 frames. ], batch size: 84, lr: 2.08e-02, grad_scale: 8.0 2023-04-16 15:37:08,042 INFO [train.py:893] (1/4) Epoch 6, batch 2300, loss[loss=0.2377, simple_loss=0.2817, pruned_loss=0.09681, over 13417.00 frames. ], tot_loss[loss=0.2529, simple_loss=0.2916, pruned_loss=0.1071, over 2651562.22 frames. ], batch size: 65, lr: 2.08e-02, grad_scale: 8.0 2023-04-16 15:37:09,376 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-04-16 15:37:16,719 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.314e+02 3.478e+02 4.218e+02 5.479e+02 1.103e+03, threshold=8.435e+02, percent-clipped=4.0 2023-04-16 15:37:52,530 INFO [train.py:893] (1/4) Epoch 6, batch 2350, loss[loss=0.2562, simple_loss=0.2925, pruned_loss=0.11, over 13238.00 frames. ], tot_loss[loss=0.2528, simple_loss=0.2918, pruned_loss=0.1069, over 2659552.75 frames. ], batch size: 124, lr: 2.08e-02, grad_scale: 8.0 2023-04-16 15:37:57,494 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=15851.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:38:16,815 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-16 15:38:26,886 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=15883.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:38:38,115 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 2023-04-16 15:38:39,142 INFO [train.py:893] (1/4) Epoch 6, batch 2400, loss[loss=0.2341, simple_loss=0.2707, pruned_loss=0.09869, over 13216.00 frames. ], tot_loss[loss=0.2519, simple_loss=0.2907, pruned_loss=0.1065, over 2662776.22 frames. ], batch size: 132, lr: 2.07e-02, grad_scale: 8.0 2023-04-16 15:38:41,816 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=15899.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:38:47,378 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.949e+02 3.560e+02 4.458e+02 5.993e+02 1.349e+03, threshold=8.916e+02, percent-clipped=5.0 2023-04-16 15:39:22,656 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=15944.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:39:24,046 INFO [train.py:893] (1/4) Epoch 6, batch 2450, loss[loss=0.2242, simple_loss=0.2678, pruned_loss=0.09024, over 13188.00 frames. ], tot_loss[loss=0.2514, simple_loss=0.2905, pruned_loss=0.1061, over 2665968.98 frames. ], batch size: 58, lr: 2.07e-02, grad_scale: 8.0 2023-04-16 15:40:04,110 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6610, 3.1397, 2.5313, 3.0632, 2.5754, 1.6842, 3.2375, 2.0252], device='cuda:1'), covar=tensor([0.0604, 0.0592, 0.0476, 0.0338, 0.0777, 0.1806, 0.0734, 0.1165], device='cuda:1'), in_proj_covar=tensor([0.0105, 0.0084, 0.0096, 0.0081, 0.0113, 0.0138, 0.0093, 0.0112], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:40:10,319 INFO [train.py:893] (1/4) Epoch 6, batch 2500, loss[loss=0.2293, simple_loss=0.2816, pruned_loss=0.08851, over 13503.00 frames. ], tot_loss[loss=0.251, simple_loss=0.2904, pruned_loss=0.1058, over 2666672.56 frames. ], batch size: 81, lr: 2.07e-02, grad_scale: 8.0 2023-04-16 15:40:17,875 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7793, 3.2786, 2.5486, 3.1436, 2.5620, 1.6400, 3.1165, 1.9009], device='cuda:1'), covar=tensor([0.0704, 0.0734, 0.0665, 0.0366, 0.1041, 0.2242, 0.1639, 0.1659], device='cuda:1'), in_proj_covar=tensor([0.0107, 0.0085, 0.0098, 0.0082, 0.0116, 0.0141, 0.0095, 0.0115], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:40:18,613 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16001.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:40:23,169 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.821e+02 3.155e+02 3.869e+02 4.967e+02 1.311e+03, threshold=7.737e+02, percent-clipped=1.0 2023-04-16 15:40:59,434 INFO [train.py:893] (1/4) Epoch 6, batch 2550, loss[loss=0.2503, simple_loss=0.2907, pruned_loss=0.1049, over 13475.00 frames. ], tot_loss[loss=0.2517, simple_loss=0.2906, pruned_loss=0.1064, over 2664149.56 frames. ], batch size: 93, lr: 2.07e-02, grad_scale: 8.0 2023-04-16 15:41:06,444 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.97 vs. limit=2.0 2023-04-16 15:41:14,421 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16062.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:41:23,968 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-16 15:41:36,216 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16085.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:41:44,943 INFO [train.py:893] (1/4) Epoch 6, batch 2600, loss[loss=0.2424, simple_loss=0.2855, pruned_loss=0.09972, over 13550.00 frames. ], tot_loss[loss=0.2526, simple_loss=0.2911, pruned_loss=0.1071, over 2666903.55 frames. ], batch size: 72, lr: 2.06e-02, grad_scale: 8.0 2023-04-16 15:41:53,824 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.362e+02 4.100e+02 4.621e+02 5.666e+02 1.546e+03, threshold=9.242e+02, percent-clipped=11.0 2023-04-16 15:42:00,652 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9482, 4.3891, 4.0673, 4.1364, 4.1392, 4.0097, 4.4917, 4.4514], device='cuda:1'), covar=tensor([0.0243, 0.0228, 0.0231, 0.0319, 0.0301, 0.0254, 0.0255, 0.0243], device='cuda:1'), in_proj_covar=tensor([0.0095, 0.0138, 0.0103, 0.0128, 0.0094, 0.0132, 0.0095, 0.0111], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:42:26,556 INFO [train.py:893] (1/4) Epoch 6, batch 2650, loss[loss=0.2271, simple_loss=0.2724, pruned_loss=0.09088, over 13460.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.2918, pruned_loss=0.1079, over 2666633.54 frames. ], batch size: 79, lr: 2.06e-02, grad_scale: 16.0 2023-04-16 15:42:26,797 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16146.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:42:29,028 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16149.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:42:46,981 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 2023-04-16 15:43:00,306 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2592, 3.6468, 3.1763, 4.0321, 1.7279, 2.7174, 3.4886, 1.8990], device='cuda:1'), covar=tensor([0.0087, 0.0424, 0.0739, 0.0339, 0.1699, 0.0860, 0.0631, 0.2051], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0148, 0.0163, 0.0131, 0.0171, 0.0170, 0.0148, 0.0170], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0003], device='cuda:1') 2023-04-16 15:43:23,487 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 15:43:33,344 INFO [train.py:893] (1/4) Epoch 7, batch 0, loss[loss=0.2426, simple_loss=0.2811, pruned_loss=0.102, over 13510.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.2811, pruned_loss=0.102, over 13510.00 frames. ], batch size: 81, lr: 1.93e-02, grad_scale: 16.0 2023-04-16 15:43:33,344 INFO [train.py:918] (1/4) Computing validation loss 2023-04-16 15:43:55,951 INFO [train.py:927] (1/4) Epoch 7, validation: loss=0.1848, simple_loss=0.2348, pruned_loss=0.06743, over 2446609.00 frames. 2023-04-16 15:43:55,952 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12761MB 2023-04-16 15:44:05,322 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.167e+02 3.559e+02 4.435e+02 5.202e+02 1.050e+03, threshold=8.871e+02, percent-clipped=1.0 2023-04-16 15:44:08,895 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16210.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:44:33,503 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-04-16 15:44:35,544 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16239.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:44:40,952 INFO [train.py:893] (1/4) Epoch 7, batch 50, loss[loss=0.2113, simple_loss=0.2613, pruned_loss=0.08067, over 13488.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.2845, pruned_loss=0.1043, over 600998.37 frames. ], batch size: 81, lr: 1.92e-02, grad_scale: 16.0 2023-04-16 15:44:50,500 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0 2023-04-16 15:45:05,844 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-16 15:45:05,845 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-16 15:45:05,845 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-16 15:45:05,851 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-16 15:45:05,866 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-16 15:45:05,886 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-16 15:45:06,623 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-16 15:45:27,290 INFO [train.py:893] (1/4) Epoch 7, batch 100, loss[loss=0.2518, simple_loss=0.2877, pruned_loss=0.1079, over 13475.00 frames. ], tot_loss[loss=0.2487, simple_loss=0.2857, pruned_loss=0.1058, over 1059572.25 frames. ], batch size: 81, lr: 1.92e-02, grad_scale: 16.0 2023-04-16 15:45:36,291 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-04-16 15:45:36,610 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.237e+02 3.339e+02 4.034e+02 5.230e+02 9.968e+02, threshold=8.069e+02, percent-clipped=1.0 2023-04-16 15:45:42,663 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-04-16 15:46:04,699 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.1575, 4.7318, 4.6094, 4.5419, 4.3479, 4.5252, 5.1021, 4.5986], device='cuda:1'), covar=tensor([0.0712, 0.0963, 0.1975, 0.3009, 0.0832, 0.1432, 0.0911, 0.0950], device='cuda:1'), in_proj_covar=tensor([0.0227, 0.0260, 0.0332, 0.0341, 0.0180, 0.0258, 0.0304, 0.0226], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 15:46:12,555 INFO [train.py:893] (1/4) Epoch 7, batch 150, loss[loss=0.2735, simple_loss=0.3141, pruned_loss=0.1164, over 13520.00 frames. ], tot_loss[loss=0.2538, simple_loss=0.29, pruned_loss=0.1088, over 1408610.88 frames. ], batch size: 85, lr: 1.92e-02, grad_scale: 16.0 2023-04-16 15:46:18,288 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.4767, 5.2509, 5.6183, 5.2370, 5.7325, 5.1932, 5.6972, 5.7469], device='cuda:1'), covar=tensor([0.0249, 0.0400, 0.0405, 0.0389, 0.0400, 0.0568, 0.0453, 0.0382], device='cuda:1'), in_proj_covar=tensor([0.0166, 0.0202, 0.0182, 0.0141, 0.0264, 0.0223, 0.0165, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0004, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:46:24,676 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16357.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:46:36,869 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16371.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:46:48,905 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0098, 4.4998, 4.2008, 4.2268, 4.1672, 4.0890, 4.5645, 4.5388], device='cuda:1'), covar=tensor([0.0221, 0.0211, 0.0188, 0.0246, 0.0272, 0.0238, 0.0223, 0.0187], device='cuda:1'), in_proj_covar=tensor([0.0097, 0.0140, 0.0105, 0.0130, 0.0097, 0.0133, 0.0098, 0.0112], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:46:49,743 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6805, 3.6112, 3.6128, 2.4128, 4.0897, 3.8290, 3.9081, 4.0379], device='cuda:1'), covar=tensor([0.0308, 0.0156, 0.0183, 0.1241, 0.0236, 0.0227, 0.0170, 0.0194], device='cuda:1'), in_proj_covar=tensor([0.0057, 0.0031, 0.0052, 0.0076, 0.0064, 0.0054, 0.0052, 0.0042], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:46:58,252 INFO [train.py:893] (1/4) Epoch 7, batch 200, loss[loss=0.2567, simple_loss=0.2917, pruned_loss=0.1108, over 13534.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.292, pruned_loss=0.1104, over 1686547.44 frames. ], batch size: 87, lr: 1.92e-02, grad_scale: 16.0 2023-04-16 15:47:08,041 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.396e+02 3.451e+02 4.179e+02 5.669e+02 1.465e+03, threshold=8.357e+02, percent-clipped=6.0 2023-04-16 15:47:25,085 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1415, 2.8393, 1.8870, 4.0089, 4.6781, 3.3789, 4.5873, 4.1754], device='cuda:1'), covar=tensor([0.0078, 0.0544, 0.0929, 0.0089, 0.0046, 0.0335, 0.0059, 0.0066], device='cuda:1'), in_proj_covar=tensor([0.0048, 0.0065, 0.0076, 0.0054, 0.0038, 0.0059, 0.0036, 0.0046], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:47:31,725 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16432.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:47:40,283 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16441.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:47:44,102 INFO [train.py:893] (1/4) Epoch 7, batch 250, loss[loss=0.2337, simple_loss=0.2847, pruned_loss=0.09133, over 13533.00 frames. ], tot_loss[loss=0.2564, simple_loss=0.2928, pruned_loss=0.11, over 1902590.32 frames. ], batch size: 91, lr: 1.91e-02, grad_scale: 16.0 2023-04-16 15:47:56,571 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-04-16 15:48:06,791 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.9398, 4.7672, 5.0794, 4.8559, 5.2822, 4.7210, 5.2365, 5.2596], device='cuda:1'), covar=tensor([0.0287, 0.0495, 0.0528, 0.0483, 0.0501, 0.0735, 0.0460, 0.0409], device='cuda:1'), in_proj_covar=tensor([0.0170, 0.0207, 0.0187, 0.0145, 0.0272, 0.0226, 0.0167, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0004, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:48:28,651 INFO [train.py:893] (1/4) Epoch 7, batch 300, loss[loss=0.255, simple_loss=0.2854, pruned_loss=0.1122, over 13528.00 frames. ], tot_loss[loss=0.2561, simple_loss=0.2933, pruned_loss=0.1095, over 2075461.76 frames. ], batch size: 72, lr: 1.91e-02, grad_scale: 16.0 2023-04-16 15:48:38,373 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16505.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:48:38,555 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4126, 3.6882, 3.3347, 4.0715, 2.1648, 2.8862, 3.7348, 1.8673], device='cuda:1'), covar=tensor([0.0092, 0.0482, 0.0762, 0.0275, 0.1711, 0.1073, 0.0596, 0.2298], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0150, 0.0164, 0.0135, 0.0170, 0.0171, 0.0151, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 15:48:39,013 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.699e+02 3.423e+02 4.208e+02 4.999e+02 1.353e+03, threshold=8.417e+02, percent-clipped=5.0 2023-04-16 15:48:55,172 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8841, 4.1254, 4.6438, 3.4464, 2.9509, 3.3770, 4.7761, 4.8997], device='cuda:1'), covar=tensor([0.0667, 0.0528, 0.0182, 0.1060, 0.1248, 0.0757, 0.0124, 0.0049], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0163, 0.0143, 0.0190, 0.0189, 0.0153, 0.0124, 0.0093], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0001], device='cuda:1') 2023-04-16 15:49:08,844 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16539.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:49:14,156 INFO [train.py:893] (1/4) Epoch 7, batch 350, loss[loss=0.2349, simple_loss=0.2803, pruned_loss=0.09469, over 13361.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.2936, pruned_loss=0.1095, over 2197876.77 frames. ], batch size: 73, lr: 1.91e-02, grad_scale: 16.0 2023-04-16 15:49:18,672 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1868, 2.0652, 2.0561, 3.3756, 3.2889, 3.4640, 2.6734, 1.9392], device='cuda:1'), covar=tensor([0.0212, 0.1139, 0.0990, 0.0086, 0.0236, 0.0058, 0.0628, 0.1188], device='cuda:1'), in_proj_covar=tensor([0.0089, 0.0142, 0.0138, 0.0073, 0.0084, 0.0066, 0.0134, 0.0132], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:49:51,234 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16587.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:49:55,156 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.9292, 4.4807, 4.3405, 4.3961, 4.2690, 4.3225, 4.8775, 4.4081], device='cuda:1'), covar=tensor([0.1069, 0.1241, 0.3146, 0.3034, 0.0706, 0.1518, 0.1032, 0.1035], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0257, 0.0334, 0.0341, 0.0178, 0.0257, 0.0303, 0.0221], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 15:49:58,975 INFO [train.py:893] (1/4) Epoch 7, batch 400, loss[loss=0.248, simple_loss=0.3034, pruned_loss=0.09628, over 13274.00 frames. ], tot_loss[loss=0.2565, simple_loss=0.2939, pruned_loss=0.1095, over 2299973.27 frames. ], batch size: 124, lr: 1.91e-02, grad_scale: 16.0 2023-04-16 15:50:00,005 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.7311, 5.3203, 5.2824, 5.1374, 4.8098, 5.1351, 5.6452, 5.0755], device='cuda:1'), covar=tensor([0.0608, 0.0909, 0.2115, 0.2702, 0.0646, 0.1320, 0.0847, 0.1011], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0258, 0.0334, 0.0343, 0.0179, 0.0258, 0.0304, 0.0222], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 15:50:09,787 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.481e+02 3.841e+02 4.424e+02 5.759e+02 1.128e+03, threshold=8.848e+02, percent-clipped=5.0 2023-04-16 15:50:18,582 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.4640, 1.9977, 1.6392, 2.3986, 1.6145, 2.2837, 2.1823, 2.1516], device='cuda:1'), covar=tensor([0.0080, 0.0197, 0.0180, 0.0135, 0.0194, 0.0138, 0.0235, 0.0207], device='cuda:1'), in_proj_covar=tensor([0.0050, 0.0052, 0.0058, 0.0048, 0.0063, 0.0047, 0.0056, 0.0054], device='cuda:1'), out_proj_covar=tensor([5.8801e-05, 6.2130e-05, 7.2282e-05, 5.9442e-05, 7.6966e-05, 5.5992e-05, 7.0541e-05, 6.5853e-05], device='cuda:1') 2023-04-16 15:50:45,081 INFO [train.py:893] (1/4) Epoch 7, batch 450, loss[loss=0.2603, simple_loss=0.2918, pruned_loss=0.1144, over 13527.00 frames. ], tot_loss[loss=0.2568, simple_loss=0.2946, pruned_loss=0.1095, over 2383505.02 frames. ], batch size: 76, lr: 1.90e-02, grad_scale: 16.0 2023-04-16 15:50:56,867 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16657.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:51:08,816 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-16 15:51:30,838 INFO [train.py:893] (1/4) Epoch 7, batch 500, loss[loss=0.2663, simple_loss=0.3063, pruned_loss=0.1131, over 13378.00 frames. ], tot_loss[loss=0.2563, simple_loss=0.2943, pruned_loss=0.1091, over 2444956.42 frames. ], batch size: 109, lr: 1.90e-02, grad_scale: 16.0 2023-04-16 15:51:40,736 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16705.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:51:41,375 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.243e+02 3.362e+02 3.949e+02 4.886e+02 8.592e+02, threshold=7.898e+02, percent-clipped=0.0 2023-04-16 15:52:00,106 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=16727.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:52:13,143 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16741.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:52:16,922 INFO [train.py:893] (1/4) Epoch 7, batch 550, loss[loss=0.2297, simple_loss=0.2752, pruned_loss=0.09213, over 13467.00 frames. ], tot_loss[loss=0.256, simple_loss=0.2943, pruned_loss=0.1088, over 2488906.86 frames. ], batch size: 106, lr: 1.90e-02, grad_scale: 16.0 2023-04-16 15:52:19,135 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.68 vs. limit=2.0 2023-04-16 15:52:32,881 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 2023-04-16 15:52:38,157 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-04-16 15:52:51,901 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-04-16 15:52:56,386 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16789.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:53:01,815 INFO [train.py:893] (1/4) Epoch 7, batch 600, loss[loss=0.213, simple_loss=0.2493, pruned_loss=0.0883, over 12473.00 frames. ], tot_loss[loss=0.2543, simple_loss=0.2925, pruned_loss=0.108, over 2527772.93 frames. ], batch size: 51, lr: 1.90e-02, grad_scale: 16.0 2023-04-16 15:53:11,227 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=16805.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:53:11,749 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.461e+02 3.645e+02 4.474e+02 6.050e+02 1.289e+03, threshold=8.949e+02, percent-clipped=7.0 2023-04-16 15:53:26,009 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.24 vs. limit=5.0 2023-04-16 15:53:47,770 INFO [train.py:893] (1/4) Epoch 7, batch 650, loss[loss=0.2574, simple_loss=0.2972, pruned_loss=0.1088, over 13421.00 frames. ], tot_loss[loss=0.2548, simple_loss=0.293, pruned_loss=0.1083, over 2559148.57 frames. ], batch size: 95, lr: 1.89e-02, grad_scale: 16.0 2023-04-16 15:53:54,704 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=16853.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:53:56,545 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6974, 3.3512, 3.7276, 2.4579, 4.0335, 3.7060, 3.7799, 4.0005], device='cuda:1'), covar=tensor([0.0181, 0.0122, 0.0119, 0.0782, 0.0112, 0.0144, 0.0116, 0.0084], device='cuda:1'), in_proj_covar=tensor([0.0057, 0.0031, 0.0051, 0.0075, 0.0062, 0.0054, 0.0052, 0.0042], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:54:08,643 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16868.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 15:54:21,454 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5475, 4.4929, 3.8099, 3.3300, 3.6532, 2.4649, 4.7473, 2.7433], device='cuda:1'), covar=tensor([0.1037, 0.0227, 0.0486, 0.0894, 0.0414, 0.2581, 0.0135, 0.2735], device='cuda:1'), in_proj_covar=tensor([0.0263, 0.0225, 0.0224, 0.0247, 0.0197, 0.0247, 0.0156, 0.0301], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 15:54:32,815 INFO [train.py:893] (1/4) Epoch 7, batch 700, loss[loss=0.2329, simple_loss=0.2756, pruned_loss=0.0951, over 13565.00 frames. ], tot_loss[loss=0.253, simple_loss=0.2915, pruned_loss=0.1072, over 2579002.74 frames. ], batch size: 78, lr: 1.89e-02, grad_scale: 16.0 2023-04-16 15:54:42,799 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.357e+02 3.467e+02 4.354e+02 5.050e+02 1.182e+03, threshold=8.708e+02, percent-clipped=3.0 2023-04-16 15:54:43,735 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=16906.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:55:04,119 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16929.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 15:55:18,215 INFO [train.py:893] (1/4) Epoch 7, batch 750, loss[loss=0.2453, simple_loss=0.289, pruned_loss=0.1008, over 13364.00 frames. ], tot_loss[loss=0.2524, simple_loss=0.2907, pruned_loss=0.107, over 2598388.00 frames. ], batch size: 109, lr: 1.89e-02, grad_scale: 16.0 2023-04-16 15:55:23,793 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2023-04-16 15:55:38,935 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=16967.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 15:56:04,365 INFO [train.py:893] (1/4) Epoch 7, batch 800, loss[loss=0.2664, simple_loss=0.3084, pruned_loss=0.1123, over 13376.00 frames. ], tot_loss[loss=0.2525, simple_loss=0.2911, pruned_loss=0.1069, over 2613578.40 frames. ], batch size: 113, lr: 1.89e-02, grad_scale: 16.0 2023-04-16 15:56:13,895 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.339e+02 3.609e+02 4.209e+02 5.245e+02 1.076e+03, threshold=8.417e+02, percent-clipped=1.0 2023-04-16 15:56:26,209 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5628, 2.2300, 2.5879, 3.8944, 3.4470, 3.8537, 3.1530, 2.0992], device='cuda:1'), covar=tensor([0.0217, 0.1220, 0.0875, 0.0053, 0.0344, 0.0050, 0.0552, 0.1297], device='cuda:1'), in_proj_covar=tensor([0.0088, 0.0146, 0.0141, 0.0075, 0.0090, 0.0065, 0.0140, 0.0134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:56:32,694 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17027.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:56:44,807 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.0208, 4.9435, 5.2187, 4.9473, 5.3940, 4.9187, 5.3727, 5.4079], device='cuda:1'), covar=tensor([0.0296, 0.0451, 0.0507, 0.0432, 0.0486, 0.0683, 0.0484, 0.0435], device='cuda:1'), in_proj_covar=tensor([0.0174, 0.0209, 0.0189, 0.0147, 0.0275, 0.0232, 0.0170, 0.0182], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 15:56:48,019 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6286, 4.1018, 3.7724, 3.8497, 3.9654, 4.2660, 4.1235, 3.9343], device='cuda:1'), covar=tensor([0.0345, 0.0255, 0.0284, 0.1181, 0.0238, 0.0286, 0.0228, 0.0445], device='cuda:1'), in_proj_covar=tensor([0.0101, 0.0106, 0.0110, 0.0196, 0.0110, 0.0124, 0.0108, 0.0087], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-16 15:56:49,411 INFO [train.py:893] (1/4) Epoch 7, batch 850, loss[loss=0.2639, simple_loss=0.2983, pruned_loss=0.1147, over 13550.00 frames. ], tot_loss[loss=0.2533, simple_loss=0.292, pruned_loss=0.1073, over 2625184.31 frames. ], batch size: 87, lr: 1.88e-02, grad_scale: 16.0 2023-04-16 15:57:09,927 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0203, 4.3554, 2.5109, 4.5108, 4.1885, 2.2437, 3.5078, 2.4772], device='cuda:1'), covar=tensor([0.0316, 0.0336, 0.1927, 0.0162, 0.0272, 0.1889, 0.0736, 0.2500], device='cuda:1'), in_proj_covar=tensor([0.0127, 0.0124, 0.0169, 0.0090, 0.0112, 0.0155, 0.0141, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 15:57:16,801 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17075.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:57:35,255 INFO [train.py:893] (1/4) Epoch 7, batch 900, loss[loss=0.2618, simple_loss=0.2962, pruned_loss=0.1137, over 13379.00 frames. ], tot_loss[loss=0.2539, simple_loss=0.2926, pruned_loss=0.1076, over 2636054.78 frames. ], batch size: 113, lr: 1.88e-02, grad_scale: 16.0 2023-04-16 15:57:44,214 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0709, 2.8856, 2.0682, 3.8533, 4.5817, 3.2499, 4.4543, 4.0862], device='cuda:1'), covar=tensor([0.0094, 0.0575, 0.1002, 0.0109, 0.0061, 0.0371, 0.0092, 0.0077], device='cuda:1'), in_proj_covar=tensor([0.0050, 0.0066, 0.0077, 0.0056, 0.0040, 0.0060, 0.0037, 0.0047], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:57:44,757 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.319e+02 3.791e+02 4.573e+02 5.198e+02 1.092e+03, threshold=9.145e+02, percent-clipped=2.0 2023-04-16 15:57:57,546 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-16 15:58:03,489 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-16 15:58:19,320 INFO [train.py:893] (1/4) Epoch 7, batch 950, loss[loss=0.2661, simple_loss=0.2972, pruned_loss=0.1175, over 13486.00 frames. ], tot_loss[loss=0.2535, simple_loss=0.2914, pruned_loss=0.1078, over 2643004.49 frames. ], batch size: 81, lr: 1.88e-02, grad_scale: 16.0 2023-04-16 15:58:32,197 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1642, 4.5335, 4.2622, 4.2568, 4.1452, 4.1306, 4.5796, 4.5762], device='cuda:1'), covar=tensor([0.0187, 0.0216, 0.0188, 0.0283, 0.0286, 0.0250, 0.0264, 0.0212], device='cuda:1'), in_proj_covar=tensor([0.0098, 0.0147, 0.0109, 0.0135, 0.0100, 0.0139, 0.0102, 0.0115], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 15:58:59,595 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17189.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:59:04,945 INFO [train.py:893] (1/4) Epoch 7, batch 1000, loss[loss=0.2756, simple_loss=0.3097, pruned_loss=0.1207, over 13578.00 frames. ], tot_loss[loss=0.251, simple_loss=0.2888, pruned_loss=0.1066, over 2647398.15 frames. ], batch size: 89, lr: 1.88e-02, grad_scale: 16.0 2023-04-16 15:59:14,952 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.147e+02 3.772e+02 4.502e+02 5.538e+02 9.283e+02, threshold=9.005e+02, percent-clipped=1.0 2023-04-16 15:59:31,939 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17224.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 15:59:50,063 INFO [train.py:893] (1/4) Epoch 7, batch 1050, loss[loss=0.2283, simple_loss=0.2831, pruned_loss=0.0867, over 13485.00 frames. ], tot_loss[loss=0.2479, simple_loss=0.2862, pruned_loss=0.1048, over 2648006.69 frames. ], batch size: 81, lr: 1.87e-02, grad_scale: 16.0 2023-04-16 15:59:55,410 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17250.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 15:59:59,372 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.28 vs. limit=5.0 2023-04-16 16:00:06,301 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17262.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 16:00:07,542 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.71 vs. limit=5.0 2023-04-16 16:00:10,388 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0774, 4.5015, 4.0718, 4.1405, 4.1855, 4.6888, 4.4312, 4.3258], device='cuda:1'), covar=tensor([0.0302, 0.0258, 0.0302, 0.0958, 0.0259, 0.0200, 0.0226, 0.0214], device='cuda:1'), in_proj_covar=tensor([0.0099, 0.0105, 0.0109, 0.0194, 0.0108, 0.0122, 0.0107, 0.0085], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 16:00:36,190 INFO [train.py:893] (1/4) Epoch 7, batch 1100, loss[loss=0.2452, simple_loss=0.2869, pruned_loss=0.1017, over 13253.00 frames. ], tot_loss[loss=0.2468, simple_loss=0.286, pruned_loss=0.1038, over 2649716.72 frames. ], batch size: 124, lr: 1.87e-02, grad_scale: 16.0 2023-04-16 16:00:46,005 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.333e+02 3.290e+02 4.098e+02 5.268e+02 1.127e+03, threshold=8.196e+02, percent-clipped=3.0 2023-04-16 16:01:21,991 INFO [train.py:893] (1/4) Epoch 7, batch 1150, loss[loss=0.2334, simple_loss=0.2782, pruned_loss=0.09429, over 13575.00 frames. ], tot_loss[loss=0.2454, simple_loss=0.2854, pruned_loss=0.1027, over 2649912.80 frames. ], batch size: 89, lr: 1.87e-02, grad_scale: 16.0 2023-04-16 16:01:25,501 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.90 vs. limit=2.0 2023-04-16 16:01:37,389 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17362.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:01:55,045 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8782, 3.5059, 3.8817, 2.8369, 4.1816, 3.9263, 3.8095, 4.2023], device='cuda:1'), covar=tensor([0.0172, 0.0153, 0.0122, 0.0750, 0.0136, 0.0141, 0.0147, 0.0091], device='cuda:1'), in_proj_covar=tensor([0.0058, 0.0032, 0.0053, 0.0078, 0.0065, 0.0057, 0.0055, 0.0044], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:01:56,710 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17383.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 16:02:06,940 INFO [train.py:893] (1/4) Epoch 7, batch 1200, loss[loss=0.2489, simple_loss=0.2927, pruned_loss=0.1025, over 13526.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.2862, pruned_loss=0.1025, over 2653744.43 frames. ], batch size: 91, lr: 1.87e-02, grad_scale: 16.0 2023-04-16 16:02:17,748 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.320e+02 3.565e+02 4.089e+02 4.949e+02 1.181e+03, threshold=8.177e+02, percent-clipped=3.0 2023-04-16 16:02:31,479 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-16 16:02:33,455 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17423.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:02:43,517 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-16 16:02:52,788 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17444.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 16:02:53,243 INFO [train.py:893] (1/4) Epoch 7, batch 1250, loss[loss=0.2437, simple_loss=0.2883, pruned_loss=0.09956, over 13438.00 frames. ], tot_loss[loss=0.2473, simple_loss=0.2877, pruned_loss=0.1035, over 2661477.30 frames. ], batch size: 106, lr: 1.86e-02, grad_scale: 16.0 2023-04-16 16:02:54,331 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8222, 4.2033, 4.0462, 3.9354, 4.0465, 3.8735, 4.3492, 4.3742], device='cuda:1'), covar=tensor([0.0267, 0.0245, 0.0195, 0.0358, 0.0242, 0.0301, 0.0257, 0.0232], device='cuda:1'), in_proj_covar=tensor([0.0100, 0.0148, 0.0111, 0.0137, 0.0101, 0.0140, 0.0102, 0.0116], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:02:57,702 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0 2023-04-16 16:03:38,911 INFO [train.py:893] (1/4) Epoch 7, batch 1300, loss[loss=0.2688, simple_loss=0.2984, pruned_loss=0.1196, over 13343.00 frames. ], tot_loss[loss=0.2494, simple_loss=0.2894, pruned_loss=0.1047, over 2661064.94 frames. ], batch size: 73, lr: 1.86e-02, grad_scale: 16.0 2023-04-16 16:03:48,783 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.402e+02 3.839e+02 4.451e+02 5.527e+02 8.705e+02, threshold=8.903e+02, percent-clipped=2.0 2023-04-16 16:04:00,448 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.4965, 1.7246, 3.7645, 3.4916, 3.6805, 2.9574, 3.3656, 2.4625], device='cuda:1'), covar=tensor([0.2266, 0.1921, 0.0095, 0.0279, 0.0163, 0.0612, 0.0230, 0.1018], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0184, 0.0096, 0.0103, 0.0105, 0.0153, 0.0105, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-16 16:04:05,375 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17524.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 16:04:18,343 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.6758, 1.9412, 4.0153, 3.7046, 3.9084, 3.2874, 3.4645, 2.7241], device='cuda:1'), covar=tensor([0.2514, 0.1923, 0.0072, 0.0342, 0.0196, 0.0490, 0.0280, 0.1067], device='cuda:1'), in_proj_covar=tensor([0.0198, 0.0185, 0.0096, 0.0105, 0.0106, 0.0154, 0.0107, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-16 16:04:24,323 INFO [train.py:893] (1/4) Epoch 7, batch 1350, loss[loss=0.2417, simple_loss=0.2849, pruned_loss=0.09925, over 13451.00 frames. ], tot_loss[loss=0.2503, simple_loss=0.29, pruned_loss=0.1053, over 2659105.61 frames. ], batch size: 95, lr: 1.86e-02, grad_scale: 16.0 2023-04-16 16:04:24,513 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17545.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:04:40,156 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17562.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 16:04:49,101 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17572.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 16:05:09,770 INFO [train.py:893] (1/4) Epoch 7, batch 1400, loss[loss=0.2238, simple_loss=0.2739, pruned_loss=0.08686, over 13460.00 frames. ], tot_loss[loss=0.2489, simple_loss=0.2887, pruned_loss=0.1045, over 2658846.91 frames. ], batch size: 79, lr: 1.86e-02, grad_scale: 16.0 2023-04-16 16:05:18,474 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.2762, 4.7780, 4.6778, 4.6496, 4.4833, 4.5488, 5.2216, 4.8014], device='cuda:1'), covar=tensor([0.0820, 0.0885, 0.2247, 0.2940, 0.0862, 0.1510, 0.0875, 0.0999], device='cuda:1'), in_proj_covar=tensor([0.0237, 0.0264, 0.0342, 0.0350, 0.0191, 0.0266, 0.0314, 0.0233], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 16:05:20,572 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.063e+02 3.603e+02 4.503e+02 5.400e+02 1.169e+03, threshold=9.005e+02, percent-clipped=1.0 2023-04-16 16:05:24,076 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17610.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:05:25,075 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3243, 2.2928, 2.5191, 3.7282, 3.3477, 3.7292, 3.0930, 2.3434], device='cuda:1'), covar=tensor([0.0265, 0.1054, 0.0782, 0.0044, 0.0282, 0.0040, 0.0504, 0.0974], device='cuda:1'), in_proj_covar=tensor([0.0094, 0.0149, 0.0147, 0.0075, 0.0089, 0.0068, 0.0142, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:05:55,706 INFO [train.py:893] (1/4) Epoch 7, batch 1450, loss[loss=0.2541, simple_loss=0.3033, pruned_loss=0.1024, over 13527.00 frames. ], tot_loss[loss=0.2486, simple_loss=0.2883, pruned_loss=0.1045, over 2660947.14 frames. ], batch size: 85, lr: 1.85e-02, grad_scale: 16.0 2023-04-16 16:06:15,078 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-04-16 16:06:40,975 INFO [train.py:893] (1/4) Epoch 7, batch 1500, loss[loss=0.261, simple_loss=0.3066, pruned_loss=0.1077, over 13519.00 frames. ], tot_loss[loss=0.2483, simple_loss=0.2882, pruned_loss=0.1041, over 2655673.32 frames. ], batch size: 98, lr: 1.85e-02, grad_scale: 16.0 2023-04-16 16:06:45,822 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17700.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:06:50,423 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.535e+02 3.584e+02 4.081e+02 5.020e+02 1.019e+03, threshold=8.163e+02, percent-clipped=1.0 2023-04-16 16:06:55,668 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5604, 4.0103, 3.6900, 3.6131, 3.7815, 4.0708, 3.8885, 3.6433], device='cuda:1'), covar=tensor([0.0299, 0.0218, 0.0291, 0.0991, 0.0237, 0.0221, 0.0256, 0.0546], device='cuda:1'), in_proj_covar=tensor([0.0103, 0.0109, 0.0113, 0.0202, 0.0115, 0.0128, 0.0111, 0.0086], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0004, 0.0003, 0.0002], device='cuda:1') 2023-04-16 16:07:01,912 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17718.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:07:13,085 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17730.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:07:20,411 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=17739.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 16:07:22,159 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.5793, 2.3806, 1.9040, 1.1587, 1.1885, 1.8625, 1.6772, 2.5255], device='cuda:1'), covar=tensor([0.0844, 0.0272, 0.0909, 0.1673, 0.0300, 0.0338, 0.0795, 0.0260], device='cuda:1'), in_proj_covar=tensor([0.0104, 0.0087, 0.0078, 0.0134, 0.0074, 0.0089, 0.0101, 0.0081], device='cuda:1'), out_proj_covar=tensor([8.2301e-05, 6.8481e-05, 6.6090e-05, 1.1265e-04, 6.5455e-05, 6.8837e-05, 8.2332e-05, 5.9996e-05], device='cuda:1') 2023-04-16 16:07:25,685 INFO [train.py:893] (1/4) Epoch 7, batch 1550, loss[loss=0.2348, simple_loss=0.2865, pruned_loss=0.09154, over 13486.00 frames. ], tot_loss[loss=0.2491, simple_loss=0.289, pruned_loss=0.1046, over 2659234.89 frames. ], batch size: 81, lr: 1.85e-02, grad_scale: 16.0 2023-04-16 16:07:30,753 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.6240, 2.3505, 1.9089, 1.1334, 1.1057, 1.7940, 1.6846, 2.5311], device='cuda:1'), covar=tensor([0.0776, 0.0262, 0.0711, 0.1619, 0.0237, 0.0418, 0.0695, 0.0211], device='cuda:1'), in_proj_covar=tensor([0.0103, 0.0086, 0.0078, 0.0134, 0.0073, 0.0088, 0.0101, 0.0081], device='cuda:1'), out_proj_covar=tensor([8.1680e-05, 6.8105e-05, 6.5805e-05, 1.1209e-04, 6.5256e-05, 6.8370e-05, 8.1949e-05, 6.0022e-05], device='cuda:1') 2023-04-16 16:07:40,508 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17761.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:08:07,175 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.5705, 2.1207, 1.6113, 1.1751, 1.2470, 1.9332, 1.6137, 2.2896], device='cuda:1'), covar=tensor([0.0628, 0.0261, 0.0943, 0.1505, 0.0152, 0.0282, 0.0669, 0.0332], device='cuda:1'), in_proj_covar=tensor([0.0102, 0.0086, 0.0078, 0.0134, 0.0073, 0.0087, 0.0100, 0.0081], device='cuda:1'), out_proj_covar=tensor([8.1170e-05, 6.7928e-05, 6.5945e-05, 1.1238e-04, 6.4940e-05, 6.7285e-05, 8.1737e-05, 6.0371e-05], device='cuda:1') 2023-04-16 16:08:07,938 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=17791.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:08:11,689 INFO [train.py:893] (1/4) Epoch 7, batch 1600, loss[loss=0.2351, simple_loss=0.2802, pruned_loss=0.09498, over 13368.00 frames. ], tot_loss[loss=0.2493, simple_loss=0.2897, pruned_loss=0.1044, over 2657283.11 frames. ], batch size: 67, lr: 1.85e-02, grad_scale: 16.0 2023-04-16 16:08:12,091 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8497, 4.0559, 2.5101, 4.2869, 3.9393, 2.2677, 3.1745, 2.5299], device='cuda:1'), covar=tensor([0.0311, 0.0365, 0.1449, 0.0129, 0.0279, 0.1389, 0.0775, 0.1609], device='cuda:1'), in_proj_covar=tensor([0.0125, 0.0129, 0.0173, 0.0096, 0.0112, 0.0159, 0.0147, 0.0163], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 16:08:21,303 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.592e+02 3.759e+02 4.311e+02 5.092e+02 1.331e+03, threshold=8.621e+02, percent-clipped=7.0 2023-04-16 16:08:35,370 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.18 vs. limit=2.0 2023-04-16 16:08:41,123 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2023-04-16 16:08:57,467 INFO [train.py:893] (1/4) Epoch 7, batch 1650, loss[loss=0.2497, simple_loss=0.2881, pruned_loss=0.1056, over 12013.00 frames. ], tot_loss[loss=0.2485, simple_loss=0.2893, pruned_loss=0.1039, over 2645496.14 frames. ], batch size: 157, lr: 1.84e-02, grad_scale: 16.0 2023-04-16 16:08:57,709 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=17845.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:09:26,500 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.04 vs. limit=5.0 2023-04-16 16:09:41,152 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=17893.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:09:42,612 INFO [train.py:893] (1/4) Epoch 7, batch 1700, loss[loss=0.222, simple_loss=0.2675, pruned_loss=0.08823, over 13339.00 frames. ], tot_loss[loss=0.2475, simple_loss=0.289, pruned_loss=0.103, over 2646228.00 frames. ], batch size: 67, lr: 1.84e-02, grad_scale: 16.0 2023-04-16 16:09:52,380 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.676e+02 3.617e+02 4.238e+02 5.324e+02 1.365e+03, threshold=8.475e+02, percent-clipped=1.0 2023-04-16 16:10:12,463 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2023-04-16 16:10:28,902 INFO [train.py:893] (1/4) Epoch 7, batch 1750, loss[loss=0.2527, simple_loss=0.2969, pruned_loss=0.1043, over 13562.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.2871, pruned_loss=0.1015, over 2650023.88 frames. ], batch size: 89, lr: 1.84e-02, grad_scale: 16.0 2023-04-16 16:10:42,299 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=17960.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:11:02,543 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-04-16 16:11:03,983 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2251, 3.9835, 4.2491, 3.1381, 4.7209, 4.3652, 4.3660, 4.6484], device='cuda:1'), covar=tensor([0.0179, 0.0104, 0.0105, 0.0671, 0.0086, 0.0138, 0.0115, 0.0073], device='cuda:1'), in_proj_covar=tensor([0.0058, 0.0033, 0.0053, 0.0078, 0.0064, 0.0058, 0.0054, 0.0045], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:11:09,852 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.45 vs. limit=5.0 2023-04-16 16:11:13,876 INFO [train.py:893] (1/4) Epoch 7, batch 1800, loss[loss=0.2609, simple_loss=0.3006, pruned_loss=0.1106, over 13459.00 frames. ], tot_loss[loss=0.2431, simple_loss=0.2856, pruned_loss=0.1003, over 2655152.45 frames. ], batch size: 100, lr: 1.84e-02, grad_scale: 16.0 2023-04-16 16:11:28,369 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.327e+02 3.248e+02 3.972e+02 4.740e+02 1.184e+03, threshold=7.944e+02, percent-clipped=3.0 2023-04-16 16:11:39,229 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18018.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:11:41,659 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18021.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:11:54,848 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.80 vs. limit=2.0 2023-04-16 16:11:58,663 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18039.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 16:12:03,912 INFO [train.py:893] (1/4) Epoch 7, batch 1850, loss[loss=0.2287, simple_loss=0.273, pruned_loss=0.09225, over 13529.00 frames. ], tot_loss[loss=0.2411, simple_loss=0.2839, pruned_loss=0.09917, over 2658433.15 frames. ], batch size: 85, lr: 1.83e-02, grad_scale: 16.0 2023-04-16 16:12:06,390 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-16 16:12:13,055 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18056.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:12:14,639 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8029, 3.4857, 3.0639, 3.4543, 2.9557, 1.7120, 3.4294, 1.9539], device='cuda:1'), covar=tensor([0.0730, 0.0644, 0.0365, 0.0312, 0.0724, 0.2027, 0.1441, 0.1636], device='cuda:1'), in_proj_covar=tensor([0.0111, 0.0094, 0.0098, 0.0087, 0.0120, 0.0147, 0.0099, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:12:22,677 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18066.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:12:40,531 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18086.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:12:41,377 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18087.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 16:12:48,929 INFO [train.py:893] (1/4) Epoch 7, batch 1900, loss[loss=0.2345, simple_loss=0.2771, pruned_loss=0.09593, over 13533.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.2832, pruned_loss=0.09928, over 2658635.86 frames. ], batch size: 76, lr: 1.83e-02, grad_scale: 16.0 2023-04-16 16:12:58,585 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.339e+02 3.460e+02 4.332e+02 5.125e+02 1.072e+03, threshold=8.663e+02, percent-clipped=3.0 2023-04-16 16:13:17,077 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.68 vs. limit=2.0 2023-04-16 16:13:24,106 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.6269, 1.9493, 3.5556, 3.3473, 3.5086, 2.9631, 3.3079, 2.5528], device='cuda:1'), covar=tensor([0.2188, 0.1689, 0.0079, 0.0260, 0.0213, 0.0581, 0.0168, 0.0956], device='cuda:1'), in_proj_covar=tensor([0.0199, 0.0190, 0.0096, 0.0107, 0.0110, 0.0155, 0.0107, 0.0164], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-16 16:13:34,749 INFO [train.py:893] (1/4) Epoch 7, batch 1950, loss[loss=0.2412, simple_loss=0.2617, pruned_loss=0.1104, over 11373.00 frames. ], tot_loss[loss=0.24, simple_loss=0.2823, pruned_loss=0.09879, over 2660704.29 frames. ], batch size: 46, lr: 1.83e-02, grad_scale: 32.0 2023-04-16 16:14:01,845 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18174.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:14:02,664 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18175.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:14:20,800 INFO [train.py:893] (1/4) Epoch 7, batch 2000, loss[loss=0.2293, simple_loss=0.2787, pruned_loss=0.08996, over 13527.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.2847, pruned_loss=0.1001, over 2662889.06 frames. ], batch size: 85, lr: 1.83e-02, grad_scale: 32.0 2023-04-16 16:14:26,662 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-16 16:14:30,676 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.042e+02 3.662e+02 4.315e+02 5.072e+02 8.289e+02, threshold=8.631e+02, percent-clipped=0.0 2023-04-16 16:14:57,863 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18235.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:14:58,693 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18236.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:15:06,526 INFO [train.py:893] (1/4) Epoch 7, batch 2050, loss[loss=0.2281, simple_loss=0.2829, pruned_loss=0.08664, over 13493.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.2867, pruned_loss=0.1012, over 2661778.52 frames. ], batch size: 81, lr: 1.82e-02, grad_scale: 32.0 2023-04-16 16:15:45,421 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.99 vs. limit=5.0 2023-04-16 16:15:45,538 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.75 vs. limit=2.0 2023-04-16 16:15:52,003 INFO [train.py:893] (1/4) Epoch 7, batch 2100, loss[loss=0.2264, simple_loss=0.276, pruned_loss=0.08841, over 13545.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.2877, pruned_loss=0.1019, over 2659027.67 frames. ], batch size: 76, lr: 1.82e-02, grad_scale: 32.0 2023-04-16 16:16:01,878 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.613e+02 3.375e+02 4.098e+02 5.034e+02 9.532e+02, threshold=8.196e+02, percent-clipped=1.0 2023-04-16 16:16:11,114 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18316.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:16:16,541 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-04-16 16:16:28,495 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.7108, 3.9954, 3.7950, 4.3819, 2.2098, 3.2097, 4.0923, 2.2396], device='cuda:1'), covar=tensor([0.0079, 0.0493, 0.0584, 0.0307, 0.1596, 0.0846, 0.0590, 0.2130], device='cuda:1'), in_proj_covar=tensor([0.0085, 0.0152, 0.0168, 0.0150, 0.0172, 0.0176, 0.0156, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 16:16:36,239 INFO [train.py:893] (1/4) Epoch 7, batch 2150, loss[loss=0.2705, simple_loss=0.3088, pruned_loss=0.116, over 13242.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.2874, pruned_loss=0.1012, over 2661616.29 frames. ], batch size: 117, lr: 1.82e-02, grad_scale: 32.0 2023-04-16 16:16:47,421 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18356.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:17:13,884 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18386.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:17:16,224 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0838, 2.6734, 2.0501, 3.9220, 4.4817, 3.3198, 4.3910, 4.0823], device='cuda:1'), covar=tensor([0.0087, 0.0696, 0.0957, 0.0086, 0.0052, 0.0378, 0.0062, 0.0073], device='cuda:1'), in_proj_covar=tensor([0.0052, 0.0071, 0.0080, 0.0058, 0.0042, 0.0064, 0.0040, 0.0050], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:17:21,659 INFO [train.py:893] (1/4) Epoch 7, batch 2200, loss[loss=0.2584, simple_loss=0.2989, pruned_loss=0.1089, over 13598.00 frames. ], tot_loss[loss=0.2439, simple_loss=0.2865, pruned_loss=0.1006, over 2654709.11 frames. ], batch size: 89, lr: 1.82e-02, grad_scale: 32.0 2023-04-16 16:17:30,261 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18404.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:17:31,680 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.091e+02 3.363e+02 3.835e+02 4.605e+02 1.288e+03, threshold=7.670e+02, percent-clipped=1.0 2023-04-16 16:17:49,272 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.95 vs. limit=2.0 2023-04-16 16:17:57,905 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18434.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:18:08,162 INFO [train.py:893] (1/4) Epoch 7, batch 2250, loss[loss=0.2652, simple_loss=0.3011, pruned_loss=0.1147, over 13530.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.2835, pruned_loss=0.09915, over 2655264.82 frames. ], batch size: 98, lr: 1.82e-02, grad_scale: 32.0 2023-04-16 16:18:44,496 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-04-16 16:18:49,097 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18490.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:18:53,482 INFO [train.py:893] (1/4) Epoch 7, batch 2300, loss[loss=0.2203, simple_loss=0.2649, pruned_loss=0.08783, over 13392.00 frames. ], tot_loss[loss=0.2399, simple_loss=0.2825, pruned_loss=0.09868, over 2657776.66 frames. ], batch size: 65, lr: 1.81e-02, grad_scale: 32.0 2023-04-16 16:19:03,357 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.168e+02 3.469e+02 4.011e+02 5.110e+02 9.671e+02, threshold=8.022e+02, percent-clipped=1.0 2023-04-16 16:19:05,303 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18508.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 16:19:14,129 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4963, 2.1885, 2.8472, 4.0097, 3.6031, 3.9952, 3.2058, 2.1997], device='cuda:1'), covar=tensor([0.0383, 0.1295, 0.0866, 0.0062, 0.0223, 0.0053, 0.0601, 0.1159], device='cuda:1'), in_proj_covar=tensor([0.0098, 0.0151, 0.0156, 0.0081, 0.0094, 0.0075, 0.0149, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:19:15,802 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.5960, 2.2252, 1.8908, 2.6138, 2.0108, 2.4170, 2.3394, 2.3992], device='cuda:1'), covar=tensor([0.0087, 0.0122, 0.0140, 0.0108, 0.0152, 0.0095, 0.0240, 0.0128], device='cuda:1'), in_proj_covar=tensor([0.0054, 0.0057, 0.0064, 0.0054, 0.0071, 0.0053, 0.0061, 0.0057], device='cuda:1'), out_proj_covar=tensor([6.1932e-05, 6.8483e-05, 7.8569e-05, 6.5564e-05, 8.5702e-05, 6.3556e-05, 7.6365e-05, 6.8484e-05], device='cuda:1') 2023-04-16 16:19:25,252 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18530.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:19:26,084 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18531.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:19:27,094 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9288, 4.1464, 2.5943, 4.2886, 3.9178, 2.0655, 3.3303, 2.4449], device='cuda:1'), covar=tensor([0.0262, 0.0303, 0.1391, 0.0131, 0.0272, 0.1567, 0.0835, 0.1783], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0128, 0.0171, 0.0100, 0.0110, 0.0156, 0.0145, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 16:19:38,752 INFO [train.py:893] (1/4) Epoch 7, batch 2350, loss[loss=0.2436, simple_loss=0.2871, pruned_loss=0.1, over 13359.00 frames. ], tot_loss[loss=0.2395, simple_loss=0.282, pruned_loss=0.09848, over 2659853.13 frames. ], batch size: 118, lr: 1.81e-02, grad_scale: 32.0 2023-04-16 16:19:41,198 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0860, 4.1897, 3.6066, 3.0294, 3.1390, 2.5379, 4.3889, 2.7017], device='cuda:1'), covar=tensor([0.1292, 0.0263, 0.0575, 0.1181, 0.0595, 0.2320, 0.0139, 0.2804], device='cuda:1'), in_proj_covar=tensor([0.0268, 0.0230, 0.0236, 0.0256, 0.0202, 0.0256, 0.0164, 0.0307], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 16:19:44,438 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18551.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:19:55,766 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5758, 3.1254, 3.5320, 2.4620, 3.6242, 3.4614, 3.4326, 3.7421], device='cuda:1'), covar=tensor([0.0188, 0.0166, 0.0157, 0.0888, 0.0141, 0.0168, 0.0145, 0.0114], device='cuda:1'), in_proj_covar=tensor([0.0061, 0.0035, 0.0056, 0.0081, 0.0067, 0.0061, 0.0056, 0.0046], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:19:59,088 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.8241, 2.5561, 2.0324, 1.4393, 1.3056, 2.1163, 1.9110, 2.6711], device='cuda:1'), covar=tensor([0.0706, 0.0266, 0.0913, 0.1643, 0.0278, 0.0538, 0.0691, 0.0343], device='cuda:1'), in_proj_covar=tensor([0.0102, 0.0087, 0.0080, 0.0137, 0.0074, 0.0091, 0.0105, 0.0087], device='cuda:1'), out_proj_covar=tensor([8.0538e-05, 6.7627e-05, 6.7046e-05, 1.1441e-04, 6.5431e-05, 7.0003e-05, 8.4035e-05, 6.5109e-05], device='cuda:1') 2023-04-16 16:20:00,556 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18569.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 16:20:01,163 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-16 16:20:23,733 INFO [train.py:893] (1/4) Epoch 7, batch 2400, loss[loss=0.2573, simple_loss=0.2963, pruned_loss=0.1092, over 13211.00 frames. ], tot_loss[loss=0.2409, simple_loss=0.283, pruned_loss=0.09936, over 2660517.10 frames. ], batch size: 132, lr: 1.81e-02, grad_scale: 32.0 2023-04-16 16:20:34,557 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.416e+02 3.464e+02 4.191e+02 4.925e+02 8.297e+02, threshold=8.381e+02, percent-clipped=2.0 2023-04-16 16:20:34,877 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4912, 2.4190, 2.7715, 4.2140, 3.8691, 4.1953, 3.2817, 2.2180], device='cuda:1'), covar=tensor([0.0372, 0.1295, 0.0871, 0.0064, 0.0174, 0.0061, 0.0626, 0.1114], device='cuda:1'), in_proj_covar=tensor([0.0097, 0.0148, 0.0152, 0.0080, 0.0093, 0.0074, 0.0146, 0.0141], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:20:43,601 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18616.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:21:09,658 INFO [train.py:893] (1/4) Epoch 7, batch 2450, loss[loss=0.2622, simple_loss=0.3033, pruned_loss=0.1105, over 13392.00 frames. ], tot_loss[loss=0.2405, simple_loss=0.2826, pruned_loss=0.09915, over 2662659.47 frames. ], batch size: 118, lr: 1.81e-02, grad_scale: 32.0 2023-04-16 16:21:26,026 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18664.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:21:36,315 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18674.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:21:53,929 INFO [train.py:893] (1/4) Epoch 7, batch 2500, loss[loss=0.233, simple_loss=0.2795, pruned_loss=0.09324, over 13521.00 frames. ], tot_loss[loss=0.242, simple_loss=0.2838, pruned_loss=0.1001, over 2660349.20 frames. ], batch size: 91, lr: 1.80e-02, grad_scale: 32.0 2023-04-16 16:22:04,436 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.081e+02 3.550e+02 4.127e+02 4.846e+02 9.992e+02, threshold=8.253e+02, percent-clipped=2.0 2023-04-16 16:22:30,360 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18735.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:22:39,103 INFO [train.py:893] (1/4) Epoch 7, batch 2550, loss[loss=0.2613, simple_loss=0.2986, pruned_loss=0.112, over 13247.00 frames. ], tot_loss[loss=0.2421, simple_loss=0.284, pruned_loss=0.1001, over 2666358.88 frames. ], batch size: 124, lr: 1.80e-02, grad_scale: 32.0 2023-04-16 16:23:04,395 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-16 16:23:25,406 INFO [train.py:893] (1/4) Epoch 7, batch 2600, loss[loss=0.2282, simple_loss=0.2741, pruned_loss=0.09119, over 13497.00 frames. ], tot_loss[loss=0.2419, simple_loss=0.2836, pruned_loss=0.1001, over 2667242.84 frames. ], batch size: 93, lr: 1.80e-02, grad_scale: 32.0 2023-04-16 16:23:35,304 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.202e+02 3.564e+02 4.195e+02 5.330e+02 9.508e+02, threshold=8.389e+02, percent-clipped=2.0 2023-04-16 16:23:54,831 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18830.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:23:55,591 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=18831.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:23:59,250 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2087, 4.7015, 4.4147, 4.3239, 4.2773, 4.2032, 4.7268, 4.8120], device='cuda:1'), covar=tensor([0.0220, 0.0185, 0.0213, 0.0323, 0.0283, 0.0226, 0.0288, 0.0165], device='cuda:1'), in_proj_covar=tensor([0.0103, 0.0142, 0.0108, 0.0137, 0.0099, 0.0139, 0.0099, 0.0114], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:24:00,894 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7213, 4.0126, 4.3664, 3.3440, 2.7568, 3.1157, 4.5045, 4.7044], device='cuda:1'), covar=tensor([0.0802, 0.0704, 0.0248, 0.1159, 0.1307, 0.0934, 0.0145, 0.0074], device='cuda:1'), in_proj_covar=tensor([0.0200, 0.0185, 0.0153, 0.0195, 0.0191, 0.0159, 0.0133, 0.0103], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0001], device='cuda:1') 2023-04-16 16:24:05,695 INFO [train.py:893] (1/4) Epoch 7, batch 2650, loss[loss=0.2398, simple_loss=0.279, pruned_loss=0.1004, over 13188.00 frames. ], tot_loss[loss=0.243, simple_loss=0.2844, pruned_loss=0.1008, over 2666261.89 frames. ], batch size: 58, lr: 1.80e-02, grad_scale: 32.0 2023-04-16 16:24:06,630 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18846.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:24:13,437 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1402, 4.5615, 4.0964, 4.2271, 4.3386, 4.7714, 4.4577, 4.4017], device='cuda:1'), covar=tensor([0.0324, 0.0273, 0.0390, 0.1108, 0.0261, 0.0240, 0.0287, 0.0274], device='cuda:1'), in_proj_covar=tensor([0.0102, 0.0108, 0.0113, 0.0205, 0.0114, 0.0132, 0.0112, 0.0089], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-16 16:24:21,350 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=18864.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 16:24:31,893 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18878.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:24:32,552 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=18879.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:25:03,579 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 16:25:13,565 INFO [train.py:893] (1/4) Epoch 8, batch 0, loss[loss=0.2449, simple_loss=0.2767, pruned_loss=0.1066, over 13231.00 frames. ], tot_loss[loss=0.2449, simple_loss=0.2767, pruned_loss=0.1066, over 13231.00 frames. ], batch size: 117, lr: 1.69e-02, grad_scale: 16.0 2023-04-16 16:25:13,565 INFO [train.py:918] (1/4) Computing validation loss 2023-04-16 16:25:35,920 INFO [train.py:927] (1/4) Epoch 8, validation: loss=0.1764, simple_loss=0.2286, pruned_loss=0.06213, over 2446609.00 frames. 2023-04-16 16:25:35,920 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12761MB 2023-04-16 16:25:47,139 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.595e+02 3.840e+02 4.521e+02 5.320e+02 9.570e+02, threshold=9.042e+02, percent-clipped=2.0 2023-04-16 16:26:07,728 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=18930.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:26:21,961 INFO [train.py:893] (1/4) Epoch 8, batch 50, loss[loss=0.2166, simple_loss=0.2599, pruned_loss=0.08662, over 13381.00 frames. ], tot_loss[loss=0.2375, simple_loss=0.2771, pruned_loss=0.099, over 599366.62 frames. ], batch size: 73, lr: 1.69e-02, grad_scale: 16.0 2023-04-16 16:26:46,540 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-16 16:26:46,541 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-16 16:26:46,541 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-16 16:26:46,555 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-16 16:26:46,563 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-16 16:26:46,587 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-16 16:26:48,047 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-16 16:26:55,468 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6382, 2.2538, 2.8739, 4.0517, 3.9253, 4.1396, 3.2479, 2.2758], device='cuda:1'), covar=tensor([0.0256, 0.1221, 0.0722, 0.0054, 0.0177, 0.0057, 0.0636, 0.1137], device='cuda:1'), in_proj_covar=tensor([0.0091, 0.0142, 0.0143, 0.0075, 0.0092, 0.0070, 0.0140, 0.0134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:27:04,751 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=18991.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:27:07,780 INFO [train.py:893] (1/4) Epoch 8, batch 100, loss[loss=0.2632, simple_loss=0.2951, pruned_loss=0.1157, over 11950.00 frames. ], tot_loss[loss=0.2383, simple_loss=0.2773, pruned_loss=0.09958, over 1047044.80 frames. ], batch size: 157, lr: 1.69e-02, grad_scale: 16.0 2023-04-16 16:27:19,877 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.147e+02 3.465e+02 4.037e+02 4.694e+02 8.484e+02, threshold=8.075e+02, percent-clipped=0.0 2023-04-16 16:27:38,810 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6797, 4.2469, 4.2789, 4.1932, 3.7948, 4.1246, 4.6522, 4.1404], device='cuda:1'), covar=tensor([0.0787, 0.1090, 0.2115, 0.2722, 0.1064, 0.1451, 0.0923, 0.1068], device='cuda:1'), in_proj_covar=tensor([0.0238, 0.0274, 0.0350, 0.0354, 0.0193, 0.0266, 0.0319, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 16:27:40,558 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19030.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:27:54,669 INFO [train.py:893] (1/4) Epoch 8, batch 150, loss[loss=0.273, simple_loss=0.3068, pruned_loss=0.1196, over 13391.00 frames. ], tot_loss[loss=0.244, simple_loss=0.2828, pruned_loss=0.1026, over 1398027.18 frames. ], batch size: 109, lr: 1.68e-02, grad_scale: 16.0 2023-04-16 16:28:41,429 INFO [train.py:893] (1/4) Epoch 8, batch 200, loss[loss=0.2746, simple_loss=0.3118, pruned_loss=0.1187, over 13445.00 frames. ], tot_loss[loss=0.2456, simple_loss=0.2849, pruned_loss=0.1032, over 1675320.77 frames. ], batch size: 103, lr: 1.68e-02, grad_scale: 16.0 2023-04-16 16:28:42,923 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.81 vs. limit=2.0 2023-04-16 16:28:53,339 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.148e+02 3.539e+02 3.896e+02 4.763e+02 1.039e+03, threshold=7.792e+02, percent-clipped=3.0 2023-04-16 16:29:07,798 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.4557, 2.4207, 1.9870, 1.3494, 1.3002, 2.1503, 1.7370, 2.5981], device='cuda:1'), covar=tensor([0.0949, 0.0402, 0.0989, 0.1782, 0.0485, 0.0397, 0.0835, 0.0329], device='cuda:1'), in_proj_covar=tensor([0.0110, 0.0095, 0.0086, 0.0150, 0.0080, 0.0097, 0.0108, 0.0090], device='cuda:1'), out_proj_covar=tensor([8.5548e-05, 7.3123e-05, 7.1955e-05, 1.2338e-04, 7.1180e-05, 7.4325e-05, 8.6402e-05, 6.7004e-05], device='cuda:1') 2023-04-16 16:29:12,893 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.98 vs. limit=2.0 2023-04-16 16:29:14,170 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19129.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 16:29:28,579 INFO [train.py:893] (1/4) Epoch 8, batch 250, loss[loss=0.2259, simple_loss=0.2694, pruned_loss=0.09125, over 13510.00 frames. ], tot_loss[loss=0.2455, simple_loss=0.2852, pruned_loss=0.1028, over 1894822.76 frames. ], batch size: 76, lr: 1.68e-02, grad_scale: 16.0 2023-04-16 16:29:29,777 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19146.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:29:33,253 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7718, 3.5420, 3.8595, 2.5412, 4.1772, 3.8179, 3.7991, 4.2659], device='cuda:1'), covar=tensor([0.0204, 0.0132, 0.0115, 0.0873, 0.0147, 0.0194, 0.0152, 0.0067], device='cuda:1'), in_proj_covar=tensor([0.0059, 0.0033, 0.0054, 0.0077, 0.0066, 0.0060, 0.0054, 0.0045], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0001, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:29:47,340 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19164.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 16:29:59,090 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3211, 2.0398, 2.5686, 3.8628, 3.5719, 3.8741, 2.9613, 2.1330], device='cuda:1'), covar=tensor([0.0275, 0.1211, 0.0809, 0.0044, 0.0200, 0.0047, 0.0612, 0.1113], device='cuda:1'), in_proj_covar=tensor([0.0094, 0.0143, 0.0145, 0.0077, 0.0094, 0.0070, 0.0142, 0.0134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:30:10,830 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19190.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 16:30:12,048 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-04-16 16:30:14,660 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19194.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:30:15,327 INFO [train.py:893] (1/4) Epoch 8, batch 300, loss[loss=0.249, simple_loss=0.2958, pruned_loss=0.1011, over 13488.00 frames. ], tot_loss[loss=0.2445, simple_loss=0.2853, pruned_loss=0.1019, over 2065629.30 frames. ], batch size: 93, lr: 1.68e-02, grad_scale: 16.0 2023-04-16 16:30:27,118 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.082e+02 3.374e+02 4.034e+02 4.942e+02 9.430e+02, threshold=8.068e+02, percent-clipped=2.0 2023-04-16 16:30:31,512 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19212.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 16:31:01,914 INFO [train.py:893] (1/4) Epoch 8, batch 350, loss[loss=0.2828, simple_loss=0.317, pruned_loss=0.1243, over 13267.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.2863, pruned_loss=0.1025, over 2199633.38 frames. ], batch size: 124, lr: 1.68e-02, grad_scale: 16.0 2023-04-16 16:31:30,410 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.88 vs. limit=5.0 2023-04-16 16:31:40,000 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19286.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:31:48,575 INFO [train.py:893] (1/4) Epoch 8, batch 400, loss[loss=0.2239, simple_loss=0.2799, pruned_loss=0.08391, over 13457.00 frames. ], tot_loss[loss=0.2457, simple_loss=0.2868, pruned_loss=0.1023, over 2301121.91 frames. ], batch size: 103, lr: 1.67e-02, grad_scale: 16.0 2023-04-16 16:31:53,620 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1299, 4.4644, 4.1779, 4.1875, 4.2506, 4.7759, 4.5491, 4.3337], device='cuda:1'), covar=tensor([0.0376, 0.0280, 0.0277, 0.1064, 0.0265, 0.0249, 0.0204, 0.0224], device='cuda:1'), in_proj_covar=tensor([0.0102, 0.0107, 0.0111, 0.0200, 0.0113, 0.0129, 0.0109, 0.0086], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0004, 0.0003, 0.0002], device='cuda:1') 2023-04-16 16:31:59,044 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.962e+02 3.520e+02 4.019e+02 4.754e+02 8.353e+02, threshold=8.037e+02, percent-clipped=1.0 2023-04-16 16:32:20,943 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19330.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:32:21,201 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-04-16 16:32:25,264 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2023-04-16 16:32:33,852 INFO [train.py:893] (1/4) Epoch 8, batch 450, loss[loss=0.2601, simple_loss=0.2989, pruned_loss=0.1106, over 13575.00 frames. ], tot_loss[loss=0.2469, simple_loss=0.2878, pruned_loss=0.103, over 2382430.62 frames. ], batch size: 89, lr: 1.67e-02, grad_scale: 16.0 2023-04-16 16:32:59,490 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-16 16:33:05,421 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19378.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:33:17,108 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.4858, 3.1839, 2.6499, 2.9450, 2.7394, 1.5460, 3.3314, 1.7560], device='cuda:1'), covar=tensor([0.0675, 0.0624, 0.0472, 0.0403, 0.0823, 0.2004, 0.0526, 0.1434], device='cuda:1'), in_proj_covar=tensor([0.0111, 0.0096, 0.0103, 0.0086, 0.0123, 0.0149, 0.0101, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:33:18,060 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-04-16 16:33:21,167 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19394.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:33:21,692 INFO [train.py:893] (1/4) Epoch 8, batch 500, loss[loss=0.1916, simple_loss=0.2406, pruned_loss=0.07135, over 13338.00 frames. ], tot_loss[loss=0.2466, simple_loss=0.2878, pruned_loss=0.1027, over 2441719.15 frames. ], batch size: 62, lr: 1.67e-02, grad_scale: 16.0 2023-04-16 16:33:27,476 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.35 vs. limit=5.0 2023-04-16 16:33:31,900 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.189e+02 3.388e+02 3.878e+02 4.820e+02 9.115e+02, threshold=7.756e+02, percent-clipped=3.0 2023-04-16 16:34:07,897 INFO [train.py:893] (1/4) Epoch 8, batch 550, loss[loss=0.2328, simple_loss=0.2819, pruned_loss=0.09184, over 13436.00 frames. ], tot_loss[loss=0.2451, simple_loss=0.2874, pruned_loss=0.1014, over 2494033.50 frames. ], batch size: 103, lr: 1.67e-02, grad_scale: 16.0 2023-04-16 16:34:18,158 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19455.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:34:30,828 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.0843, 4.9455, 5.2049, 4.9426, 5.4070, 4.8465, 5.3887, 5.3816], device='cuda:1'), covar=tensor([0.0253, 0.0394, 0.0453, 0.0409, 0.0471, 0.0721, 0.0421, 0.0393], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0213, 0.0199, 0.0153, 0.0287, 0.0242, 0.0174, 0.0189], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 16:34:31,777 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5488, 3.9044, 3.6924, 4.2867, 2.0973, 3.0642, 3.8997, 2.1124], device='cuda:1'), covar=tensor([0.0079, 0.0457, 0.0633, 0.0391, 0.1624, 0.0905, 0.0523, 0.2140], device='cuda:1'), in_proj_covar=tensor([0.0088, 0.0162, 0.0173, 0.0158, 0.0177, 0.0179, 0.0161, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 16:34:46,505 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19485.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 16:34:55,336 INFO [train.py:893] (1/4) Epoch 8, batch 600, loss[loss=0.2601, simple_loss=0.3054, pruned_loss=0.1074, over 13240.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.2853, pruned_loss=0.1002, over 2525424.53 frames. ], batch size: 117, lr: 1.67e-02, grad_scale: 16.0 2023-04-16 16:35:01,442 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.1464, 2.0245, 1.8032, 2.3316, 1.5440, 2.2516, 2.2338, 1.9530], device='cuda:1'), covar=tensor([0.0098, 0.0123, 0.0127, 0.0086, 0.0165, 0.0120, 0.0153, 0.0169], device='cuda:1'), in_proj_covar=tensor([0.0056, 0.0058, 0.0065, 0.0053, 0.0070, 0.0053, 0.0060, 0.0059], device='cuda:1'), out_proj_covar=tensor([6.5292e-05, 6.8048e-05, 7.9523e-05, 6.4168e-05, 8.4854e-05, 6.1885e-05, 7.3063e-05, 7.0161e-05], device='cuda:1') 2023-04-16 16:35:05,980 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.294e+02 3.228e+02 3.690e+02 4.422e+02 7.814e+02, threshold=7.380e+02, percent-clipped=1.0 2023-04-16 16:35:41,873 INFO [train.py:893] (1/4) Epoch 8, batch 650, loss[loss=0.2566, simple_loss=0.2981, pruned_loss=0.1075, over 13561.00 frames. ], tot_loss[loss=0.2425, simple_loss=0.2846, pruned_loss=0.1002, over 2554368.80 frames. ], batch size: 89, lr: 1.66e-02, grad_scale: 16.0 2023-04-16 16:36:20,072 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19586.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:36:27,515 INFO [train.py:893] (1/4) Epoch 8, batch 700, loss[loss=0.2256, simple_loss=0.2701, pruned_loss=0.09051, over 13423.00 frames. ], tot_loss[loss=0.2406, simple_loss=0.2832, pruned_loss=0.09897, over 2579582.79 frames. ], batch size: 106, lr: 1.66e-02, grad_scale: 16.0 2023-04-16 16:36:40,424 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.363e+02 3.372e+02 3.978e+02 4.795e+02 1.054e+03, threshold=7.956e+02, percent-clipped=5.0 2023-04-16 16:36:50,677 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=19618.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:37:05,793 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19634.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:37:09,092 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2382, 4.6431, 4.2620, 4.3309, 4.3821, 4.8324, 4.5393, 4.3907], device='cuda:1'), covar=tensor([0.0342, 0.0298, 0.0350, 0.1157, 0.0302, 0.0268, 0.0287, 0.0262], device='cuda:1'), in_proj_covar=tensor([0.0106, 0.0112, 0.0119, 0.0210, 0.0118, 0.0136, 0.0118, 0.0090], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-16 16:37:16,242 INFO [train.py:893] (1/4) Epoch 8, batch 750, loss[loss=0.2583, simple_loss=0.2962, pruned_loss=0.1102, over 12052.00 frames. ], tot_loss[loss=0.2406, simple_loss=0.2824, pruned_loss=0.09939, over 2595519.34 frames. ], batch size: 157, lr: 1.66e-02, grad_scale: 16.0 2023-04-16 16:37:47,835 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=19679.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:38:03,035 INFO [train.py:893] (1/4) Epoch 8, batch 800, loss[loss=0.2397, simple_loss=0.281, pruned_loss=0.09924, over 13519.00 frames. ], tot_loss[loss=0.2418, simple_loss=0.2839, pruned_loss=0.09981, over 2611565.67 frames. ], batch size: 70, lr: 1.66e-02, grad_scale: 16.0 2023-04-16 16:38:14,376 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.290e+02 3.416e+02 4.034e+02 4.840e+02 9.110e+02, threshold=8.068e+02, percent-clipped=3.0 2023-04-16 16:38:17,254 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7546, 4.2322, 3.8160, 3.9644, 3.9785, 4.4118, 4.1336, 4.1193], device='cuda:1'), covar=tensor([0.0347, 0.0249, 0.0323, 0.1055, 0.0237, 0.0209, 0.0291, 0.0393], device='cuda:1'), in_proj_covar=tensor([0.0107, 0.0112, 0.0119, 0.0212, 0.0118, 0.0135, 0.0118, 0.0090], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-16 16:38:49,580 INFO [train.py:893] (1/4) Epoch 8, batch 850, loss[loss=0.2485, simple_loss=0.291, pruned_loss=0.103, over 13364.00 frames. ], tot_loss[loss=0.2429, simple_loss=0.2848, pruned_loss=0.1006, over 2620762.41 frames. ], batch size: 109, lr: 1.66e-02, grad_scale: 16.0 2023-04-16 16:38:53,937 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19750.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:39:17,894 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1759, 2.7979, 2.2428, 4.0113, 4.6145, 3.4630, 4.5176, 4.1644], device='cuda:1'), covar=tensor([0.0104, 0.0690, 0.0973, 0.0095, 0.0062, 0.0388, 0.0085, 0.0067], device='cuda:1'), in_proj_covar=tensor([0.0053, 0.0070, 0.0080, 0.0059, 0.0042, 0.0064, 0.0040, 0.0050], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:39:18,722 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1422, 2.7906, 2.0948, 3.9585, 4.5879, 3.3589, 4.4916, 4.1374], device='cuda:1'), covar=tensor([0.0116, 0.0711, 0.1062, 0.0100, 0.0064, 0.0423, 0.0079, 0.0065], device='cuda:1'), in_proj_covar=tensor([0.0053, 0.0070, 0.0080, 0.0059, 0.0042, 0.0064, 0.0040, 0.0050], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:39:26,254 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=19785.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 16:39:35,730 INFO [train.py:893] (1/4) Epoch 8, batch 900, loss[loss=0.2606, simple_loss=0.2916, pruned_loss=0.1147, over 11908.00 frames. ], tot_loss[loss=0.2422, simple_loss=0.2838, pruned_loss=0.1002, over 2629097.61 frames. ], batch size: 157, lr: 1.65e-02, grad_scale: 16.0 2023-04-16 16:39:47,274 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.192e+02 3.363e+02 4.160e+02 5.148e+02 9.333e+02, threshold=8.319e+02, percent-clipped=2.0 2023-04-16 16:40:06,865 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-16 16:40:12,470 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=19833.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 16:40:22,072 INFO [train.py:893] (1/4) Epoch 8, batch 950, loss[loss=0.2225, simple_loss=0.2676, pruned_loss=0.08869, over 13450.00 frames. ], tot_loss[loss=0.2426, simple_loss=0.283, pruned_loss=0.1011, over 2634073.08 frames. ], batch size: 79, lr: 1.65e-02, grad_scale: 16.0 2023-04-16 16:40:42,224 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0688, 2.0216, 2.1163, 3.3470, 3.0962, 3.3630, 2.6812, 2.1042], device='cuda:1'), covar=tensor([0.0192, 0.1134, 0.0850, 0.0054, 0.0272, 0.0052, 0.0645, 0.0946], device='cuda:1'), in_proj_covar=tensor([0.0096, 0.0149, 0.0149, 0.0078, 0.0100, 0.0073, 0.0149, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:41:08,912 INFO [train.py:893] (1/4) Epoch 8, batch 1000, loss[loss=0.2289, simple_loss=0.2703, pruned_loss=0.09378, over 13513.00 frames. ], tot_loss[loss=0.2402, simple_loss=0.2811, pruned_loss=0.09969, over 2639641.99 frames. ], batch size: 70, lr: 1.65e-02, grad_scale: 16.0 2023-04-16 16:41:19,647 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.219e+02 3.520e+02 4.323e+02 5.176e+02 1.444e+03, threshold=8.646e+02, percent-clipped=3.0 2023-04-16 16:41:54,456 INFO [train.py:893] (1/4) Epoch 8, batch 1050, loss[loss=0.2316, simple_loss=0.2897, pruned_loss=0.08672, over 13222.00 frames. ], tot_loss[loss=0.237, simple_loss=0.2788, pruned_loss=0.09757, over 2645354.92 frames. ], batch size: 132, lr: 1.65e-02, grad_scale: 16.0 2023-04-16 16:41:56,563 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7200, 1.7640, 3.7675, 3.6938, 3.6470, 2.8895, 3.5272, 2.5983], device='cuda:1'), covar=tensor([0.2258, 0.1738, 0.0094, 0.0145, 0.0168, 0.0622, 0.0206, 0.0915], device='cuda:1'), in_proj_covar=tensor([0.0200, 0.0186, 0.0097, 0.0106, 0.0110, 0.0152, 0.0111, 0.0165], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-16 16:42:21,736 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=19974.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:42:41,838 INFO [train.py:893] (1/4) Epoch 8, batch 1100, loss[loss=0.2327, simple_loss=0.274, pruned_loss=0.09567, over 13460.00 frames. ], tot_loss[loss=0.2372, simple_loss=0.2796, pruned_loss=0.09739, over 2648614.43 frames. ], batch size: 65, lr: 1.65e-02, grad_scale: 16.0 2023-04-16 16:42:55,814 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.721e+02 3.510e+02 4.200e+02 4.928e+02 1.765e+03, threshold=8.400e+02, percent-clipped=2.0 2023-04-16 16:43:09,084 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20020.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:43:09,223 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1527, 4.2103, 3.5902, 2.9618, 2.9242, 2.3714, 4.3243, 2.5144], device='cuda:1'), covar=tensor([0.1103, 0.0270, 0.0472, 0.1098, 0.0583, 0.2380, 0.0154, 0.2711], device='cuda:1'), in_proj_covar=tensor([0.0276, 0.0245, 0.0246, 0.0267, 0.0210, 0.0265, 0.0173, 0.0318], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 16:43:16,812 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-04-16 16:43:31,183 INFO [train.py:893] (1/4) Epoch 8, batch 1150, loss[loss=0.2227, simple_loss=0.2727, pruned_loss=0.08637, over 13262.00 frames. ], tot_loss[loss=0.2362, simple_loss=0.2794, pruned_loss=0.09651, over 2649263.80 frames. ], batch size: 124, lr: 1.64e-02, grad_scale: 16.0 2023-04-16 16:43:34,128 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-04-16 16:43:36,387 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20050.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:43:49,790 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6247, 2.2366, 2.0454, 2.6149, 1.8951, 2.4852, 2.4372, 2.3686], device='cuda:1'), covar=tensor([0.0067, 0.0166, 0.0139, 0.0129, 0.0169, 0.0120, 0.0199, 0.0134], device='cuda:1'), in_proj_covar=tensor([0.0061, 0.0062, 0.0070, 0.0058, 0.0074, 0.0055, 0.0062, 0.0060], device='cuda:1'), out_proj_covar=tensor([7.0749e-05, 7.3096e-05, 8.6214e-05, 7.0209e-05, 9.0349e-05, 6.4324e-05, 7.5352e-05, 7.1074e-05], device='cuda:1') 2023-04-16 16:43:52,416 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5117, 3.7303, 4.3403, 3.1009, 2.6547, 3.0349, 4.4398, 4.5857], device='cuda:1'), covar=tensor([0.0927, 0.1059, 0.0297, 0.1258, 0.1484, 0.1074, 0.0148, 0.0109], device='cuda:1'), in_proj_covar=tensor([0.0205, 0.0194, 0.0153, 0.0198, 0.0194, 0.0161, 0.0142, 0.0105], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:44:03,322 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20081.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:44:11,234 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20089.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:44:15,943 INFO [train.py:893] (1/4) Epoch 8, batch 1200, loss[loss=0.2412, simple_loss=0.285, pruned_loss=0.09863, over 13025.00 frames. ], tot_loss[loss=0.2364, simple_loss=0.2799, pruned_loss=0.09641, over 2652911.62 frames. ], batch size: 142, lr: 1.64e-02, grad_scale: 16.0 2023-04-16 16:44:18,579 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20098.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:44:26,837 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.337e+02 3.392e+02 3.993e+02 4.511e+02 8.412e+02, threshold=7.987e+02, percent-clipped=1.0 2023-04-16 16:44:31,296 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20112.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:44:36,339 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8437, 2.4307, 2.0623, 3.6983, 4.2178, 3.1150, 4.1865, 3.8608], device='cuda:1'), covar=tensor([0.0088, 0.0730, 0.0916, 0.0086, 0.0057, 0.0359, 0.0068, 0.0060], device='cuda:1'), in_proj_covar=tensor([0.0053, 0.0071, 0.0081, 0.0060, 0.0042, 0.0064, 0.0040, 0.0051], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:44:42,161 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-16 16:44:45,877 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6073, 3.3970, 2.6830, 3.2624, 2.8400, 1.8299, 3.3376, 1.7795], device='cuda:1'), covar=tensor([0.0806, 0.0485, 0.0508, 0.0306, 0.0777, 0.2048, 0.0989, 0.1614], device='cuda:1'), in_proj_covar=tensor([0.0115, 0.0099, 0.0106, 0.0086, 0.0124, 0.0154, 0.0106, 0.0122], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:44:54,488 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-16 16:45:00,991 INFO [train.py:893] (1/4) Epoch 8, batch 1250, loss[loss=0.2866, simple_loss=0.3112, pruned_loss=0.131, over 11643.00 frames. ], tot_loss[loss=0.2382, simple_loss=0.2813, pruned_loss=0.09753, over 2653025.32 frames. ], batch size: 157, lr: 1.64e-02, grad_scale: 16.0 2023-04-16 16:45:06,239 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20150.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:45:27,076 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20173.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:45:41,050 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9022, 2.4659, 1.8709, 1.3067, 1.0960, 2.0944, 1.8514, 2.6172], device='cuda:1'), covar=tensor([0.0705, 0.0334, 0.1027, 0.1744, 0.0224, 0.0452, 0.0672, 0.0214], device='cuda:1'), in_proj_covar=tensor([0.0110, 0.0092, 0.0085, 0.0154, 0.0081, 0.0101, 0.0110, 0.0088], device='cuda:1'), out_proj_covar=tensor([8.6318e-05, 7.0883e-05, 7.0876e-05, 1.2636e-04, 7.1141e-05, 7.7816e-05, 8.7476e-05, 6.6448e-05], device='cuda:1') 2023-04-16 16:45:47,247 INFO [train.py:893] (1/4) Epoch 8, batch 1300, loss[loss=0.2712, simple_loss=0.3006, pruned_loss=0.1209, over 13248.00 frames. ], tot_loss[loss=0.24, simple_loss=0.2832, pruned_loss=0.09841, over 2657492.61 frames. ], batch size: 132, lr: 1.64e-02, grad_scale: 16.0 2023-04-16 16:45:51,396 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.3240, 2.9791, 2.6381, 1.6595, 1.7761, 2.5106, 2.4688, 3.2222], device='cuda:1'), covar=tensor([0.0609, 0.0295, 0.0660, 0.1568, 0.0574, 0.0340, 0.0522, 0.0264], device='cuda:1'), in_proj_covar=tensor([0.0111, 0.0093, 0.0085, 0.0155, 0.0081, 0.0101, 0.0110, 0.0089], device='cuda:1'), out_proj_covar=tensor([8.6654e-05, 7.1329e-05, 7.1132e-05, 1.2731e-04, 7.1084e-05, 7.8112e-05, 8.7708e-05, 6.7130e-05], device='cuda:1') 2023-04-16 16:45:55,896 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20204.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:45:59,489 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.442e+02 3.397e+02 3.954e+02 4.859e+02 9.243e+02, threshold=7.909e+02, percent-clipped=2.0 2023-04-16 16:46:30,175 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2023-04-16 16:46:33,712 INFO [train.py:893] (1/4) Epoch 8, batch 1350, loss[loss=0.237, simple_loss=0.2802, pruned_loss=0.09693, over 13520.00 frames. ], tot_loss[loss=0.2407, simple_loss=0.2837, pruned_loss=0.09881, over 2659503.06 frames. ], batch size: 76, lr: 1.64e-02, grad_scale: 16.0 2023-04-16 16:46:52,195 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20265.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:47:01,015 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20274.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:47:19,760 INFO [train.py:893] (1/4) Epoch 8, batch 1400, loss[loss=0.2463, simple_loss=0.2741, pruned_loss=0.1093, over 13374.00 frames. ], tot_loss[loss=0.2391, simple_loss=0.2823, pruned_loss=0.09794, over 2660647.51 frames. ], batch size: 62, lr: 1.64e-02, grad_scale: 16.0 2023-04-16 16:47:31,201 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.220e+02 3.226e+02 3.715e+02 4.443e+02 7.884e+02, threshold=7.431e+02, percent-clipped=0.0 2023-04-16 16:47:44,599 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20322.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:48:00,097 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0363, 4.4066, 4.0608, 4.1107, 4.1762, 3.9341, 4.4028, 4.4448], device='cuda:1'), covar=tensor([0.0193, 0.0250, 0.0219, 0.0306, 0.0279, 0.0274, 0.0310, 0.0250], device='cuda:1'), in_proj_covar=tensor([0.0101, 0.0148, 0.0111, 0.0139, 0.0102, 0.0142, 0.0101, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 16:48:05,621 INFO [train.py:893] (1/4) Epoch 8, batch 1450, loss[loss=0.2265, simple_loss=0.2756, pruned_loss=0.08874, over 13456.00 frames. ], tot_loss[loss=0.2394, simple_loss=0.2822, pruned_loss=0.09826, over 2661199.68 frames. ], batch size: 79, lr: 1.63e-02, grad_scale: 16.0 2023-04-16 16:48:30,733 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.05 vs. limit=5.0 2023-04-16 16:48:33,531 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20376.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:48:44,984 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20388.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:48:51,087 INFO [train.py:893] (1/4) Epoch 8, batch 1500, loss[loss=0.2499, simple_loss=0.2731, pruned_loss=0.1133, over 12060.00 frames. ], tot_loss[loss=0.2391, simple_loss=0.2821, pruned_loss=0.09805, over 2662222.59 frames. ], batch size: 49, lr: 1.63e-02, grad_scale: 16.0 2023-04-16 16:49:02,074 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.502e+02 3.447e+02 3.995e+02 5.100e+02 1.034e+03, threshold=7.990e+02, percent-clipped=5.0 2023-04-16 16:49:12,906 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20418.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:49:37,033 INFO [train.py:893] (1/4) Epoch 8, batch 1550, loss[loss=0.2103, simple_loss=0.2652, pruned_loss=0.07773, over 13498.00 frames. ], tot_loss[loss=0.2393, simple_loss=0.2825, pruned_loss=0.0981, over 2659788.58 frames. ], batch size: 81, lr: 1.63e-02, grad_scale: 16.0 2023-04-16 16:49:37,294 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20445.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:49:40,808 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20449.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:49:57,716 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20468.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:50:08,411 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20479.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:50:22,576 INFO [train.py:893] (1/4) Epoch 8, batch 1600, loss[loss=0.2768, simple_loss=0.3164, pruned_loss=0.1186, over 13541.00 frames. ], tot_loss[loss=0.2384, simple_loss=0.2819, pruned_loss=0.0974, over 2660379.95 frames. ], batch size: 91, lr: 1.63e-02, grad_scale: 16.0 2023-04-16 16:50:33,979 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.804e+02 3.330e+02 4.045e+02 4.918e+02 9.492e+02, threshold=8.090e+02, percent-clipped=1.0 2023-04-16 16:50:40,390 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9065, 4.1567, 2.6381, 4.2873, 3.9471, 2.1310, 3.2880, 2.5548], device='cuda:1'), covar=tensor([0.0252, 0.0244, 0.1199, 0.0189, 0.0243, 0.1445, 0.0660, 0.1429], device='cuda:1'), in_proj_covar=tensor([0.0128, 0.0129, 0.0170, 0.0106, 0.0112, 0.0158, 0.0148, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 16:51:08,737 INFO [train.py:893] (1/4) Epoch 8, batch 1650, loss[loss=0.235, simple_loss=0.2773, pruned_loss=0.0963, over 13217.00 frames. ], tot_loss[loss=0.2373, simple_loss=0.2816, pruned_loss=0.09653, over 2659404.73 frames. ], batch size: 132, lr: 1.63e-02, grad_scale: 16.0 2023-04-16 16:51:22,956 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20560.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:51:49,137 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=20588.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:51:52,467 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7479, 2.4132, 2.0436, 1.4220, 1.2452, 1.9696, 1.8791, 2.5695], device='cuda:1'), covar=tensor([0.0754, 0.0317, 0.0808, 0.1614, 0.0246, 0.0394, 0.0655, 0.0236], device='cuda:1'), in_proj_covar=tensor([0.0112, 0.0096, 0.0086, 0.0153, 0.0082, 0.0102, 0.0114, 0.0089], device='cuda:1'), out_proj_covar=tensor([8.7244e-05, 7.3364e-05, 7.1717e-05, 1.2515e-04, 7.1920e-05, 7.8028e-05, 9.0303e-05, 6.7404e-05], device='cuda:1') 2023-04-16 16:51:54,567 INFO [train.py:893] (1/4) Epoch 8, batch 1700, loss[loss=0.2269, simple_loss=0.2735, pruned_loss=0.09015, over 13368.00 frames. ], tot_loss[loss=0.2367, simple_loss=0.2813, pruned_loss=0.09604, over 2657128.76 frames. ], batch size: 113, lr: 1.62e-02, grad_scale: 16.0 2023-04-16 16:51:56,624 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.8016, 1.9997, 3.8223, 3.6058, 3.7289, 2.9617, 3.5532, 2.8149], device='cuda:1'), covar=tensor([0.2173, 0.1745, 0.0071, 0.0186, 0.0168, 0.0677, 0.0270, 0.0769], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0189, 0.0096, 0.0110, 0.0109, 0.0152, 0.0113, 0.0164], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-16 16:52:03,294 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4756, 3.7166, 4.2146, 2.9421, 2.7285, 2.9312, 4.2909, 4.4433], device='cuda:1'), covar=tensor([0.0885, 0.0936, 0.0295, 0.1451, 0.1406, 0.1195, 0.0217, 0.0118], device='cuda:1'), in_proj_covar=tensor([0.0207, 0.0197, 0.0157, 0.0198, 0.0196, 0.0162, 0.0142, 0.0106], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:52:06,199 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.424e+02 3.366e+02 4.046e+02 4.775e+02 1.003e+03, threshold=8.092e+02, percent-clipped=1.0 2023-04-16 16:52:15,479 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0420, 4.3797, 2.7207, 4.4409, 4.0844, 2.4148, 3.5959, 2.5888], device='cuda:1'), covar=tensor([0.0304, 0.0243, 0.1405, 0.0208, 0.0251, 0.1543, 0.0636, 0.1920], device='cuda:1'), in_proj_covar=tensor([0.0127, 0.0128, 0.0170, 0.0107, 0.0111, 0.0156, 0.0146, 0.0158], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 16:52:40,939 INFO [train.py:893] (1/4) Epoch 8, batch 1750, loss[loss=0.2273, simple_loss=0.2719, pruned_loss=0.09138, over 13310.00 frames. ], tot_loss[loss=0.2345, simple_loss=0.2794, pruned_loss=0.09477, over 2658684.17 frames. ], batch size: 67, lr: 1.62e-02, grad_scale: 16.0 2023-04-16 16:52:45,274 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=20649.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:53:09,311 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20676.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:53:26,203 INFO [train.py:893] (1/4) Epoch 8, batch 1800, loss[loss=0.2065, simple_loss=0.265, pruned_loss=0.074, over 13561.00 frames. ], tot_loss[loss=0.2332, simple_loss=0.2784, pruned_loss=0.09404, over 2663855.92 frames. ], batch size: 87, lr: 1.62e-02, grad_scale: 16.0 2023-04-16 16:53:37,551 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.294e+02 3.249e+02 3.823e+02 4.877e+02 1.148e+03, threshold=7.645e+02, percent-clipped=3.0 2023-04-16 16:53:53,242 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20724.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:54:07,401 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-04-16 16:54:11,932 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20744.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:54:12,585 INFO [train.py:893] (1/4) Epoch 8, batch 1850, loss[loss=0.2361, simple_loss=0.2762, pruned_loss=0.09804, over 13382.00 frames. ], tot_loss[loss=0.2337, simple_loss=0.2787, pruned_loss=0.09436, over 2665936.69 frames. ], batch size: 113, lr: 1.62e-02, grad_scale: 16.0 2023-04-16 16:54:12,875 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20745.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:54:15,155 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-16 16:54:33,357 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20768.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:54:39,671 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20774.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:54:48,839 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7188, 4.1423, 3.8567, 3.8955, 3.8911, 4.3641, 4.1218, 4.0130], device='cuda:1'), covar=tensor([0.0342, 0.0269, 0.0305, 0.1146, 0.0309, 0.0239, 0.0275, 0.0381], device='cuda:1'), in_proj_covar=tensor([0.0107, 0.0112, 0.0120, 0.0214, 0.0119, 0.0138, 0.0118, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-16 16:54:56,663 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20793.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:54:58,136 INFO [train.py:893] (1/4) Epoch 8, batch 1900, loss[loss=0.2361, simple_loss=0.2815, pruned_loss=0.09533, over 13451.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.2779, pruned_loss=0.09431, over 2667460.23 frames. ], batch size: 103, lr: 1.62e-02, grad_scale: 16.0 2023-04-16 16:55:06,008 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1045, 4.3486, 2.7859, 4.4625, 4.0676, 2.4316, 3.6570, 2.7768], device='cuda:1'), covar=tensor([0.0248, 0.0214, 0.1196, 0.0136, 0.0272, 0.1304, 0.0580, 0.1394], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0127, 0.0170, 0.0106, 0.0111, 0.0155, 0.0146, 0.0158], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 16:55:09,577 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.241e+02 3.078e+02 3.600e+02 4.474e+02 8.582e+02, threshold=7.200e+02, percent-clipped=1.0 2023-04-16 16:55:17,441 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20816.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:55:43,899 INFO [train.py:893] (1/4) Epoch 8, batch 1950, loss[loss=0.247, simple_loss=0.2948, pruned_loss=0.09961, over 13536.00 frames. ], tot_loss[loss=0.2333, simple_loss=0.2778, pruned_loss=0.09442, over 2663224.44 frames. ], batch size: 98, lr: 1.61e-02, grad_scale: 16.0 2023-04-16 16:55:58,004 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=20860.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:56:28,675 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4673, 1.8999, 2.2719, 3.8720, 3.5852, 3.9426, 2.9455, 2.0206], device='cuda:1'), covar=tensor([0.0248, 0.1481, 0.1154, 0.0068, 0.0213, 0.0039, 0.0631, 0.1251], device='cuda:1'), in_proj_covar=tensor([0.0102, 0.0150, 0.0153, 0.0081, 0.0104, 0.0073, 0.0150, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:56:29,847 INFO [train.py:893] (1/4) Epoch 8, batch 2000, loss[loss=0.3066, simple_loss=0.335, pruned_loss=0.1391, over 13530.00 frames. ], tot_loss[loss=0.235, simple_loss=0.2792, pruned_loss=0.09539, over 2661845.98 frames. ], batch size: 98, lr: 1.61e-02, grad_scale: 32.0 2023-04-16 16:56:36,108 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-16 16:56:40,199 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.374e+02 3.423e+02 4.035e+02 4.744e+02 7.852e+02, threshold=8.070e+02, percent-clipped=3.0 2023-04-16 16:56:41,229 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=20908.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:56:55,079 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7075, 2.6651, 1.9797, 3.5905, 4.0145, 3.1610, 3.9718, 3.6854], device='cuda:1'), covar=tensor([0.0077, 0.0620, 0.0850, 0.0080, 0.0054, 0.0324, 0.0050, 0.0067], device='cuda:1'), in_proj_covar=tensor([0.0054, 0.0073, 0.0081, 0.0062, 0.0044, 0.0066, 0.0040, 0.0053], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:57:11,588 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.2906, 3.5051, 3.8488, 2.8376, 2.5496, 2.7231, 4.0522, 4.2287], device='cuda:1'), covar=tensor([0.1037, 0.0952, 0.0374, 0.1409, 0.1526, 0.1158, 0.0291, 0.0123], device='cuda:1'), in_proj_covar=tensor([0.0208, 0.0201, 0.0161, 0.0199, 0.0199, 0.0161, 0.0147, 0.0109], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 16:57:13,956 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=20944.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:57:15,186 INFO [train.py:893] (1/4) Epoch 8, batch 2050, loss[loss=0.2031, simple_loss=0.2591, pruned_loss=0.07356, over 13475.00 frames. ], tot_loss[loss=0.2374, simple_loss=0.2817, pruned_loss=0.09658, over 2661595.94 frames. ], batch size: 79, lr: 1.61e-02, grad_scale: 32.0 2023-04-16 16:58:00,253 INFO [train.py:893] (1/4) Epoch 8, batch 2100, loss[loss=0.203, simple_loss=0.2516, pruned_loss=0.07722, over 13374.00 frames. ], tot_loss[loss=0.236, simple_loss=0.2805, pruned_loss=0.09569, over 2664131.47 frames. ], batch size: 73, lr: 1.61e-02, grad_scale: 32.0 2023-04-16 16:58:12,044 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.172e+02 3.303e+02 3.918e+02 4.795e+02 7.015e+02, threshold=7.837e+02, percent-clipped=0.0 2023-04-16 16:58:46,019 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21044.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:58:46,604 INFO [train.py:893] (1/4) Epoch 8, batch 2150, loss[loss=0.2387, simple_loss=0.2852, pruned_loss=0.09607, over 13533.00 frames. ], tot_loss[loss=0.2353, simple_loss=0.2803, pruned_loss=0.0952, over 2664884.75 frames. ], batch size: 85, lr: 1.61e-02, grad_scale: 32.0 2023-04-16 16:58:51,599 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1175, 4.3600, 4.0574, 4.1878, 4.1317, 4.6155, 4.3984, 4.2524], device='cuda:1'), covar=tensor([0.0258, 0.0290, 0.0272, 0.0980, 0.0288, 0.0270, 0.0284, 0.0280], device='cuda:1'), in_proj_covar=tensor([0.0105, 0.0113, 0.0120, 0.0213, 0.0119, 0.0137, 0.0120, 0.0092], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-16 16:58:53,287 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21052.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:59:13,003 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21074.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:59:28,930 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21092.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:59:31,167 INFO [train.py:893] (1/4) Epoch 8, batch 2200, loss[loss=0.2188, simple_loss=0.2701, pruned_loss=0.08381, over 13462.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.2802, pruned_loss=0.09484, over 2665641.56 frames. ], batch size: 79, lr: 1.61e-02, grad_scale: 32.0 2023-04-16 16:59:42,195 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.587e+02 3.381e+02 4.062e+02 4.939e+02 9.669e+02, threshold=8.123e+02, percent-clipped=2.0 2023-04-16 16:59:47,542 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21113.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:59:52,450 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21118.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 16:59:55,589 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21122.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:00:17,220 INFO [train.py:893] (1/4) Epoch 8, batch 2250, loss[loss=0.2058, simple_loss=0.2574, pruned_loss=0.07713, over 13474.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.2788, pruned_loss=0.09452, over 2664107.21 frames. ], batch size: 81, lr: 1.60e-02, grad_scale: 32.0 2023-04-16 17:00:48,386 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21179.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:00:49,131 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.8883, 4.3803, 4.2824, 4.4123, 4.0513, 4.2692, 4.8693, 4.3802], device='cuda:1'), covar=tensor([0.0977, 0.1088, 0.2676, 0.2982, 0.1097, 0.1620, 0.1019, 0.1057], device='cuda:1'), in_proj_covar=tensor([0.0251, 0.0283, 0.0360, 0.0363, 0.0202, 0.0275, 0.0330, 0.0244], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 17:01:02,557 INFO [train.py:893] (1/4) Epoch 8, batch 2300, loss[loss=0.2146, simple_loss=0.2616, pruned_loss=0.08375, over 13510.00 frames. ], tot_loss[loss=0.2318, simple_loss=0.2769, pruned_loss=0.0934, over 2665842.58 frames. ], batch size: 70, lr: 1.60e-02, grad_scale: 32.0 2023-04-16 17:01:14,068 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.369e+02 3.357e+02 4.098e+02 4.938e+02 8.843e+02, threshold=8.196e+02, percent-clipped=1.0 2023-04-16 17:01:36,198 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6578, 4.3942, 4.6897, 4.5614, 4.8790, 4.4233, 4.8867, 4.8890], device='cuda:1'), covar=tensor([0.0272, 0.0506, 0.0485, 0.0405, 0.0458, 0.0646, 0.0407, 0.0341], device='cuda:1'), in_proj_covar=tensor([0.0189, 0.0218, 0.0207, 0.0158, 0.0298, 0.0242, 0.0183, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 17:01:44,449 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5720, 4.6470, 4.0607, 3.4244, 3.5666, 2.6902, 4.8224, 2.9641], device='cuda:1'), covar=tensor([0.1026, 0.0191, 0.0421, 0.0896, 0.0420, 0.2149, 0.0124, 0.2388], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0246, 0.0247, 0.0266, 0.0209, 0.0264, 0.0169, 0.0311], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 17:01:47,490 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21244.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:01:48,036 INFO [train.py:893] (1/4) Epoch 8, batch 2350, loss[loss=0.2181, simple_loss=0.262, pruned_loss=0.08706, over 13543.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.2757, pruned_loss=0.09242, over 2667265.86 frames. ], batch size: 87, lr: 1.60e-02, grad_scale: 32.0 2023-04-16 17:01:49,987 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3683, 4.6864, 4.3854, 4.3344, 4.4362, 4.8441, 4.6585, 4.4520], device='cuda:1'), covar=tensor([0.0268, 0.0261, 0.0263, 0.1105, 0.0230, 0.0235, 0.0265, 0.0233], device='cuda:1'), in_proj_covar=tensor([0.0108, 0.0116, 0.0123, 0.0218, 0.0121, 0.0139, 0.0120, 0.0093], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-16 17:02:08,920 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-04-16 17:02:12,335 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-16 17:02:26,557 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5594, 4.3130, 4.5955, 4.6029, 4.8468, 4.3417, 4.9613, 4.8813], device='cuda:1'), covar=tensor([0.0337, 0.0579, 0.0645, 0.0461, 0.0515, 0.0848, 0.0332, 0.0396], device='cuda:1'), in_proj_covar=tensor([0.0191, 0.0221, 0.0210, 0.0159, 0.0302, 0.0248, 0.0185, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 17:02:31,269 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21292.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:02:34,366 INFO [train.py:893] (1/4) Epoch 8, batch 2400, loss[loss=0.2208, simple_loss=0.252, pruned_loss=0.09482, over 12830.00 frames. ], tot_loss[loss=0.2298, simple_loss=0.2752, pruned_loss=0.09217, over 2666879.63 frames. ], batch size: 52, lr: 1.60e-02, grad_scale: 32.0 2023-04-16 17:02:44,935 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.442e+02 3.489e+02 4.000e+02 4.780e+02 1.257e+03, threshold=8.000e+02, percent-clipped=2.0 2023-04-16 17:02:49,323 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21312.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:03:04,418 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21328.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:03:18,898 INFO [train.py:893] (1/4) Epoch 8, batch 2450, loss[loss=0.2292, simple_loss=0.2743, pruned_loss=0.09203, over 13401.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.2758, pruned_loss=0.09295, over 2661387.45 frames. ], batch size: 113, lr: 1.60e-02, grad_scale: 32.0 2023-04-16 17:03:44,778 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21373.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:04:00,100 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21389.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:04:04,619 INFO [train.py:893] (1/4) Epoch 8, batch 2500, loss[loss=0.2332, simple_loss=0.2806, pruned_loss=0.0929, over 13476.00 frames. ], tot_loss[loss=0.2309, simple_loss=0.276, pruned_loss=0.09289, over 2665690.22 frames. ], batch size: 79, lr: 1.60e-02, grad_scale: 32.0 2023-04-16 17:04:15,350 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.088e+02 3.094e+02 3.679e+02 4.725e+02 7.841e+02, threshold=7.359e+02, percent-clipped=0.0 2023-04-16 17:04:16,432 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21408.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:04:20,624 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0057, 3.7345, 3.9463, 2.5852, 4.4337, 4.2082, 4.0515, 4.5057], device='cuda:1'), covar=tensor([0.0205, 0.0133, 0.0123, 0.0951, 0.0129, 0.0159, 0.0142, 0.0078], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0037, 0.0057, 0.0083, 0.0072, 0.0066, 0.0057, 0.0051], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:04:49,328 INFO [train.py:893] (1/4) Epoch 8, batch 2550, loss[loss=0.2479, simple_loss=0.288, pruned_loss=0.1039, over 13045.00 frames. ], tot_loss[loss=0.2305, simple_loss=0.2758, pruned_loss=0.09263, over 2669144.83 frames. ], batch size: 142, lr: 1.59e-02, grad_scale: 32.0 2023-04-16 17:05:03,370 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.45 vs. limit=5.0 2023-04-16 17:05:15,275 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-16 17:05:16,183 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21474.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:05:36,126 INFO [train.py:893] (1/4) Epoch 8, batch 2600, loss[loss=0.2665, simple_loss=0.3016, pruned_loss=0.1157, over 13483.00 frames. ], tot_loss[loss=0.2311, simple_loss=0.2757, pruned_loss=0.09321, over 2666222.63 frames. ], batch size: 93, lr: 1.59e-02, grad_scale: 32.0 2023-04-16 17:05:45,870 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.283e+02 3.241e+02 4.304e+02 5.606e+02 8.528e+02, threshold=8.608e+02, percent-clipped=3.0 2023-04-16 17:06:17,820 INFO [train.py:893] (1/4) Epoch 8, batch 2650, loss[loss=0.2174, simple_loss=0.2675, pruned_loss=0.08362, over 13424.00 frames. ], tot_loss[loss=0.2316, simple_loss=0.2759, pruned_loss=0.09365, over 2665936.61 frames. ], batch size: 106, lr: 1.59e-02, grad_scale: 16.0 2023-04-16 17:07:13,578 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 17:07:23,773 INFO [train.py:893] (1/4) Epoch 9, batch 0, loss[loss=0.2389, simple_loss=0.2816, pruned_loss=0.09805, over 13458.00 frames. ], tot_loss[loss=0.2389, simple_loss=0.2816, pruned_loss=0.09805, over 13458.00 frames. ], batch size: 95, lr: 1.50e-02, grad_scale: 16.0 2023-04-16 17:07:23,773 INFO [train.py:918] (1/4) Computing validation loss 2023-04-16 17:07:38,739 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2108, 4.5989, 4.5436, 4.3819, 4.4322, 4.4007, 4.7061, 4.7288], device='cuda:1'), covar=tensor([0.0195, 0.0216, 0.0165, 0.0272, 0.0286, 0.0237, 0.0217, 0.0183], device='cuda:1'), in_proj_covar=tensor([0.0102, 0.0149, 0.0114, 0.0140, 0.0109, 0.0148, 0.0102, 0.0123], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 17:07:46,617 INFO [train.py:927] (1/4) Epoch 9, validation: loss=0.1695, simple_loss=0.223, pruned_loss=0.05802, over 2446609.00 frames. 2023-04-16 17:07:46,618 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12773MB 2023-04-16 17:07:48,573 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21595.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:08:00,062 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.409e+02 3.326e+02 3.891e+02 4.627e+02 8.563e+02, threshold=7.782e+02, percent-clipped=0.0 2023-04-16 17:08:04,385 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21613.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:08:32,552 INFO [train.py:893] (1/4) Epoch 9, batch 50, loss[loss=0.2057, simple_loss=0.2453, pruned_loss=0.083, over 13213.00 frames. ], tot_loss[loss=0.2285, simple_loss=0.2726, pruned_loss=0.09221, over 603731.81 frames. ], batch size: 58, lr: 1.50e-02, grad_scale: 16.0 2023-04-16 17:08:43,548 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21656.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:08:51,719 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-04-16 17:08:55,512 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21668.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:08:57,831 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-16 17:08:57,831 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-16 17:08:57,832 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-16 17:08:57,838 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-16 17:08:57,855 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-16 17:08:57,874 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-16 17:08:57,883 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-16 17:09:00,583 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21674.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:09:09,266 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21684.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:09:13,669 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.5740, 3.2699, 2.7723, 3.1581, 2.9083, 1.8334, 3.3084, 1.8156], device='cuda:1'), covar=tensor([0.0719, 0.0690, 0.0473, 0.0315, 0.0733, 0.1912, 0.1043, 0.1433], device='cuda:1'), in_proj_covar=tensor([0.0116, 0.0101, 0.0107, 0.0091, 0.0129, 0.0156, 0.0113, 0.0124], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:09:18,942 INFO [train.py:893] (1/4) Epoch 9, batch 100, loss[loss=0.2327, simple_loss=0.2532, pruned_loss=0.1062, over 11119.00 frames. ], tot_loss[loss=0.23, simple_loss=0.2728, pruned_loss=0.09354, over 1063557.40 frames. ], batch size: 45, lr: 1.50e-02, grad_scale: 16.0 2023-04-16 17:09:31,064 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.350e+02 3.037e+02 3.556e+02 4.306e+02 7.126e+02, threshold=7.112e+02, percent-clipped=0.0 2023-04-16 17:09:31,298 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21708.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:09:53,606 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3759, 4.2873, 3.7874, 3.0914, 2.9653, 2.3948, 4.4934, 2.5960], device='cuda:1'), covar=tensor([0.0979, 0.0261, 0.0458, 0.1035, 0.0561, 0.2360, 0.0153, 0.2443], device='cuda:1'), in_proj_covar=tensor([0.0272, 0.0247, 0.0245, 0.0264, 0.0208, 0.0262, 0.0169, 0.0311], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 17:09:57,596 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6771, 3.9201, 4.2798, 3.0464, 2.5708, 3.0006, 4.4861, 4.5967], device='cuda:1'), covar=tensor([0.0844, 0.0985, 0.0362, 0.1413, 0.1500, 0.1198, 0.0192, 0.0112], device='cuda:1'), in_proj_covar=tensor([0.0206, 0.0202, 0.0161, 0.0197, 0.0194, 0.0162, 0.0149, 0.0111], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:10:03,142 INFO [train.py:893] (1/4) Epoch 9, batch 150, loss[loss=0.2179, simple_loss=0.2664, pruned_loss=0.08471, over 13468.00 frames. ], tot_loss[loss=0.2343, simple_loss=0.2767, pruned_loss=0.09592, over 1422496.06 frames. ], batch size: 79, lr: 1.50e-02, grad_scale: 16.0 2023-04-16 17:10:05,447 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.02 vs. limit=5.0 2023-04-16 17:10:14,139 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21756.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:10:22,385 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21763.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:10:31,526 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21774.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:10:50,072 INFO [train.py:893] (1/4) Epoch 9, batch 200, loss[loss=0.2532, simple_loss=0.294, pruned_loss=0.1062, over 11875.00 frames. ], tot_loss[loss=0.236, simple_loss=0.278, pruned_loss=0.09699, over 1685107.92 frames. ], batch size: 157, lr: 1.50e-02, grad_scale: 16.0 2023-04-16 17:10:52,928 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-04-16 17:11:02,744 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.061e+02 3.486e+02 4.224e+02 5.229e+02 1.059e+03, threshold=8.447e+02, percent-clipped=3.0 2023-04-16 17:11:16,560 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=21822.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:11:18,353 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21824.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:11:35,551 INFO [train.py:893] (1/4) Epoch 9, batch 250, loss[loss=0.2242, simple_loss=0.2765, pruned_loss=0.08592, over 13458.00 frames. ], tot_loss[loss=0.2351, simple_loss=0.2772, pruned_loss=0.09647, over 1898004.29 frames. ], batch size: 79, lr: 1.50e-02, grad_scale: 16.0 2023-04-16 17:11:41,588 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21850.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:11:55,337 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1872, 2.1792, 2.4916, 3.8035, 3.5241, 3.8900, 3.0764, 2.2270], device='cuda:1'), covar=tensor([0.0385, 0.1209, 0.0932, 0.0058, 0.0228, 0.0045, 0.0650, 0.1203], device='cuda:1'), in_proj_covar=tensor([0.0100, 0.0145, 0.0149, 0.0080, 0.0101, 0.0073, 0.0148, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:12:22,908 INFO [train.py:893] (1/4) Epoch 9, batch 300, loss[loss=0.2459, simple_loss=0.2944, pruned_loss=0.09866, over 13447.00 frames. ], tot_loss[loss=0.2349, simple_loss=0.2778, pruned_loss=0.09605, over 2064123.75 frames. ], batch size: 100, lr: 1.49e-02, grad_scale: 8.0 2023-04-16 17:12:35,096 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.144e+02 3.113e+02 3.883e+02 4.592e+02 8.864e+02, threshold=7.767e+02, percent-clipped=2.0 2023-04-16 17:12:37,826 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=21911.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:13:07,829 INFO [train.py:893] (1/4) Epoch 9, batch 350, loss[loss=0.2309, simple_loss=0.2712, pruned_loss=0.09528, over 13225.00 frames. ], tot_loss[loss=0.2346, simple_loss=0.2775, pruned_loss=0.09585, over 2197066.17 frames. ], batch size: 132, lr: 1.49e-02, grad_scale: 8.0 2023-04-16 17:13:12,981 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6954, 3.3743, 2.6878, 3.1615, 2.8318, 1.7376, 3.3575, 1.7829], device='cuda:1'), covar=tensor([0.0628, 0.0514, 0.0548, 0.0349, 0.0738, 0.2053, 0.0911, 0.1527], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0100, 0.0107, 0.0092, 0.0128, 0.0155, 0.0114, 0.0124], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:13:15,256 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21951.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:13:22,765 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1019, 4.5055, 4.1485, 4.1244, 4.1174, 4.5532, 4.3366, 4.1413], device='cuda:1'), covar=tensor([0.0272, 0.0185, 0.0265, 0.1097, 0.0274, 0.0247, 0.0260, 0.0314], device='cuda:1'), in_proj_covar=tensor([0.0112, 0.0118, 0.0123, 0.0221, 0.0123, 0.0141, 0.0123, 0.0096], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-16 17:13:29,410 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21968.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:13:30,128 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=21969.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:13:31,061 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=21970.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:13:32,819 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.1563, 1.9951, 1.9932, 2.3627, 1.6850, 2.4234, 2.3466, 2.1605], device='cuda:1'), covar=tensor([0.0082, 0.0175, 0.0145, 0.0163, 0.0201, 0.0086, 0.0192, 0.0121], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0066, 0.0073, 0.0062, 0.0079, 0.0058, 0.0067, 0.0064], device='cuda:1'), out_proj_covar=tensor([7.2996e-05, 7.7101e-05, 8.7961e-05, 7.3674e-05, 9.5679e-05, 6.6933e-05, 7.9788e-05, 7.4139e-05], device='cuda:1') 2023-04-16 17:13:45,535 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=21984.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:13:53,621 INFO [train.py:893] (1/4) Epoch 9, batch 400, loss[loss=0.2149, simple_loss=0.2501, pruned_loss=0.08989, over 12720.00 frames. ], tot_loss[loss=0.2354, simple_loss=0.279, pruned_loss=0.0959, over 2298604.00 frames. ], batch size: 52, lr: 1.49e-02, grad_scale: 8.0 2023-04-16 17:14:08,549 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0173, 3.6255, 3.2003, 3.6918, 3.1417, 2.2091, 3.6554, 2.0975], device='cuda:1'), covar=tensor([0.0585, 0.0411, 0.0445, 0.0219, 0.0630, 0.1726, 0.0905, 0.1420], device='cuda:1'), in_proj_covar=tensor([0.0115, 0.0099, 0.0105, 0.0090, 0.0125, 0.0153, 0.0113, 0.0122], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:14:12,158 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.421e+02 3.487e+02 4.119e+02 4.911e+02 7.347e+02, threshold=8.238e+02, percent-clipped=0.0 2023-04-16 17:14:18,897 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22016.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:14:28,370 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.65 vs. limit=2.0 2023-04-16 17:14:31,272 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22031.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:14:32,009 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22032.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:14:44,306 INFO [train.py:893] (1/4) Epoch 9, batch 450, loss[loss=0.2485, simple_loss=0.295, pruned_loss=0.101, over 13455.00 frames. ], tot_loss[loss=0.2364, simple_loss=0.2802, pruned_loss=0.09632, over 2380004.08 frames. ], batch size: 103, lr: 1.49e-02, grad_scale: 8.0 2023-04-16 17:15:02,832 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1885, 3.8511, 4.1187, 2.7913, 4.6017, 4.1834, 4.3047, 4.5133], device='cuda:1'), covar=tensor([0.0181, 0.0123, 0.0125, 0.0789, 0.0107, 0.0185, 0.0110, 0.0100], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0036, 0.0055, 0.0079, 0.0070, 0.0066, 0.0056, 0.0049], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:15:08,193 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-16 17:15:27,918 INFO [train.py:893] (1/4) Epoch 9, batch 500, loss[loss=0.2376, simple_loss=0.2915, pruned_loss=0.09186, over 13319.00 frames. ], tot_loss[loss=0.2365, simple_loss=0.2806, pruned_loss=0.09622, over 2439410.60 frames. ], batch size: 118, lr: 1.49e-02, grad_scale: 8.0 2023-04-16 17:15:42,973 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.552e+02 3.311e+02 3.830e+02 4.684e+02 7.371e+02, threshold=7.659e+02, percent-clipped=0.0 2023-04-16 17:15:51,401 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22119.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:16:01,950 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22130.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:16:14,797 INFO [train.py:893] (1/4) Epoch 9, batch 550, loss[loss=0.2146, simple_loss=0.2678, pruned_loss=0.0807, over 13349.00 frames. ], tot_loss[loss=0.2351, simple_loss=0.2796, pruned_loss=0.09533, over 2487517.26 frames. ], batch size: 73, lr: 1.49e-02, grad_scale: 8.0 2023-04-16 17:16:44,379 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22176.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:16:52,954 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-04-16 17:16:56,933 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22191.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:16:59,766 INFO [train.py:893] (1/4) Epoch 9, batch 600, loss[loss=0.2437, simple_loss=0.2905, pruned_loss=0.09851, over 13441.00 frames. ], tot_loss[loss=0.2339, simple_loss=0.278, pruned_loss=0.09491, over 2527573.47 frames. ], batch size: 103, lr: 1.48e-02, grad_scale: 8.0 2023-04-16 17:17:11,981 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22206.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:17:13,748 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9868, 3.7638, 4.0217, 2.6026, 4.4083, 4.0955, 4.1958, 4.3475], device='cuda:1'), covar=tensor([0.0223, 0.0144, 0.0131, 0.0991, 0.0137, 0.0185, 0.0114, 0.0097], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0036, 0.0057, 0.0081, 0.0070, 0.0067, 0.0057, 0.0049], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:17:14,238 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.989e+02 3.192e+02 3.802e+02 4.959e+02 1.314e+03, threshold=7.604e+02, percent-clipped=2.0 2023-04-16 17:17:19,450 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22215.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:17:24,454 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.4883, 2.8739, 2.4655, 2.6984, 2.7233, 1.6839, 2.9535, 1.7962], device='cuda:1'), covar=tensor([0.0590, 0.0883, 0.0456, 0.0421, 0.0613, 0.1808, 0.0763, 0.1461], device='cuda:1'), in_proj_covar=tensor([0.0116, 0.0102, 0.0105, 0.0089, 0.0125, 0.0153, 0.0114, 0.0123], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:17:39,853 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22237.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:17:39,875 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22237.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:17:45,621 INFO [train.py:893] (1/4) Epoch 9, batch 650, loss[loss=0.2174, simple_loss=0.2679, pruned_loss=0.08343, over 13538.00 frames. ], tot_loss[loss=0.2329, simple_loss=0.2774, pruned_loss=0.09424, over 2555834.83 frames. ], batch size: 85, lr: 1.48e-02, grad_scale: 8.0 2023-04-16 17:17:46,046 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.55 vs. limit=5.0 2023-04-16 17:17:51,413 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22251.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:17:52,424 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.2786, 2.0307, 1.9136, 2.4173, 1.7304, 2.4228, 2.2142, 2.1109], device='cuda:1'), covar=tensor([0.0091, 0.0164, 0.0141, 0.0114, 0.0189, 0.0111, 0.0205, 0.0149], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0068, 0.0076, 0.0064, 0.0082, 0.0059, 0.0069, 0.0065], device='cuda:1'), out_proj_covar=tensor([7.5200e-05, 8.0257e-05, 9.1855e-05, 7.5933e-05, 9.8662e-05, 6.8106e-05, 8.2283e-05, 7.4900e-05], device='cuda:1') 2023-04-16 17:18:04,604 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 2023-04-16 17:18:08,521 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22269.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:18:14,406 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22276.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:18:16,171 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.2462, 4.2338, 3.6258, 3.0979, 3.1646, 2.5093, 4.4381, 2.6127], device='cuda:1'), covar=tensor([0.1212, 0.0238, 0.0523, 0.1095, 0.0534, 0.2567, 0.0149, 0.2828], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0251, 0.0247, 0.0267, 0.0212, 0.0266, 0.0172, 0.0316], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 17:18:30,010 INFO [train.py:893] (1/4) Epoch 9, batch 700, loss[loss=0.2416, simple_loss=0.2892, pruned_loss=0.09702, over 13540.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.2764, pruned_loss=0.09332, over 2582959.44 frames. ], batch size: 98, lr: 1.48e-02, grad_scale: 8.0 2023-04-16 17:18:34,220 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22298.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:18:34,886 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22299.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:18:42,936 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.092e+02 3.305e+02 4.012e+02 4.954e+02 1.106e+03, threshold=8.025e+02, percent-clipped=3.0 2023-04-16 17:18:43,335 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6589, 3.7839, 4.2616, 3.1000, 2.8137, 3.0835, 4.4773, 4.6128], device='cuda:1'), covar=tensor([0.1000, 0.1152, 0.0407, 0.1424, 0.1508, 0.1217, 0.0213, 0.0145], device='cuda:1'), in_proj_covar=tensor([0.0206, 0.0204, 0.0161, 0.0198, 0.0196, 0.0163, 0.0148, 0.0112], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:18:49,734 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22317.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:18:59,341 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22326.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:19:14,743 INFO [train.py:893] (1/4) Epoch 9, batch 750, loss[loss=0.2202, simple_loss=0.2676, pruned_loss=0.08638, over 13464.00 frames. ], tot_loss[loss=0.2313, simple_loss=0.2759, pruned_loss=0.09332, over 2599378.15 frames. ], batch size: 79, lr: 1.48e-02, grad_scale: 8.0 2023-04-16 17:19:46,341 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-04-16 17:19:53,915 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22385.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:20:01,674 INFO [train.py:893] (1/4) Epoch 9, batch 800, loss[loss=0.2281, simple_loss=0.2777, pruned_loss=0.08925, over 13523.00 frames. ], tot_loss[loss=0.2331, simple_loss=0.2775, pruned_loss=0.09434, over 2614413.82 frames. ], batch size: 85, lr: 1.48e-02, grad_scale: 8.0 2023-04-16 17:20:14,430 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.259e+02 3.347e+02 3.901e+02 4.564e+02 8.768e+02, threshold=7.802e+02, percent-clipped=1.0 2023-04-16 17:20:24,338 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22419.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:20:45,106 INFO [train.py:893] (1/4) Epoch 9, batch 850, loss[loss=0.2277, simple_loss=0.2823, pruned_loss=0.08652, over 13454.00 frames. ], tot_loss[loss=0.234, simple_loss=0.279, pruned_loss=0.09454, over 2629547.22 frames. ], batch size: 103, lr: 1.48e-02, grad_scale: 8.0 2023-04-16 17:20:47,863 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22446.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:21:06,433 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22467.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:21:24,565 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22486.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:21:31,014 INFO [train.py:893] (1/4) Epoch 9, batch 900, loss[loss=0.2197, simple_loss=0.2647, pruned_loss=0.08732, over 13386.00 frames. ], tot_loss[loss=0.2341, simple_loss=0.2784, pruned_loss=0.09483, over 2631782.81 frames. ], batch size: 113, lr: 1.47e-02, grad_scale: 8.0 2023-04-16 17:21:40,968 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22506.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:21:43,899 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.166e+02 3.283e+02 3.837e+02 4.867e+02 8.567e+02, threshold=7.675e+02, percent-clipped=1.0 2023-04-16 17:22:00,013 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-16 17:22:04,284 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22532.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:22:15,460 INFO [train.py:893] (1/4) Epoch 9, batch 950, loss[loss=0.2379, simple_loss=0.2784, pruned_loss=0.09871, over 13536.00 frames. ], tot_loss[loss=0.2337, simple_loss=0.2775, pruned_loss=0.09501, over 2641066.02 frames. ], batch size: 83, lr: 1.47e-02, grad_scale: 8.0 2023-04-16 17:22:24,614 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22554.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:22:39,546 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22571.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:22:45,146 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1097, 4.2289, 2.5935, 4.3050, 4.0686, 2.3410, 3.5924, 2.5977], device='cuda:1'), covar=tensor([0.0267, 0.0303, 0.1696, 0.0166, 0.0258, 0.1657, 0.0579, 0.2027], device='cuda:1'), in_proj_covar=tensor([0.0133, 0.0134, 0.0174, 0.0116, 0.0115, 0.0160, 0.0149, 0.0163], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 17:22:59,600 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.1455, 4.9455, 5.1944, 4.9892, 5.4578, 4.9837, 5.4758, 5.4478], device='cuda:1'), covar=tensor([0.0304, 0.0467, 0.0484, 0.0434, 0.0429, 0.0675, 0.0398, 0.0412], device='cuda:1'), in_proj_covar=tensor([0.0197, 0.0223, 0.0209, 0.0157, 0.0306, 0.0253, 0.0191, 0.0205], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 17:22:59,605 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22593.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:23:00,239 INFO [train.py:893] (1/4) Epoch 9, batch 1000, loss[loss=0.2155, simple_loss=0.2632, pruned_loss=0.08393, over 13547.00 frames. ], tot_loss[loss=0.2317, simple_loss=0.2754, pruned_loss=0.09398, over 2641881.62 frames. ], batch size: 72, lr: 1.47e-02, grad_scale: 8.0 2023-04-16 17:23:01,408 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1914, 4.3336, 2.6526, 4.4006, 4.1620, 2.6201, 3.6546, 2.6870], device='cuda:1'), covar=tensor([0.0195, 0.0293, 0.1322, 0.0136, 0.0202, 0.1182, 0.0462, 0.1570], device='cuda:1'), in_proj_covar=tensor([0.0131, 0.0133, 0.0172, 0.0115, 0.0114, 0.0157, 0.0147, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 17:23:09,756 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7811, 3.7944, 4.3899, 3.3350, 2.9817, 3.0776, 4.6304, 4.7380], device='cuda:1'), covar=tensor([0.0843, 0.0992, 0.0341, 0.1252, 0.1274, 0.1116, 0.0195, 0.0093], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0205, 0.0163, 0.0201, 0.0195, 0.0165, 0.0151, 0.0114], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:23:15,071 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.089e+02 2.969e+02 3.521e+02 4.471e+02 1.035e+03, threshold=7.042e+02, percent-clipped=1.0 2023-04-16 17:23:30,032 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22626.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:23:42,413 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22639.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:23:46,956 INFO [train.py:893] (1/4) Epoch 9, batch 1050, loss[loss=0.2367, simple_loss=0.2849, pruned_loss=0.09421, over 13482.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.2733, pruned_loss=0.09247, over 2649739.63 frames. ], batch size: 93, lr: 1.47e-02, grad_scale: 8.0 2023-04-16 17:23:52,230 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9981, 3.7451, 4.1023, 2.8222, 4.3520, 4.0427, 4.1339, 4.3744], device='cuda:1'), covar=tensor([0.0232, 0.0129, 0.0137, 0.0872, 0.0178, 0.0195, 0.0157, 0.0120], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0036, 0.0057, 0.0081, 0.0071, 0.0068, 0.0058, 0.0051], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:24:13,846 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22674.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:24:30,773 INFO [train.py:893] (1/4) Epoch 9, batch 1100, loss[loss=0.2419, simple_loss=0.2999, pruned_loss=0.09192, over 13355.00 frames. ], tot_loss[loss=0.2287, simple_loss=0.2737, pruned_loss=0.09183, over 2649851.26 frames. ], batch size: 109, lr: 1.47e-02, grad_scale: 8.0 2023-04-16 17:24:38,067 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22700.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 17:24:45,842 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.277e+02 3.274e+02 3.908e+02 4.877e+02 1.233e+03, threshold=7.817e+02, percent-clipped=3.0 2023-04-16 17:25:15,064 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22741.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:25:17,336 INFO [train.py:893] (1/4) Epoch 9, batch 1150, loss[loss=0.2125, simple_loss=0.2641, pruned_loss=0.08042, over 13347.00 frames. ], tot_loss[loss=0.2283, simple_loss=0.2738, pruned_loss=0.09144, over 2649208.42 frames. ], batch size: 73, lr: 1.47e-02, grad_scale: 8.0 2023-04-16 17:25:36,924 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.1123, 4.9408, 5.2439, 4.9128, 5.4509, 4.9368, 5.5125, 5.5185], device='cuda:1'), covar=tensor([0.0319, 0.0470, 0.0521, 0.0520, 0.0496, 0.0745, 0.0455, 0.0406], device='cuda:1'), in_proj_covar=tensor([0.0201, 0.0230, 0.0215, 0.0164, 0.0315, 0.0260, 0.0196, 0.0208], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 17:25:54,962 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22786.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:26:02,609 INFO [train.py:893] (1/4) Epoch 9, batch 1200, loss[loss=0.2254, simple_loss=0.2687, pruned_loss=0.091, over 13538.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.2735, pruned_loss=0.0911, over 2647898.32 frames. ], batch size: 87, lr: 1.47e-02, grad_scale: 8.0 2023-04-16 17:26:02,900 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22794.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:26:12,967 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3928, 2.2275, 2.4853, 3.7471, 3.4203, 3.8480, 2.9868, 2.0942], device='cuda:1'), covar=tensor([0.0227, 0.0968, 0.0814, 0.0043, 0.0254, 0.0027, 0.0557, 0.1049], device='cuda:1'), in_proj_covar=tensor([0.0103, 0.0147, 0.0151, 0.0082, 0.0103, 0.0071, 0.0149, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:26:16,760 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.433e+02 3.622e+02 4.040e+02 4.982e+02 8.406e+02, threshold=8.080e+02, percent-clipped=3.0 2023-04-16 17:26:28,763 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-16 17:26:31,455 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=22825.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:26:38,612 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22832.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:26:40,121 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-16 17:26:40,207 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22834.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:26:48,374 INFO [train.py:893] (1/4) Epoch 9, batch 1250, loss[loss=0.2194, simple_loss=0.2669, pruned_loss=0.08601, over 13543.00 frames. ], tot_loss[loss=0.2303, simple_loss=0.2751, pruned_loss=0.09268, over 2643226.97 frames. ], batch size: 87, lr: 1.46e-02, grad_scale: 8.0 2023-04-16 17:26:59,423 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22855.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:27:14,027 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22871.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:27:21,385 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22880.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:27:27,944 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=22886.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:27:35,085 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=22893.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:27:35,641 INFO [train.py:893] (1/4) Epoch 9, batch 1300, loss[loss=0.2373, simple_loss=0.2763, pruned_loss=0.09911, over 13550.00 frames. ], tot_loss[loss=0.2315, simple_loss=0.2763, pruned_loss=0.09335, over 2643971.97 frames. ], batch size: 76, lr: 1.46e-02, grad_scale: 8.0 2023-04-16 17:27:48,046 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.165e+02 3.359e+02 3.921e+02 4.678e+02 8.201e+02, threshold=7.842e+02, percent-clipped=1.0 2023-04-16 17:27:55,435 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0180, 4.2309, 2.8467, 4.1086, 3.9596, 2.4934, 3.5467, 2.7147], device='cuda:1'), covar=tensor([0.0227, 0.0258, 0.1119, 0.0176, 0.0238, 0.1291, 0.0564, 0.1498], device='cuda:1'), in_proj_covar=tensor([0.0130, 0.0131, 0.0168, 0.0115, 0.0113, 0.0154, 0.0146, 0.0159], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 17:27:57,733 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22919.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:28:10,569 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.5281, 3.1304, 2.6697, 3.0251, 2.7999, 1.6963, 3.2317, 1.7069], device='cuda:1'), covar=tensor([0.0709, 0.0843, 0.0444, 0.0435, 0.0717, 0.2190, 0.0788, 0.1639], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0105, 0.0106, 0.0092, 0.0131, 0.0160, 0.0118, 0.0127], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0001, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:28:16,920 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=22941.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:28:19,273 INFO [train.py:893] (1/4) Epoch 9, batch 1350, loss[loss=0.2444, simple_loss=0.2847, pruned_loss=0.1021, over 13510.00 frames. ], tot_loss[loss=0.2316, simple_loss=0.2764, pruned_loss=0.09336, over 2648953.42 frames. ], batch size: 87, lr: 1.46e-02, grad_scale: 8.0 2023-04-16 17:28:52,475 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4536, 3.4230, 2.6761, 4.3952, 5.0265, 3.7216, 4.9683, 4.5256], device='cuda:1'), covar=tensor([0.0085, 0.0491, 0.0720, 0.0094, 0.0045, 0.0304, 0.0052, 0.0066], device='cuda:1'), in_proj_covar=tensor([0.0058, 0.0078, 0.0085, 0.0063, 0.0046, 0.0068, 0.0042, 0.0056], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:29:04,923 INFO [train.py:893] (1/4) Epoch 9, batch 1400, loss[loss=0.226, simple_loss=0.2652, pruned_loss=0.09339, over 13364.00 frames. ], tot_loss[loss=0.2313, simple_loss=0.2763, pruned_loss=0.09313, over 2653386.67 frames. ], batch size: 67, lr: 1.46e-02, grad_scale: 8.0 2023-04-16 17:29:05,941 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=22995.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 17:29:12,842 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.7121, 4.5462, 4.7633, 4.6322, 5.0081, 4.5661, 4.9980, 4.9535], device='cuda:1'), covar=tensor([0.0313, 0.0458, 0.0569, 0.0402, 0.0472, 0.0675, 0.0394, 0.0404], device='cuda:1'), in_proj_covar=tensor([0.0199, 0.0226, 0.0208, 0.0161, 0.0307, 0.0254, 0.0191, 0.0206], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0004, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 17:29:18,240 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.543e+02 3.220e+02 3.733e+02 4.427e+02 7.256e+02, threshold=7.465e+02, percent-clipped=0.0 2023-04-16 17:29:47,653 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23041.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:29:49,858 INFO [train.py:893] (1/4) Epoch 9, batch 1450, loss[loss=0.2567, simple_loss=0.3018, pruned_loss=0.1058, over 13492.00 frames. ], tot_loss[loss=0.231, simple_loss=0.2761, pruned_loss=0.0929, over 2659089.16 frames. ], batch size: 93, lr: 1.46e-02, grad_scale: 8.0 2023-04-16 17:30:29,750 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23089.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:30:29,917 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.2313, 1.9797, 1.8280, 2.1883, 1.6852, 2.1663, 1.9835, 2.0122], device='cuda:1'), covar=tensor([0.0066, 0.0121, 0.0112, 0.0100, 0.0138, 0.0083, 0.0177, 0.0110], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0068, 0.0076, 0.0065, 0.0082, 0.0060, 0.0069, 0.0064], device='cuda:1'), out_proj_covar=tensor([7.3663e-05, 7.9216e-05, 9.1023e-05, 7.7410e-05, 9.8010e-05, 6.8823e-05, 8.1737e-05, 7.3211e-05], device='cuda:1') 2023-04-16 17:30:33,729 INFO [train.py:893] (1/4) Epoch 9, batch 1500, loss[loss=0.2174, simple_loss=0.2597, pruned_loss=0.08758, over 12792.00 frames. ], tot_loss[loss=0.2302, simple_loss=0.2753, pruned_loss=0.09249, over 2657083.63 frames. ], batch size: 52, lr: 1.46e-02, grad_scale: 8.0 2023-04-16 17:30:47,111 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23108.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:30:47,595 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.350e+02 3.219e+02 3.782e+02 4.835e+02 1.139e+03, threshold=7.564e+02, percent-clipped=5.0 2023-04-16 17:31:20,073 INFO [train.py:893] (1/4) Epoch 9, batch 1550, loss[loss=0.2342, simple_loss=0.2817, pruned_loss=0.09332, over 13563.00 frames. ], tot_loss[loss=0.2287, simple_loss=0.2745, pruned_loss=0.09147, over 2661479.05 frames. ], batch size: 89, lr: 1.45e-02, grad_scale: 8.0 2023-04-16 17:31:21,575 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2023-04-16 17:31:25,192 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23150.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:31:29,387 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8344, 4.2414, 3.8942, 3.9717, 3.9802, 4.4215, 4.2279, 4.1167], device='cuda:1'), covar=tensor([0.0337, 0.0257, 0.0319, 0.1046, 0.0321, 0.0261, 0.0256, 0.0389], device='cuda:1'), in_proj_covar=tensor([0.0114, 0.0118, 0.0127, 0.0227, 0.0128, 0.0143, 0.0126, 0.0097], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-16 17:31:41,569 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1599, 2.7878, 2.3954, 3.9954, 4.6367, 3.5054, 4.5449, 4.2371], device='cuda:1'), covar=tensor([0.0100, 0.0775, 0.0907, 0.0101, 0.0062, 0.0391, 0.0097, 0.0072], device='cuda:1'), in_proj_covar=tensor([0.0056, 0.0076, 0.0082, 0.0061, 0.0045, 0.0066, 0.0041, 0.0054], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:31:42,365 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23169.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:31:43,358 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-04-16 17:31:53,627 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23181.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:32:04,383 INFO [train.py:893] (1/4) Epoch 9, batch 1600, loss[loss=0.2768, simple_loss=0.3169, pruned_loss=0.1183, over 13387.00 frames. ], tot_loss[loss=0.2284, simple_loss=0.2746, pruned_loss=0.09108, over 2667271.74 frames. ], batch size: 113, lr: 1.45e-02, grad_scale: 8.0 2023-04-16 17:32:19,471 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.289e+02 3.375e+02 3.850e+02 4.710e+02 9.191e+02, threshold=7.699e+02, percent-clipped=3.0 2023-04-16 17:32:26,405 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23217.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:32:37,739 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23229.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:32:50,308 INFO [train.py:893] (1/4) Epoch 9, batch 1650, loss[loss=0.2203, simple_loss=0.2679, pruned_loss=0.08634, over 13359.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.276, pruned_loss=0.09084, over 2667586.98 frames. ], batch size: 73, lr: 1.45e-02, grad_scale: 8.0 2023-04-16 17:33:20,723 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23278.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:33:31,087 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23290.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:33:34,769 INFO [train.py:893] (1/4) Epoch 9, batch 1700, loss[loss=0.2041, simple_loss=0.2379, pruned_loss=0.08512, over 7030.00 frames. ], tot_loss[loss=0.228, simple_loss=0.2751, pruned_loss=0.09045, over 2653593.87 frames. ], batch size: 28, lr: 1.45e-02, grad_scale: 8.0 2023-04-16 17:33:35,862 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23295.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:33:48,284 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.390e+02 3.407e+02 3.875e+02 4.640e+02 9.292e+02, threshold=7.749e+02, percent-clipped=4.0 2023-04-16 17:33:51,979 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3238, 4.4670, 3.8250, 2.9841, 3.1602, 2.4173, 4.5844, 2.4889], device='cuda:1'), covar=tensor([0.1155, 0.0243, 0.0512, 0.1439, 0.0512, 0.2702, 0.0141, 0.3156], device='cuda:1'), in_proj_covar=tensor([0.0283, 0.0254, 0.0254, 0.0272, 0.0217, 0.0271, 0.0174, 0.0325], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 17:34:19,235 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23343.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:34:19,885 INFO [train.py:893] (1/4) Epoch 9, batch 1750, loss[loss=0.2063, simple_loss=0.2528, pruned_loss=0.07988, over 13421.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.2741, pruned_loss=0.08981, over 2652669.28 frames. ], batch size: 65, lr: 1.45e-02, grad_scale: 8.0 2023-04-16 17:34:26,071 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.83 vs. limit=5.0 2023-04-16 17:34:41,354 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2023-04-16 17:35:05,499 INFO [train.py:893] (1/4) Epoch 9, batch 1800, loss[loss=0.2503, simple_loss=0.2982, pruned_loss=0.1013, over 13402.00 frames. ], tot_loss[loss=0.226, simple_loss=0.2735, pruned_loss=0.08929, over 2651545.93 frames. ], batch size: 113, lr: 1.45e-02, grad_scale: 8.0 2023-04-16 17:35:18,931 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.368e+02 3.089e+02 3.721e+02 4.545e+02 8.418e+02, threshold=7.442e+02, percent-clipped=1.0 2023-04-16 17:35:42,816 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9205, 4.3954, 4.1729, 4.0432, 4.2039, 3.9683, 4.3641, 4.4240], device='cuda:1'), covar=tensor([0.0267, 0.0362, 0.0269, 0.0424, 0.0317, 0.0375, 0.0455, 0.0318], device='cuda:1'), in_proj_covar=tensor([0.0102, 0.0153, 0.0118, 0.0142, 0.0110, 0.0146, 0.0103, 0.0123], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 17:35:45,388 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6068, 3.9961, 3.7553, 4.3986, 2.1593, 3.1019, 3.8702, 2.1856], device='cuda:1'), covar=tensor([0.0098, 0.0402, 0.0532, 0.0353, 0.1397, 0.0789, 0.0493, 0.1877], device='cuda:1'), in_proj_covar=tensor([0.0089, 0.0162, 0.0176, 0.0176, 0.0172, 0.0183, 0.0164, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 17:35:49,970 INFO [train.py:893] (1/4) Epoch 9, batch 1850, loss[loss=0.2342, simple_loss=0.2762, pruned_loss=0.09608, over 13359.00 frames. ], tot_loss[loss=0.2258, simple_loss=0.2732, pruned_loss=0.08924, over 2655706.65 frames. ], batch size: 109, lr: 1.45e-02, grad_scale: 8.0 2023-04-16 17:35:51,498 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-16 17:35:55,727 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23450.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:36:08,521 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23464.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:36:13,192 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 2023-04-16 17:36:23,955 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23481.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:36:31,097 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=23488.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:36:35,764 INFO [train.py:893] (1/4) Epoch 9, batch 1900, loss[loss=0.2206, simple_loss=0.2515, pruned_loss=0.09487, over 12826.00 frames. ], tot_loss[loss=0.2251, simple_loss=0.272, pruned_loss=0.08907, over 2659808.05 frames. ], batch size: 52, lr: 1.44e-02, grad_scale: 8.0 2023-04-16 17:36:39,374 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23498.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:36:49,681 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.068e+02 3.397e+02 4.025e+02 4.847e+02 8.504e+02, threshold=8.050e+02, percent-clipped=4.0 2023-04-16 17:37:07,612 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23529.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:37:21,169 INFO [train.py:893] (1/4) Epoch 9, batch 1950, loss[loss=0.22, simple_loss=0.2762, pruned_loss=0.0819, over 13525.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.2715, pruned_loss=0.08885, over 2659951.95 frames. ], batch size: 91, lr: 1.44e-02, grad_scale: 8.0 2023-04-16 17:37:26,193 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=23549.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 17:37:47,803 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23573.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:37:50,405 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.5256, 2.2466, 2.0440, 2.5745, 1.7762, 2.7354, 2.6054, 2.3039], device='cuda:1'), covar=tensor([0.0073, 0.0178, 0.0169, 0.0141, 0.0197, 0.0093, 0.0160, 0.0106], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0069, 0.0078, 0.0067, 0.0084, 0.0061, 0.0070, 0.0065], device='cuda:1'), out_proj_covar=tensor([7.3166e-05, 8.1401e-05, 9.3421e-05, 7.8925e-05, 1.0026e-04, 6.8523e-05, 8.2969e-05, 7.4602e-05], device='cuda:1') 2023-04-16 17:37:59,118 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23585.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:38:06,245 INFO [train.py:893] (1/4) Epoch 9, batch 2000, loss[loss=0.2565, simple_loss=0.2985, pruned_loss=0.1072, over 13359.00 frames. ], tot_loss[loss=0.227, simple_loss=0.2738, pruned_loss=0.09012, over 2661808.79 frames. ], batch size: 118, lr: 1.44e-02, grad_scale: 8.0 2023-04-16 17:38:07,561 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1477, 4.3209, 2.6086, 4.1622, 4.0496, 2.2315, 3.5773, 2.5970], device='cuda:1'), covar=tensor([0.0261, 0.0191, 0.1452, 0.0160, 0.0228, 0.1571, 0.0474, 0.1773], device='cuda:1'), in_proj_covar=tensor([0.0132, 0.0132, 0.0170, 0.0115, 0.0113, 0.0156, 0.0148, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 17:38:09,763 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-16 17:38:21,022 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.996e+02 3.137e+02 3.681e+02 4.329e+02 7.461e+02, threshold=7.363e+02, percent-clipped=0.0 2023-04-16 17:38:52,214 INFO [train.py:893] (1/4) Epoch 9, batch 2050, loss[loss=0.2305, simple_loss=0.2676, pruned_loss=0.09668, over 13526.00 frames. ], tot_loss[loss=0.2291, simple_loss=0.2755, pruned_loss=0.09136, over 2665757.93 frames. ], batch size: 72, lr: 1.44e-02, grad_scale: 8.0 2023-04-16 17:39:29,143 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4965, 3.7195, 4.0876, 2.9653, 2.4651, 2.7553, 4.3711, 4.4428], device='cuda:1'), covar=tensor([0.0929, 0.0909, 0.0380, 0.1417, 0.1607, 0.1335, 0.0200, 0.0122], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0213, 0.0165, 0.0202, 0.0200, 0.0164, 0.0154, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:39:35,171 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2023-04-16 17:39:38,269 INFO [train.py:893] (1/4) Epoch 9, batch 2100, loss[loss=0.2376, simple_loss=0.2842, pruned_loss=0.09555, over 13464.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.2752, pruned_loss=0.09124, over 2660762.08 frames. ], batch size: 100, lr: 1.44e-02, grad_scale: 8.0 2023-04-16 17:39:51,107 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.062e+02 3.176e+02 3.858e+02 4.612e+02 8.885e+02, threshold=7.716e+02, percent-clipped=1.0 2023-04-16 17:40:05,977 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8992, 2.4501, 1.9500, 3.7193, 4.3615, 3.2215, 4.3059, 3.9198], device='cuda:1'), covar=tensor([0.0141, 0.0881, 0.1108, 0.0131, 0.0089, 0.0466, 0.0084, 0.0111], device='cuda:1'), in_proj_covar=tensor([0.0058, 0.0076, 0.0085, 0.0063, 0.0047, 0.0068, 0.0042, 0.0056], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:40:23,387 INFO [train.py:893] (1/4) Epoch 9, batch 2150, loss[loss=0.2312, simple_loss=0.2831, pruned_loss=0.08965, over 13194.00 frames. ], tot_loss[loss=0.2288, simple_loss=0.2753, pruned_loss=0.09114, over 2659722.28 frames. ], batch size: 132, lr: 1.44e-02, grad_scale: 8.0 2023-04-16 17:40:36,578 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-04-16 17:40:41,958 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23764.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:41:09,293 INFO [train.py:893] (1/4) Epoch 9, batch 2200, loss[loss=0.2268, simple_loss=0.279, pruned_loss=0.08733, over 13201.00 frames. ], tot_loss[loss=0.227, simple_loss=0.274, pruned_loss=0.09, over 2657513.54 frames. ], batch size: 132, lr: 1.44e-02, grad_scale: 8.0 2023-04-16 17:41:22,564 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.574e+02 3.265e+02 4.064e+02 4.864e+02 9.339e+02, threshold=8.128e+02, percent-clipped=4.0 2023-04-16 17:41:25,119 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23812.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:41:25,359 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9894, 4.1424, 3.5302, 2.7835, 2.9582, 2.3416, 4.2690, 2.4367], device='cuda:1'), covar=tensor([0.1195, 0.0221, 0.0492, 0.1258, 0.0591, 0.2612, 0.0155, 0.3251], device='cuda:1'), in_proj_covar=tensor([0.0282, 0.0250, 0.0257, 0.0273, 0.0217, 0.0272, 0.0176, 0.0323], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 17:41:36,196 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 2023-04-16 17:41:54,113 INFO [train.py:893] (1/4) Epoch 9, batch 2250, loss[loss=0.1961, simple_loss=0.2506, pruned_loss=0.0708, over 13533.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.2721, pruned_loss=0.08884, over 2662161.15 frames. ], batch size: 85, lr: 1.43e-02, grad_scale: 8.0 2023-04-16 17:41:54,286 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=23844.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 17:42:21,207 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23873.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:42:29,317 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.8218, 2.6037, 2.1841, 1.5034, 1.3209, 2.1287, 2.0927, 2.7909], device='cuda:1'), covar=tensor([0.0686, 0.0304, 0.0599, 0.1496, 0.0198, 0.0380, 0.0607, 0.0214], device='cuda:1'), in_proj_covar=tensor([0.0123, 0.0098, 0.0091, 0.0168, 0.0083, 0.0109, 0.0118, 0.0090], device='cuda:1'), out_proj_covar=tensor([9.5036e-05, 7.5215e-05, 7.4731e-05, 1.3406e-04, 6.9146e-05, 8.2343e-05, 9.3074e-05, 6.7356e-05], device='cuda:1') 2023-04-16 17:42:32,550 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=23885.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:42:32,678 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6064, 3.6946, 4.3317, 3.1383, 2.8028, 2.9694, 4.4873, 4.6065], device='cuda:1'), covar=tensor([0.0868, 0.1058, 0.0311, 0.1267, 0.1390, 0.1210, 0.0226, 0.0099], device='cuda:1'), in_proj_covar=tensor([0.0215, 0.0214, 0.0166, 0.0204, 0.0203, 0.0166, 0.0156, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:42:40,219 INFO [train.py:893] (1/4) Epoch 9, batch 2300, loss[loss=0.2258, simple_loss=0.2756, pruned_loss=0.08803, over 13487.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.2717, pruned_loss=0.08847, over 2656243.68 frames. ], batch size: 93, lr: 1.43e-02, grad_scale: 16.0 2023-04-16 17:42:53,341 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.220e+02 3.044e+02 3.754e+02 4.701e+02 8.519e+02, threshold=7.507e+02, percent-clipped=1.0 2023-04-16 17:43:04,667 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23921.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:43:12,366 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.18 vs. limit=5.0 2023-04-16 17:43:15,945 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=23933.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:43:17,774 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1902, 2.6276, 1.9409, 3.9226, 4.6464, 3.5086, 4.5170, 4.2513], device='cuda:1'), covar=tensor([0.0084, 0.0744, 0.1020, 0.0122, 0.0050, 0.0388, 0.0067, 0.0071], device='cuda:1'), in_proj_covar=tensor([0.0058, 0.0077, 0.0086, 0.0063, 0.0047, 0.0069, 0.0042, 0.0057], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:43:25,984 INFO [train.py:893] (1/4) Epoch 9, batch 2350, loss[loss=0.2325, simple_loss=0.2761, pruned_loss=0.09447, over 13436.00 frames. ], tot_loss[loss=0.2236, simple_loss=0.2709, pruned_loss=0.08818, over 2658396.65 frames. ], batch size: 106, lr: 1.43e-02, grad_scale: 16.0 2023-04-16 17:43:38,864 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-04-16 17:43:44,360 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3697, 2.3556, 2.3716, 3.7519, 3.4551, 3.8163, 2.9089, 2.3062], device='cuda:1'), covar=tensor([0.0254, 0.1004, 0.0942, 0.0045, 0.0235, 0.0029, 0.0736, 0.1030], device='cuda:1'), in_proj_covar=tensor([0.0109, 0.0149, 0.0155, 0.0084, 0.0105, 0.0071, 0.0156, 0.0145], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0001, 0.0003, 0.0002], device='cuda:1') 2023-04-16 17:43:45,704 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-16 17:44:02,813 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3446, 4.0518, 4.2512, 2.7336, 4.6992, 4.4404, 4.3590, 4.6241], device='cuda:1'), covar=tensor([0.0214, 0.0105, 0.0136, 0.1066, 0.0114, 0.0204, 0.0141, 0.0081], device='cuda:1'), in_proj_covar=tensor([0.0067, 0.0038, 0.0059, 0.0085, 0.0074, 0.0070, 0.0060, 0.0052], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:44:10,690 INFO [train.py:893] (1/4) Epoch 9, batch 2400, loss[loss=0.2133, simple_loss=0.2652, pruned_loss=0.08071, over 13522.00 frames. ], tot_loss[loss=0.2229, simple_loss=0.2699, pruned_loss=0.08792, over 2659745.46 frames. ], batch size: 76, lr: 1.43e-02, grad_scale: 16.0 2023-04-16 17:44:29,305 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.691e+02 3.069e+02 3.860e+02 4.518e+02 6.769e+02, threshold=7.720e+02, percent-clipped=0.0 2023-04-16 17:44:52,795 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-04-16 17:44:54,825 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24037.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:45:01,077 INFO [train.py:893] (1/4) Epoch 9, batch 2450, loss[loss=0.2417, simple_loss=0.2869, pruned_loss=0.09825, over 13533.00 frames. ], tot_loss[loss=0.2237, simple_loss=0.2706, pruned_loss=0.08837, over 2658094.02 frames. ], batch size: 91, lr: 1.43e-02, grad_scale: 16.0 2023-04-16 17:45:46,320 INFO [train.py:893] (1/4) Epoch 9, batch 2500, loss[loss=0.2339, simple_loss=0.279, pruned_loss=0.09437, over 13333.00 frames. ], tot_loss[loss=0.2235, simple_loss=0.2706, pruned_loss=0.08815, over 2660887.03 frames. ], batch size: 118, lr: 1.43e-02, grad_scale: 16.0 2023-04-16 17:45:49,971 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24098.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:46:00,029 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.936e+02 3.188e+02 3.790e+02 4.783e+02 9.771e+02, threshold=7.581e+02, percent-clipped=3.0 2023-04-16 17:46:01,352 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-16 17:46:31,824 INFO [train.py:893] (1/4) Epoch 9, batch 2550, loss[loss=0.2315, simple_loss=0.2819, pruned_loss=0.09057, over 13416.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.271, pruned_loss=0.0879, over 2660713.10 frames. ], batch size: 95, lr: 1.43e-02, grad_scale: 16.0 2023-04-16 17:46:32,042 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=24144.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 17:46:52,923 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-16 17:47:00,902 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-04-16 17:47:01,516 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4101, 2.8912, 2.3076, 4.1941, 4.8130, 3.5475, 4.6630, 4.4115], device='cuda:1'), covar=tensor([0.0091, 0.0692, 0.0896, 0.0101, 0.0064, 0.0405, 0.0079, 0.0064], device='cuda:1'), in_proj_covar=tensor([0.0058, 0.0078, 0.0086, 0.0064, 0.0047, 0.0069, 0.0042, 0.0057], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:47:10,342 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-04-16 17:47:15,067 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=24192.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:47:16,499 INFO [train.py:893] (1/4) Epoch 9, batch 2600, loss[loss=0.2106, simple_loss=0.2576, pruned_loss=0.08184, over 13235.00 frames. ], tot_loss[loss=0.2239, simple_loss=0.2709, pruned_loss=0.08843, over 2660028.24 frames. ], batch size: 117, lr: 1.42e-02, grad_scale: 16.0 2023-04-16 17:47:30,176 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.381e+02 3.194e+02 3.780e+02 4.738e+02 1.181e+03, threshold=7.561e+02, percent-clipped=2.0 2023-04-16 17:47:58,315 INFO [train.py:893] (1/4) Epoch 9, batch 2650, loss[loss=0.2031, simple_loss=0.2554, pruned_loss=0.07537, over 13518.00 frames. ], tot_loss[loss=0.2249, simple_loss=0.2716, pruned_loss=0.08903, over 2660472.38 frames. ], batch size: 98, lr: 1.42e-02, grad_scale: 16.0 2023-04-16 17:47:59,984 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.6254, 5.1991, 5.0961, 5.1765, 4.7460, 5.0568, 5.6253, 5.0854], device='cuda:1'), covar=tensor([0.0634, 0.0993, 0.1826, 0.2372, 0.0832, 0.1281, 0.0879, 0.0891], device='cuda:1'), in_proj_covar=tensor([0.0251, 0.0283, 0.0357, 0.0370, 0.0207, 0.0279, 0.0327, 0.0252], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 17:48:55,567 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 17:49:04,931 INFO [train.py:893] (1/4) Epoch 10, batch 0, loss[loss=0.1928, simple_loss=0.2525, pruned_loss=0.06654, over 13462.00 frames. ], tot_loss[loss=0.1928, simple_loss=0.2525, pruned_loss=0.06654, over 13462.00 frames. ], batch size: 79, lr: 1.35e-02, grad_scale: 16.0 2023-04-16 17:49:04,931 INFO [train.py:918] (1/4) Computing validation loss 2023-04-16 17:49:27,008 INFO [train.py:927] (1/4) Epoch 10, validation: loss=0.1639, simple_loss=0.2187, pruned_loss=0.05457, over 2446609.00 frames. 2023-04-16 17:49:27,009 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12829MB 2023-04-16 17:49:41,914 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.104e+02 3.192e+02 3.835e+02 4.467e+02 7.040e+02, threshold=7.670e+02, percent-clipped=0.0 2023-04-16 17:49:47,119 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1421, 3.4343, 3.2884, 3.7616, 2.1220, 2.6270, 3.4617, 1.9613], device='cuda:1'), covar=tensor([0.0095, 0.0504, 0.0656, 0.0467, 0.1396, 0.1017, 0.0612, 0.1935], device='cuda:1'), in_proj_covar=tensor([0.0091, 0.0164, 0.0181, 0.0183, 0.0173, 0.0184, 0.0167, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 17:50:12,909 INFO [train.py:893] (1/4) Epoch 10, batch 50, loss[loss=0.1837, simple_loss=0.2357, pruned_loss=0.06582, over 13512.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.2617, pruned_loss=0.08433, over 604069.30 frames. ], batch size: 76, lr: 1.35e-02, grad_scale: 16.0 2023-04-16 17:50:29,813 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-04-16 17:50:31,172 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7024, 1.9185, 3.7277, 3.5933, 3.4897, 2.7979, 3.4151, 2.5055], device='cuda:1'), covar=tensor([0.2391, 0.1658, 0.0138, 0.0144, 0.0265, 0.0724, 0.0229, 0.0991], device='cuda:1'), in_proj_covar=tensor([0.0202, 0.0183, 0.0104, 0.0112, 0.0112, 0.0157, 0.0114, 0.0170], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-16 17:50:36,706 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-16 17:50:36,706 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-16 17:50:36,706 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-16 17:50:36,713 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-16 17:50:37,439 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-16 17:50:37,454 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-16 17:50:37,464 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-16 17:50:41,862 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.8319, 1.9538, 3.8985, 3.7495, 3.7283, 2.8557, 3.6246, 2.4934], device='cuda:1'), covar=tensor([0.2214, 0.1586, 0.0132, 0.0179, 0.0209, 0.0629, 0.0189, 0.0971], device='cuda:1'), in_proj_covar=tensor([0.0201, 0.0182, 0.0103, 0.0111, 0.0111, 0.0156, 0.0113, 0.0169], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0002], device='cuda:1') 2023-04-16 17:50:58,020 INFO [train.py:893] (1/4) Epoch 10, batch 100, loss[loss=0.2096, simple_loss=0.2593, pruned_loss=0.07994, over 13465.00 frames. ], tot_loss[loss=0.2222, simple_loss=0.2666, pruned_loss=0.08886, over 1060345.96 frames. ], batch size: 106, lr: 1.35e-02, grad_scale: 16.0 2023-04-16 17:50:58,223 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24393.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:51:13,610 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.019e+02 3.298e+02 3.891e+02 4.543e+02 7.563e+02, threshold=7.783e+02, percent-clipped=0.0 2023-04-16 17:51:38,257 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.3979, 2.0577, 2.0157, 2.4694, 1.6939, 2.4050, 2.3763, 2.1808], device='cuda:1'), covar=tensor([0.0088, 0.0177, 0.0163, 0.0115, 0.0198, 0.0120, 0.0168, 0.0144], device='cuda:1'), in_proj_covar=tensor([0.0067, 0.0072, 0.0079, 0.0068, 0.0087, 0.0065, 0.0071, 0.0066], device='cuda:1'), out_proj_covar=tensor([7.6082e-05, 8.3408e-05, 9.3628e-05, 8.0237e-05, 1.0294e-04, 7.4233e-05, 8.3364e-05, 7.5450e-05], device='cuda:1') 2023-04-16 17:51:44,493 INFO [train.py:893] (1/4) Epoch 10, batch 150, loss[loss=0.2278, simple_loss=0.2766, pruned_loss=0.08953, over 13532.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.2692, pruned_loss=0.09001, over 1416687.95 frames. ], batch size: 87, lr: 1.35e-02, grad_scale: 16.0 2023-04-16 17:52:30,652 INFO [train.py:893] (1/4) Epoch 10, batch 200, loss[loss=0.2493, simple_loss=0.2877, pruned_loss=0.1054, over 13532.00 frames. ], tot_loss[loss=0.2278, simple_loss=0.2723, pruned_loss=0.09165, over 1685134.72 frames. ], batch size: 91, lr: 1.35e-02, grad_scale: 16.0 2023-04-16 17:52:44,380 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.196e+02 3.067e+02 3.720e+02 4.944e+02 1.446e+03, threshold=7.439e+02, percent-clipped=2.0 2023-04-16 17:53:03,834 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7966, 3.6515, 3.1102, 3.4401, 3.0295, 1.9566, 3.6418, 1.9080], device='cuda:1'), covar=tensor([0.0737, 0.0533, 0.0406, 0.0284, 0.0717, 0.1872, 0.0854, 0.1378], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0107, 0.0107, 0.0095, 0.0130, 0.0160, 0.0118, 0.0126], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:53:16,042 INFO [train.py:893] (1/4) Epoch 10, batch 250, loss[loss=0.2327, simple_loss=0.2809, pruned_loss=0.09222, over 13505.00 frames. ], tot_loss[loss=0.2276, simple_loss=0.2725, pruned_loss=0.09139, over 1899450.71 frames. ], batch size: 98, lr: 1.34e-02, grad_scale: 16.0 2023-04-16 17:53:19,769 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24547.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:53:26,950 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4099, 2.9036, 2.1347, 4.2769, 4.8481, 3.7204, 4.7025, 4.3569], device='cuda:1'), covar=tensor([0.0093, 0.0702, 0.0966, 0.0086, 0.0052, 0.0352, 0.0077, 0.0070], device='cuda:1'), in_proj_covar=tensor([0.0059, 0.0078, 0.0085, 0.0064, 0.0047, 0.0070, 0.0043, 0.0057], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:53:41,235 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24569.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:54:03,217 INFO [train.py:893] (1/4) Epoch 10, batch 300, loss[loss=0.2082, simple_loss=0.264, pruned_loss=0.0762, over 13382.00 frames. ], tot_loss[loss=0.2277, simple_loss=0.273, pruned_loss=0.0912, over 2065389.23 frames. ], batch size: 109, lr: 1.34e-02, grad_scale: 16.0 2023-04-16 17:54:17,102 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24608.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:54:17,540 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.057e+02 2.986e+02 3.516e+02 4.571e+02 9.026e+02, threshold=7.032e+02, percent-clipped=2.0 2023-04-16 17:54:27,320 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.5567, 3.4167, 2.9355, 3.1523, 2.8592, 1.8566, 3.4591, 1.6415], device='cuda:1'), covar=tensor([0.0880, 0.0619, 0.0406, 0.0486, 0.0920, 0.2268, 0.0907, 0.1778], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0107, 0.0107, 0.0095, 0.0131, 0.0161, 0.0118, 0.0125], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 17:54:38,199 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24630.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:54:40,683 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3578, 3.4601, 3.2118, 3.8852, 1.9619, 2.6655, 3.5805, 2.0784], device='cuda:1'), covar=tensor([0.0108, 0.0575, 0.0749, 0.0409, 0.1670, 0.1026, 0.0659, 0.2019], device='cuda:1'), in_proj_covar=tensor([0.0093, 0.0168, 0.0182, 0.0185, 0.0175, 0.0186, 0.0168, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 17:54:48,634 INFO [train.py:893] (1/4) Epoch 10, batch 350, loss[loss=0.1968, simple_loss=0.2554, pruned_loss=0.06912, over 13449.00 frames. ], tot_loss[loss=0.2298, simple_loss=0.2751, pruned_loss=0.09223, over 2199386.67 frames. ], batch size: 79, lr: 1.34e-02, grad_scale: 16.0 2023-04-16 17:55:06,850 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2023-04-16 17:55:10,949 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6075, 3.8915, 3.5188, 4.1951, 2.3699, 2.9095, 3.9045, 2.2286], device='cuda:1'), covar=tensor([0.0085, 0.0467, 0.0741, 0.0434, 0.1474, 0.1008, 0.0598, 0.2053], device='cuda:1'), in_proj_covar=tensor([0.0092, 0.0165, 0.0179, 0.0183, 0.0173, 0.0183, 0.0165, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 17:55:34,164 INFO [train.py:893] (1/4) Epoch 10, batch 400, loss[loss=0.2343, simple_loss=0.2838, pruned_loss=0.09236, over 13271.00 frames. ], tot_loss[loss=0.2293, simple_loss=0.2747, pruned_loss=0.09194, over 2298705.62 frames. ], batch size: 124, lr: 1.34e-02, grad_scale: 16.0 2023-04-16 17:55:34,421 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=24693.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:55:48,031 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.056e+02 3.256e+02 3.949e+02 5.125e+02 8.205e+02, threshold=7.898e+02, percent-clipped=4.0 2023-04-16 17:55:48,497 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8239, 3.6835, 3.2372, 2.6049, 2.6366, 2.1495, 3.7563, 2.1699], device='cuda:1'), covar=tensor([0.1010, 0.0286, 0.0499, 0.1269, 0.0579, 0.2610, 0.0218, 0.2919], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0251, 0.0254, 0.0271, 0.0215, 0.0271, 0.0180, 0.0321], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 17:56:17,962 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=24741.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:56:19,383 INFO [train.py:893] (1/4) Epoch 10, batch 450, loss[loss=0.2396, simple_loss=0.2757, pruned_loss=0.1018, over 13357.00 frames. ], tot_loss[loss=0.228, simple_loss=0.2734, pruned_loss=0.09129, over 2376219.80 frames. ], batch size: 67, lr: 1.34e-02, grad_scale: 16.0 2023-04-16 17:56:39,179 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24765.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:56:43,000 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-16 17:57:05,059 INFO [train.py:893] (1/4) Epoch 10, batch 500, loss[loss=0.2361, simple_loss=0.2869, pruned_loss=0.09268, over 13442.00 frames. ], tot_loss[loss=0.2259, simple_loss=0.2722, pruned_loss=0.08975, over 2443156.65 frames. ], batch size: 95, lr: 1.34e-02, grad_scale: 16.0 2023-04-16 17:57:20,448 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.148e+02 3.204e+02 3.731e+02 4.118e+02 8.079e+02, threshold=7.462e+02, percent-clipped=1.0 2023-04-16 17:57:35,602 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24826.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:57:51,373 INFO [train.py:893] (1/4) Epoch 10, batch 550, loss[loss=0.2336, simple_loss=0.2792, pruned_loss=0.09398, over 13531.00 frames. ], tot_loss[loss=0.2247, simple_loss=0.2716, pruned_loss=0.08889, over 2494450.23 frames. ], batch size: 98, lr: 1.34e-02, grad_scale: 16.0 2023-04-16 17:57:56,412 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24848.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:58:10,429 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9644, 4.2964, 3.9961, 3.9197, 3.9660, 4.3743, 4.2136, 3.9606], device='cuda:1'), covar=tensor([0.0250, 0.0223, 0.0285, 0.1085, 0.0269, 0.0210, 0.0247, 0.0404], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0124, 0.0135, 0.0234, 0.0134, 0.0150, 0.0131, 0.0100], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-16 17:58:11,521 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-04-16 17:58:35,980 INFO [train.py:893] (1/4) Epoch 10, batch 600, loss[loss=0.2114, simple_loss=0.2562, pruned_loss=0.08332, over 13520.00 frames. ], tot_loss[loss=0.2246, simple_loss=0.271, pruned_loss=0.08912, over 2526971.72 frames. ], batch size: 70, lr: 1.34e-02, grad_scale: 16.0 2023-04-16 17:58:45,622 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24903.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:58:50,925 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.078e+02 3.175e+02 3.637e+02 4.549e+02 1.249e+03, threshold=7.275e+02, percent-clipped=2.0 2023-04-16 17:58:51,243 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=24909.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:59:00,206 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3091, 4.5190, 2.9184, 4.4298, 4.3477, 2.5255, 3.7798, 2.8827], device='cuda:1'), covar=tensor([0.0194, 0.0172, 0.1040, 0.0182, 0.0146, 0.1148, 0.0389, 0.1315], device='cuda:1'), in_proj_covar=tensor([0.0134, 0.0136, 0.0173, 0.0120, 0.0114, 0.0155, 0.0148, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 17:59:04,931 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=24925.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 17:59:21,835 INFO [train.py:893] (1/4) Epoch 10, batch 650, loss[loss=0.2295, simple_loss=0.2754, pruned_loss=0.09183, over 13553.00 frames. ], tot_loss[loss=0.2243, simple_loss=0.2708, pruned_loss=0.08888, over 2559027.41 frames. ], batch size: 76, lr: 1.33e-02, grad_scale: 16.0 2023-04-16 17:59:34,396 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7425, 2.2699, 2.0345, 1.3986, 1.2539, 2.0674, 1.9876, 2.6072], device='cuda:1'), covar=tensor([0.0753, 0.0339, 0.0725, 0.1713, 0.0190, 0.0376, 0.0611, 0.0295], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0102, 0.0092, 0.0167, 0.0087, 0.0113, 0.0119, 0.0097], device='cuda:1'), out_proj_covar=tensor([9.5357e-05, 7.7862e-05, 7.5275e-05, 1.3265e-04, 7.1869e-05, 8.5301e-05, 9.3508e-05, 7.1973e-05], device='cuda:1') 2023-04-16 17:59:36,816 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7987, 2.7072, 3.1392, 4.0994, 3.8156, 4.1316, 3.4531, 2.6057], device='cuda:1'), covar=tensor([0.0211, 0.0888, 0.0744, 0.0055, 0.0196, 0.0034, 0.0638, 0.0987], device='cuda:1'), in_proj_covar=tensor([0.0105, 0.0149, 0.0151, 0.0083, 0.0104, 0.0073, 0.0153, 0.0144], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0001, 0.0003, 0.0002], device='cuda:1') 2023-04-16 17:59:57,908 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-16 18:00:06,629 INFO [train.py:893] (1/4) Epoch 10, batch 700, loss[loss=0.2353, simple_loss=0.2752, pruned_loss=0.09768, over 13536.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.2695, pruned_loss=0.08735, over 2584374.50 frames. ], batch size: 98, lr: 1.33e-02, grad_scale: 16.0 2023-04-16 18:00:09,193 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=24995.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:00:22,693 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.375e+02 3.121e+02 3.580e+02 4.230e+02 8.275e+02, threshold=7.159e+02, percent-clipped=1.0 2023-04-16 18:00:53,438 INFO [train.py:893] (1/4) Epoch 10, batch 750, loss[loss=0.209, simple_loss=0.2632, pruned_loss=0.07738, over 13462.00 frames. ], tot_loss[loss=0.2233, simple_loss=0.2702, pruned_loss=0.08821, over 2601933.57 frames. ], batch size: 100, lr: 1.33e-02, grad_scale: 16.0 2023-04-16 18:01:04,625 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25056.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 18:01:39,401 INFO [train.py:893] (1/4) Epoch 10, batch 800, loss[loss=0.2291, simple_loss=0.2799, pruned_loss=0.08913, over 13570.00 frames. ], tot_loss[loss=0.2252, simple_loss=0.2719, pruned_loss=0.08925, over 2612675.47 frames. ], batch size: 89, lr: 1.33e-02, grad_scale: 16.0 2023-04-16 18:01:53,420 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.389e+02 3.119e+02 3.527e+02 4.411e+02 6.666e+02, threshold=7.054e+02, percent-clipped=0.0 2023-04-16 18:01:53,736 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2903, 2.6329, 2.3350, 4.2380, 4.8010, 3.6241, 4.6488, 4.4023], device='cuda:1'), covar=tensor([0.0112, 0.0724, 0.0869, 0.0087, 0.0045, 0.0359, 0.0072, 0.0060], device='cuda:1'), in_proj_covar=tensor([0.0059, 0.0077, 0.0084, 0.0063, 0.0047, 0.0069, 0.0042, 0.0057], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 18:02:04,156 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25121.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:02:24,251 INFO [train.py:893] (1/4) Epoch 10, batch 850, loss[loss=0.2303, simple_loss=0.2752, pruned_loss=0.09272, over 13453.00 frames. ], tot_loss[loss=0.226, simple_loss=0.2729, pruned_loss=0.08958, over 2625435.97 frames. ], batch size: 79, lr: 1.33e-02, grad_scale: 16.0 2023-04-16 18:02:38,147 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4026, 4.6104, 3.1923, 4.6069, 4.4412, 2.8494, 4.0570, 2.9292], device='cuda:1'), covar=tensor([0.0248, 0.0271, 0.1101, 0.0207, 0.0172, 0.1065, 0.0416, 0.1412], device='cuda:1'), in_proj_covar=tensor([0.0133, 0.0135, 0.0172, 0.0120, 0.0114, 0.0152, 0.0147, 0.0158], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 18:03:10,129 INFO [train.py:893] (1/4) Epoch 10, batch 900, loss[loss=0.2438, simple_loss=0.2796, pruned_loss=0.104, over 11954.00 frames. ], tot_loss[loss=0.2269, simple_loss=0.2731, pruned_loss=0.09036, over 2625769.40 frames. ], batch size: 157, lr: 1.33e-02, grad_scale: 16.0 2023-04-16 18:03:19,474 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25203.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:03:20,184 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25204.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:03:24,266 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.167e+02 3.308e+02 3.841e+02 4.546e+02 7.372e+02, threshold=7.682e+02, percent-clipped=2.0 2023-04-16 18:03:25,593 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-16 18:03:39,623 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-16 18:03:39,851 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25225.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:03:55,700 INFO [train.py:893] (1/4) Epoch 10, batch 950, loss[loss=0.2282, simple_loss=0.269, pruned_loss=0.09373, over 13530.00 frames. ], tot_loss[loss=0.2271, simple_loss=0.2726, pruned_loss=0.09079, over 2633579.28 frames. ], batch size: 83, lr: 1.33e-02, grad_scale: 16.0 2023-04-16 18:04:03,682 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25251.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:04:22,606 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25273.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:04:40,936 INFO [train.py:893] (1/4) Epoch 10, batch 1000, loss[loss=0.1981, simple_loss=0.2423, pruned_loss=0.07692, over 13231.00 frames. ], tot_loss[loss=0.2248, simple_loss=0.2701, pruned_loss=0.08976, over 2635080.68 frames. ], batch size: 58, lr: 1.33e-02, grad_scale: 16.0 2023-04-16 18:04:46,392 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.68 vs. limit=2.0 2023-04-16 18:04:52,132 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-04-16 18:04:55,572 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.932e+02 3.162e+02 3.915e+02 4.659e+02 6.893e+02, threshold=7.829e+02, percent-clipped=0.0 2023-04-16 18:05:26,386 INFO [train.py:893] (1/4) Epoch 10, batch 1050, loss[loss=0.1965, simple_loss=0.255, pruned_loss=0.06902, over 13469.00 frames. ], tot_loss[loss=0.2212, simple_loss=0.2674, pruned_loss=0.08743, over 2644788.69 frames. ], batch size: 100, lr: 1.32e-02, grad_scale: 16.0 2023-04-16 18:05:33,858 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25351.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 18:05:41,640 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2023-04-16 18:06:11,653 INFO [train.py:893] (1/4) Epoch 10, batch 1100, loss[loss=0.2424, simple_loss=0.2882, pruned_loss=0.09826, over 13446.00 frames. ], tot_loss[loss=0.2203, simple_loss=0.2675, pruned_loss=0.08656, over 2651911.57 frames. ], batch size: 106, lr: 1.32e-02, grad_scale: 16.0 2023-04-16 18:06:26,835 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.017e+02 3.163e+02 3.560e+02 4.411e+02 7.950e+02, threshold=7.120e+02, percent-clipped=1.0 2023-04-16 18:06:38,147 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25421.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:06:44,024 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25428.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 18:06:55,582 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.5728, 3.4855, 2.9656, 3.1825, 2.8288, 1.8154, 3.4995, 1.7586], device='cuda:1'), covar=tensor([0.0679, 0.0479, 0.0381, 0.0372, 0.0671, 0.1913, 0.0705, 0.1495], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0109, 0.0108, 0.0095, 0.0129, 0.0161, 0.0118, 0.0128], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 18:06:58,302 INFO [train.py:893] (1/4) Epoch 10, batch 1150, loss[loss=0.2246, simple_loss=0.2709, pruned_loss=0.08912, over 13435.00 frames. ], tot_loss[loss=0.2193, simple_loss=0.2672, pruned_loss=0.08565, over 2658106.08 frames. ], batch size: 106, lr: 1.32e-02, grad_scale: 16.0 2023-04-16 18:07:08,247 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7434, 2.6129, 2.3724, 1.3440, 1.4684, 2.1542, 1.9702, 2.7697], device='cuda:1'), covar=tensor([0.0918, 0.0288, 0.0581, 0.1886, 0.0465, 0.0445, 0.0870, 0.0253], device='cuda:1'), in_proj_covar=tensor([0.0127, 0.0101, 0.0093, 0.0169, 0.0087, 0.0116, 0.0122, 0.0098], device='cuda:1'), out_proj_covar=tensor([9.7328e-05, 7.6979e-05, 7.6001e-05, 1.3356e-04, 7.2198e-05, 8.8253e-05, 9.5358e-05, 7.2977e-05], device='cuda:1') 2023-04-16 18:07:21,781 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25469.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:07:39,636 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25489.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 18:07:42,563 INFO [train.py:893] (1/4) Epoch 10, batch 1200, loss[loss=0.243, simple_loss=0.2895, pruned_loss=0.09821, over 13417.00 frames. ], tot_loss[loss=0.2196, simple_loss=0.2679, pruned_loss=0.08567, over 2655650.17 frames. ], batch size: 95, lr: 1.32e-02, grad_scale: 16.0 2023-04-16 18:07:53,652 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25504.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:07:57,438 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.109e+02 3.224e+02 3.678e+02 4.471e+02 9.246e+02, threshold=7.356e+02, percent-clipped=2.0 2023-04-16 18:08:09,677 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-16 18:08:22,494 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-16 18:08:28,385 INFO [train.py:893] (1/4) Epoch 10, batch 1250, loss[loss=0.2129, simple_loss=0.2639, pruned_loss=0.08098, over 13549.00 frames. ], tot_loss[loss=0.2213, simple_loss=0.2692, pruned_loss=0.08675, over 2658484.70 frames. ], batch size: 78, lr: 1.32e-02, grad_scale: 16.0 2023-04-16 18:08:36,616 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25552.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:08:51,338 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.70 vs. limit=2.0 2023-04-16 18:09:13,502 INFO [train.py:893] (1/4) Epoch 10, batch 1300, loss[loss=0.217, simple_loss=0.27, pruned_loss=0.08197, over 13357.00 frames. ], tot_loss[loss=0.2221, simple_loss=0.2702, pruned_loss=0.08701, over 2654575.38 frames. ], batch size: 109, lr: 1.32e-02, grad_scale: 16.0 2023-04-16 18:09:28,835 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.392e+02 3.558e+02 4.001e+02 4.769e+02 9.079e+02, threshold=8.002e+02, percent-clipped=3.0 2023-04-16 18:09:59,313 INFO [train.py:893] (1/4) Epoch 10, batch 1350, loss[loss=0.244, simple_loss=0.2895, pruned_loss=0.09928, over 13449.00 frames. ], tot_loss[loss=0.2224, simple_loss=0.2706, pruned_loss=0.08705, over 2656451.71 frames. ], batch size: 100, lr: 1.32e-02, grad_scale: 16.0 2023-04-16 18:10:00,427 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7602, 2.6146, 2.1471, 2.7787, 2.2819, 2.9765, 2.7193, 2.6470], device='cuda:1'), covar=tensor([0.0111, 0.0126, 0.0171, 0.0152, 0.0156, 0.0074, 0.0247, 0.0131], device='cuda:1'), in_proj_covar=tensor([0.0068, 0.0071, 0.0079, 0.0070, 0.0086, 0.0063, 0.0070, 0.0067], device='cuda:1'), out_proj_covar=tensor([7.5133e-05, 8.2163e-05, 9.2915e-05, 8.1040e-05, 1.0167e-04, 7.1123e-05, 8.1314e-05, 7.5975e-05], device='cuda:1') 2023-04-16 18:10:06,139 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=25651.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 18:10:45,422 INFO [train.py:893] (1/4) Epoch 10, batch 1400, loss[loss=0.2244, simple_loss=0.265, pruned_loss=0.09185, over 13453.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.2696, pruned_loss=0.08673, over 2660347.61 frames. ], batch size: 65, lr: 1.32e-02, grad_scale: 16.0 2023-04-16 18:10:50,744 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=25699.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:10:59,444 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.903e+02 3.130e+02 3.640e+02 4.432e+02 7.364e+02, threshold=7.280e+02, percent-clipped=0.0 2023-04-16 18:11:00,778 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.21 vs. limit=5.0 2023-04-16 18:11:08,639 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7937, 1.7539, 4.0586, 3.7568, 3.8544, 3.1384, 3.7394, 2.7548], device='cuda:1'), covar=tensor([0.2382, 0.2130, 0.0079, 0.0251, 0.0173, 0.0628, 0.0164, 0.0952], device='cuda:1'), in_proj_covar=tensor([0.0205, 0.0188, 0.0102, 0.0113, 0.0117, 0.0160, 0.0117, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 18:11:30,586 INFO [train.py:893] (1/4) Epoch 10, batch 1450, loss[loss=0.2343, simple_loss=0.2844, pruned_loss=0.09214, over 13533.00 frames. ], tot_loss[loss=0.2219, simple_loss=0.2701, pruned_loss=0.08684, over 2663461.88 frames. ], batch size: 98, lr: 1.31e-02, grad_scale: 16.0 2023-04-16 18:12:09,688 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=25784.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 18:12:16,779 INFO [train.py:893] (1/4) Epoch 10, batch 1500, loss[loss=0.1933, simple_loss=0.2409, pruned_loss=0.07286, over 13405.00 frames. ], tot_loss[loss=0.2214, simple_loss=0.2697, pruned_loss=0.08651, over 2660571.33 frames. ], batch size: 62, lr: 1.31e-02, grad_scale: 16.0 2023-04-16 18:12:22,969 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2023-04-16 18:12:26,466 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3560, 4.4113, 3.7522, 3.1186, 3.2941, 2.5768, 4.7349, 2.6081], device='cuda:1'), covar=tensor([0.1148, 0.0207, 0.0570, 0.1258, 0.0548, 0.2468, 0.0125, 0.2843], device='cuda:1'), in_proj_covar=tensor([0.0286, 0.0260, 0.0261, 0.0278, 0.0221, 0.0280, 0.0182, 0.0329], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 18:12:31,671 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.085e+02 3.068e+02 3.732e+02 4.562e+02 1.216e+03, threshold=7.464e+02, percent-clipped=1.0 2023-04-16 18:13:02,774 INFO [train.py:893] (1/4) Epoch 10, batch 1550, loss[loss=0.2218, simple_loss=0.2766, pruned_loss=0.08351, over 13378.00 frames. ], tot_loss[loss=0.2216, simple_loss=0.2698, pruned_loss=0.08666, over 2661148.30 frames. ], batch size: 109, lr: 1.31e-02, grad_scale: 16.0 2023-04-16 18:13:16,719 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=25858.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 18:13:19,304 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-04-16 18:13:48,264 INFO [train.py:893] (1/4) Epoch 10, batch 1600, loss[loss=0.2163, simple_loss=0.2676, pruned_loss=0.08254, over 13528.00 frames. ], tot_loss[loss=0.2217, simple_loss=0.2702, pruned_loss=0.08663, over 2660572.30 frames. ], batch size: 76, lr: 1.31e-02, grad_scale: 32.0 2023-04-16 18:13:55,108 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0107, 2.1483, 2.3410, 3.3267, 3.0218, 3.3514, 2.6543, 2.1988], device='cuda:1'), covar=tensor([0.0226, 0.0813, 0.0729, 0.0065, 0.0315, 0.0053, 0.0538, 0.0807], device='cuda:1'), in_proj_covar=tensor([0.0103, 0.0143, 0.0149, 0.0082, 0.0103, 0.0073, 0.0149, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0001, 0.0002, 0.0001, 0.0002, 0.0002], device='cuda:1') 2023-04-16 18:14:04,227 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.197e+02 3.045e+02 3.726e+02 4.313e+02 1.035e+03, threshold=7.453e+02, percent-clipped=2.0 2023-04-16 18:14:13,090 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=25919.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 18:14:23,242 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-04-16 18:14:35,512 INFO [train.py:893] (1/4) Epoch 10, batch 1650, loss[loss=0.2465, simple_loss=0.3066, pruned_loss=0.09321, over 13404.00 frames. ], tot_loss[loss=0.2212, simple_loss=0.2703, pruned_loss=0.08604, over 2658686.83 frames. ], batch size: 113, lr: 1.31e-02, grad_scale: 32.0 2023-04-16 18:15:20,314 INFO [train.py:893] (1/4) Epoch 10, batch 1700, loss[loss=0.1921, simple_loss=0.2505, pruned_loss=0.06682, over 13378.00 frames. ], tot_loss[loss=0.2204, simple_loss=0.2699, pruned_loss=0.08542, over 2651472.96 frames. ], batch size: 73, lr: 1.31e-02, grad_scale: 32.0 2023-04-16 18:15:39,376 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.206e+02 3.199e+02 3.766e+02 4.421e+02 9.906e+02, threshold=7.531e+02, percent-clipped=2.0 2023-04-16 18:15:46,065 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.78 vs. limit=5.0 2023-04-16 18:15:47,286 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4676, 3.9373, 3.6863, 4.1178, 2.1296, 3.1381, 3.8000, 2.1134], device='cuda:1'), covar=tensor([0.0090, 0.0382, 0.0604, 0.0508, 0.1579, 0.0850, 0.0524, 0.1956], device='cuda:1'), in_proj_covar=tensor([0.0094, 0.0161, 0.0183, 0.0186, 0.0173, 0.0182, 0.0166, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 18:16:02,014 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.71 vs. limit=5.0 2023-04-16 18:16:03,368 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8627, 3.9711, 3.3284, 2.7644, 2.6318, 2.3189, 4.1289, 2.2781], device='cuda:1'), covar=tensor([0.1192, 0.0251, 0.0618, 0.1375, 0.0663, 0.2643, 0.0177, 0.3142], device='cuda:1'), in_proj_covar=tensor([0.0282, 0.0255, 0.0258, 0.0274, 0.0218, 0.0276, 0.0179, 0.0325], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 18:16:10,315 INFO [train.py:893] (1/4) Epoch 10, batch 1750, loss[loss=0.2226, simple_loss=0.2732, pruned_loss=0.08603, over 13540.00 frames. ], tot_loss[loss=0.2181, simple_loss=0.2678, pruned_loss=0.08422, over 2656234.58 frames. ], batch size: 85, lr: 1.31e-02, grad_scale: 32.0 2023-04-16 18:16:21,383 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.08 vs. limit=5.0 2023-04-16 18:16:47,961 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26084.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 18:16:55,981 INFO [train.py:893] (1/4) Epoch 10, batch 1800, loss[loss=0.2134, simple_loss=0.2702, pruned_loss=0.07826, over 13340.00 frames. ], tot_loss[loss=0.2169, simple_loss=0.2669, pruned_loss=0.08349, over 2661894.29 frames. ], batch size: 118, lr: 1.31e-02, grad_scale: 32.0 2023-04-16 18:17:04,912 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26102.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:17:10,493 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.103e+02 2.999e+02 3.760e+02 4.430e+02 9.683e+02, threshold=7.520e+02, percent-clipped=2.0 2023-04-16 18:17:28,782 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-04-16 18:17:31,660 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26132.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 18:17:33,328 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8635, 4.3052, 4.0025, 4.0195, 4.0271, 4.4411, 4.2486, 4.1419], device='cuda:1'), covar=tensor([0.0524, 0.0380, 0.0389, 0.1259, 0.0412, 0.0367, 0.0371, 0.0367], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0126, 0.0136, 0.0230, 0.0135, 0.0151, 0.0132, 0.0096], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-16 18:17:37,523 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4981, 4.0638, 3.3624, 3.9119, 3.5080, 2.1548, 4.0337, 2.3513], device='cuda:1'), covar=tensor([0.0476, 0.0321, 0.0429, 0.0145, 0.0481, 0.1659, 0.0766, 0.1204], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0109, 0.0112, 0.0096, 0.0130, 0.0161, 0.0121, 0.0125], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 18:17:41,230 INFO [train.py:893] (1/4) Epoch 10, batch 1850, loss[loss=0.2349, simple_loss=0.2882, pruned_loss=0.0908, over 13462.00 frames. ], tot_loss[loss=0.217, simple_loss=0.2666, pruned_loss=0.08367, over 2661940.56 frames. ], batch size: 100, lr: 1.30e-02, grad_scale: 32.0 2023-04-16 18:17:43,755 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-16 18:17:59,296 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26163.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:18:03,406 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26168.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:18:08,352 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3576, 2.5280, 2.0497, 4.1499, 4.7739, 3.6757, 4.6158, 4.4055], device='cuda:1'), covar=tensor([0.0095, 0.0795, 0.1061, 0.0098, 0.0057, 0.0341, 0.0090, 0.0064], device='cuda:1'), in_proj_covar=tensor([0.0060, 0.0078, 0.0087, 0.0065, 0.0049, 0.0070, 0.0043, 0.0058], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 18:18:25,584 INFO [train.py:893] (1/4) Epoch 10, batch 1900, loss[loss=0.1981, simple_loss=0.2541, pruned_loss=0.07105, over 13349.00 frames. ], tot_loss[loss=0.2165, simple_loss=0.2658, pruned_loss=0.08362, over 2665835.24 frames. ], batch size: 73, lr: 1.30e-02, grad_scale: 32.0 2023-04-16 18:18:39,848 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.726e+02 3.263e+02 3.745e+02 4.273e+02 6.858e+02, threshold=7.490e+02, percent-clipped=0.0 2023-04-16 18:18:44,155 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26214.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 18:18:57,954 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26229.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:19:09,752 INFO [train.py:893] (1/4) Epoch 10, batch 1950, loss[loss=0.1966, simple_loss=0.2485, pruned_loss=0.07236, over 13351.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2655, pruned_loss=0.08352, over 2661892.36 frames. ], batch size: 73, lr: 1.30e-02, grad_scale: 32.0 2023-04-16 18:19:16,535 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26250.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:19:54,767 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.6055, 2.3050, 2.0481, 1.4046, 1.1886, 2.0819, 1.7949, 2.5367], device='cuda:1'), covar=tensor([0.0883, 0.0364, 0.0697, 0.1698, 0.0181, 0.0463, 0.0776, 0.0275], device='cuda:1'), in_proj_covar=tensor([0.0132, 0.0102, 0.0095, 0.0174, 0.0089, 0.0120, 0.0125, 0.0099], device='cuda:1'), out_proj_covar=tensor([1.0123e-04, 7.8736e-05, 7.7247e-05, 1.3762e-04, 7.1890e-05, 9.1745e-05, 9.7966e-05, 7.3828e-05], device='cuda:1') 2023-04-16 18:19:55,250 INFO [train.py:893] (1/4) Epoch 10, batch 2000, loss[loss=0.2134, simple_loss=0.2637, pruned_loss=0.08158, over 13266.00 frames. ], tot_loss[loss=0.218, simple_loss=0.2669, pruned_loss=0.08455, over 2663110.38 frames. ], batch size: 124, lr: 1.30e-02, grad_scale: 32.0 2023-04-16 18:20:01,840 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-16 18:20:09,140 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.941e+02 3.261e+02 3.830e+02 4.478e+02 8.145e+02, threshold=7.659e+02, percent-clipped=2.0 2023-04-16 18:20:11,764 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26311.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:20:40,243 INFO [train.py:893] (1/4) Epoch 10, batch 2050, loss[loss=0.2384, simple_loss=0.2867, pruned_loss=0.09504, over 13542.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.268, pruned_loss=0.08517, over 2661132.39 frames. ], batch size: 87, lr: 1.30e-02, grad_scale: 32.0 2023-04-16 18:21:19,912 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.66 vs. limit=5.0 2023-04-16 18:21:24,479 INFO [train.py:893] (1/4) Epoch 10, batch 2100, loss[loss=0.2029, simple_loss=0.2388, pruned_loss=0.08347, over 12707.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2676, pruned_loss=0.08466, over 2656949.03 frames. ], batch size: 52, lr: 1.30e-02, grad_scale: 32.0 2023-04-16 18:21:40,090 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.897e+02 2.995e+02 3.490e+02 4.198e+02 1.045e+03, threshold=6.980e+02, percent-clipped=1.0 2023-04-16 18:21:44,385 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26414.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:21:49,440 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4858, 3.5366, 4.1807, 2.8571, 2.6617, 2.8214, 4.3102, 4.4534], device='cuda:1'), covar=tensor([0.0971, 0.1296, 0.0319, 0.1534, 0.1573, 0.1388, 0.0173, 0.0139], device='cuda:1'), in_proj_covar=tensor([0.0218, 0.0222, 0.0168, 0.0208, 0.0204, 0.0170, 0.0164, 0.0132], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 18:21:55,933 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.8359, 2.6541, 2.1449, 1.4932, 1.4653, 2.2022, 2.0538, 2.8619], device='cuda:1'), covar=tensor([0.0819, 0.0293, 0.0766, 0.1684, 0.0410, 0.0341, 0.0733, 0.0239], device='cuda:1'), in_proj_covar=tensor([0.0129, 0.0099, 0.0091, 0.0167, 0.0087, 0.0116, 0.0121, 0.0096], device='cuda:1'), out_proj_covar=tensor([9.8982e-05, 7.5830e-05, 7.4329e-05, 1.3162e-04, 7.0715e-05, 8.7801e-05, 9.4996e-05, 7.1749e-05], device='cuda:1') 2023-04-16 18:22:11,095 INFO [train.py:893] (1/4) Epoch 10, batch 2150, loss[loss=0.2168, simple_loss=0.2678, pruned_loss=0.08291, over 13367.00 frames. ], tot_loss[loss=0.2163, simple_loss=0.2662, pruned_loss=0.08321, over 2659011.02 frames. ], batch size: 73, lr: 1.30e-02, grad_scale: 32.0 2023-04-16 18:22:24,408 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26458.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:22:40,849 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26475.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:22:55,981 INFO [train.py:893] (1/4) Epoch 10, batch 2200, loss[loss=0.1806, simple_loss=0.2279, pruned_loss=0.0666, over 13384.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.2654, pruned_loss=0.08253, over 2660885.01 frames. ], batch size: 62, lr: 1.30e-02, grad_scale: 32.0 2023-04-16 18:23:10,979 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.188e+02 3.062e+02 3.589e+02 4.255e+02 9.664e+02, threshold=7.177e+02, percent-clipped=3.0 2023-04-16 18:23:15,429 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26514.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 18:23:18,747 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7346, 2.4633, 2.3156, 1.6074, 1.3510, 2.2872, 2.0289, 2.8235], device='cuda:1'), covar=tensor([0.0876, 0.0383, 0.0550, 0.1512, 0.0324, 0.0335, 0.0685, 0.0230], device='cuda:1'), in_proj_covar=tensor([0.0128, 0.0100, 0.0090, 0.0166, 0.0086, 0.0115, 0.0121, 0.0095], device='cuda:1'), out_proj_covar=tensor([9.7992e-05, 7.6310e-05, 7.3650e-05, 1.3075e-04, 6.9786e-05, 8.7975e-05, 9.4897e-05, 7.1273e-05], device='cuda:1') 2023-04-16 18:23:24,154 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26524.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:23:41,893 INFO [train.py:893] (1/4) Epoch 10, batch 2250, loss[loss=0.1821, simple_loss=0.2313, pruned_loss=0.06642, over 13527.00 frames. ], tot_loss[loss=0.2152, simple_loss=0.2648, pruned_loss=0.08277, over 2654853.27 frames. ], batch size: 70, lr: 1.30e-02, grad_scale: 32.0 2023-04-16 18:23:58,588 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26562.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 18:24:26,369 INFO [train.py:893] (1/4) Epoch 10, batch 2300, loss[loss=0.2309, simple_loss=0.2841, pruned_loss=0.08882, over 13357.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.2641, pruned_loss=0.0826, over 2651978.87 frames. ], batch size: 118, lr: 1.29e-02, grad_scale: 32.0 2023-04-16 18:24:38,806 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26606.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:24:41,108 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.014e+02 2.982e+02 3.568e+02 4.617e+02 1.178e+03, threshold=7.136e+02, percent-clipped=4.0 2023-04-16 18:24:57,453 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.2647, 5.1507, 5.4571, 5.1153, 5.6647, 5.2034, 5.6925, 5.6372], device='cuda:1'), covar=tensor([0.0307, 0.0408, 0.0436, 0.0409, 0.0410, 0.0575, 0.0389, 0.0394], device='cuda:1'), in_proj_covar=tensor([0.0214, 0.0238, 0.0223, 0.0170, 0.0326, 0.0268, 0.0204, 0.0220], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 18:25:11,873 INFO [train.py:893] (1/4) Epoch 10, batch 2350, loss[loss=0.1974, simple_loss=0.2537, pruned_loss=0.07053, over 13477.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.2641, pruned_loss=0.08261, over 2653986.60 frames. ], batch size: 79, lr: 1.29e-02, grad_scale: 16.0 2023-04-16 18:25:36,274 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-16 18:25:57,728 INFO [train.py:893] (1/4) Epoch 10, batch 2400, loss[loss=0.2508, simple_loss=0.2957, pruned_loss=0.103, over 13532.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2644, pruned_loss=0.08304, over 2655406.98 frames. ], batch size: 83, lr: 1.29e-02, grad_scale: 16.0 2023-04-16 18:26:07,132 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2494, 4.5544, 2.9203, 4.4129, 4.3231, 2.6390, 3.9047, 2.9671], device='cuda:1'), covar=tensor([0.0250, 0.0195, 0.1163, 0.0259, 0.0185, 0.1284, 0.0482, 0.1455], device='cuda:1'), in_proj_covar=tensor([0.0135, 0.0138, 0.0172, 0.0126, 0.0116, 0.0154, 0.0145, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 18:26:12,465 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.295e+02 3.122e+02 3.581e+02 4.550e+02 8.173e+02, threshold=7.162e+02, percent-clipped=1.0 2023-04-16 18:26:42,375 INFO [train.py:893] (1/4) Epoch 10, batch 2450, loss[loss=0.2233, simple_loss=0.269, pruned_loss=0.08876, over 13493.00 frames. ], tot_loss[loss=0.2162, simple_loss=0.2654, pruned_loss=0.08354, over 2660770.96 frames. ], batch size: 81, lr: 1.29e-02, grad_scale: 16.0 2023-04-16 18:26:47,482 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7670, 2.1193, 4.2338, 3.8016, 3.9624, 3.1572, 3.6998, 2.8188], device='cuda:1'), covar=tensor([0.2428, 0.1786, 0.0081, 0.0216, 0.0156, 0.0684, 0.0239, 0.0923], device='cuda:1'), in_proj_covar=tensor([0.0204, 0.0188, 0.0102, 0.0113, 0.0118, 0.0161, 0.0123, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 18:26:54,749 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0623, 4.0962, 3.5679, 2.9445, 3.0651, 2.5121, 4.3063, 2.5499], device='cuda:1'), covar=tensor([0.1184, 0.0281, 0.0553, 0.1195, 0.0579, 0.2300, 0.0180, 0.2816], device='cuda:1'), in_proj_covar=tensor([0.0285, 0.0260, 0.0261, 0.0280, 0.0221, 0.0279, 0.0182, 0.0329], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 18:26:57,732 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26758.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:27:07,622 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=26770.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:27:20,590 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.5467, 2.4779, 2.1059, 1.3037, 1.1334, 1.8466, 1.7493, 2.5957], device='cuda:1'), covar=tensor([0.0935, 0.0325, 0.0678, 0.1818, 0.0275, 0.0456, 0.0866, 0.0248], device='cuda:1'), in_proj_covar=tensor([0.0132, 0.0103, 0.0094, 0.0173, 0.0090, 0.0121, 0.0125, 0.0099], device='cuda:1'), out_proj_covar=tensor([1.0111e-04, 7.8791e-05, 7.6544e-05, 1.3632e-04, 7.2100e-05, 9.2533e-05, 9.7920e-05, 7.3457e-05], device='cuda:1') 2023-04-16 18:27:28,693 INFO [train.py:893] (1/4) Epoch 10, batch 2500, loss[loss=0.245, simple_loss=0.2847, pruned_loss=0.1026, over 13277.00 frames. ], tot_loss[loss=0.2155, simple_loss=0.2648, pruned_loss=0.08308, over 2663043.56 frames. ], batch size: 124, lr: 1.29e-02, grad_scale: 16.0 2023-04-16 18:27:40,104 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26806.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:27:43,916 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.938e+02 3.049e+02 3.538e+02 4.460e+02 1.180e+03, threshold=7.075e+02, percent-clipped=3.0 2023-04-16 18:27:57,951 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26824.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:28:14,261 INFO [train.py:893] (1/4) Epoch 10, batch 2550, loss[loss=0.1818, simple_loss=0.2347, pruned_loss=0.0645, over 13448.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2646, pruned_loss=0.08264, over 2666515.97 frames. ], batch size: 65, lr: 1.29e-02, grad_scale: 16.0 2023-04-16 18:28:38,984 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-16 18:28:41,400 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26872.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:29:00,097 INFO [train.py:893] (1/4) Epoch 10, batch 2600, loss[loss=0.2175, simple_loss=0.268, pruned_loss=0.08348, over 13201.00 frames. ], tot_loss[loss=0.2153, simple_loss=0.2648, pruned_loss=0.0829, over 2659577.57 frames. ], batch size: 132, lr: 1.29e-02, grad_scale: 16.0 2023-04-16 18:29:10,073 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26904.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 18:29:11,603 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=26906.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:29:14,773 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.365e+02 3.149e+02 3.762e+02 4.546e+02 1.526e+03, threshold=7.525e+02, percent-clipped=2.0 2023-04-16 18:29:41,120 INFO [train.py:893] (1/4) Epoch 10, batch 2650, loss[loss=0.228, simple_loss=0.2723, pruned_loss=0.09188, over 13540.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2664, pruned_loss=0.08466, over 2658582.14 frames. ], batch size: 83, lr: 1.29e-02, grad_scale: 16.0 2023-04-16 18:29:46,409 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=26950.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:29:49,389 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=26954.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:29:57,761 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=26965.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 18:30:09,241 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.13 vs. limit=5.0 2023-04-16 18:30:09,739 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7818, 2.6319, 3.0139, 4.3263, 3.9782, 4.3264, 3.6628, 2.8572], device='cuda:1'), covar=tensor([0.0255, 0.0910, 0.0654, 0.0034, 0.0152, 0.0036, 0.0510, 0.0722], device='cuda:1'), in_proj_covar=tensor([0.0103, 0.0146, 0.0153, 0.0082, 0.0102, 0.0075, 0.0157, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0001, 0.0003, 0.0002], device='cuda:1') 2023-04-16 18:30:36,756 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 18:30:46,268 INFO [train.py:893] (1/4) Epoch 11, batch 0, loss[loss=0.239, simple_loss=0.271, pruned_loss=0.1035, over 13048.00 frames. ], tot_loss[loss=0.239, simple_loss=0.271, pruned_loss=0.1035, over 13048.00 frames. ], batch size: 142, lr: 1.23e-02, grad_scale: 16.0 2023-04-16 18:30:46,269 INFO [train.py:918] (1/4) Computing validation loss 2023-04-16 18:30:56,249 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3207, 4.6537, 3.1702, 4.5295, 4.3736, 2.8312, 3.9641, 3.0588], device='cuda:1'), covar=tensor([0.0228, 0.0201, 0.0917, 0.0246, 0.0262, 0.1081, 0.0516, 0.1409], device='cuda:1'), in_proj_covar=tensor([0.0136, 0.0138, 0.0171, 0.0127, 0.0116, 0.0152, 0.0145, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 18:31:08,657 INFO [train.py:927] (1/4) Epoch 11, validation: loss=0.1601, simple_loss=0.2155, pruned_loss=0.05235, over 2446609.00 frames. 2023-04-16 18:31:08,657 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12829MB 2023-04-16 18:31:24,695 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.130e+02 3.114e+02 4.055e+02 4.999e+02 1.494e+03, threshold=8.109e+02, percent-clipped=6.0 2023-04-16 18:31:26,624 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27011.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:31:32,369 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5658, 2.4003, 2.8784, 4.1238, 3.8155, 4.1677, 3.5856, 2.5125], device='cuda:1'), covar=tensor([0.0375, 0.1016, 0.0746, 0.0046, 0.0201, 0.0038, 0.0530, 0.1020], device='cuda:1'), in_proj_covar=tensor([0.0104, 0.0147, 0.0153, 0.0083, 0.0103, 0.0076, 0.0157, 0.0141], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0001, 0.0003, 0.0002], device='cuda:1') 2023-04-16 18:31:47,036 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-04-16 18:31:53,718 INFO [train.py:893] (1/4) Epoch 11, batch 50, loss[loss=0.2359, simple_loss=0.2772, pruned_loss=0.09734, over 13529.00 frames. ], tot_loss[loss=0.2109, simple_loss=0.2582, pruned_loss=0.08178, over 601398.91 frames. ], batch size: 85, lr: 1.23e-02, grad_scale: 16.0 2023-04-16 18:32:15,147 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0865, 4.4230, 4.1447, 4.1806, 4.2416, 4.6327, 4.4202, 4.2343], device='cuda:1'), covar=tensor([0.0344, 0.0329, 0.0288, 0.1064, 0.0258, 0.0247, 0.0265, 0.0334], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0124, 0.0134, 0.0228, 0.0134, 0.0149, 0.0132, 0.0096], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-16 18:32:19,821 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-16 18:32:19,822 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-16 18:32:19,822 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-16 18:32:19,829 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-16 18:32:19,845 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-16 18:32:19,871 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-16 18:32:19,882 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-16 18:32:20,018 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27070.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:32:34,307 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7434, 3.7018, 4.3790, 3.0790, 2.4946, 2.8435, 4.5721, 4.7297], device='cuda:1'), covar=tensor([0.0973, 0.1263, 0.0278, 0.1383, 0.1671, 0.1335, 0.0211, 0.0121], device='cuda:1'), in_proj_covar=tensor([0.0218, 0.0224, 0.0168, 0.0205, 0.0202, 0.0168, 0.0166, 0.0131], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 18:32:39,615 INFO [train.py:893] (1/4) Epoch 11, batch 100, loss[loss=0.2186, simple_loss=0.2722, pruned_loss=0.08252, over 13496.00 frames. ], tot_loss[loss=0.2136, simple_loss=0.2604, pruned_loss=0.08343, over 1062456.69 frames. ], batch size: 81, lr: 1.22e-02, grad_scale: 16.0 2023-04-16 18:32:56,092 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.133e+02 2.776e+02 3.223e+02 3.818e+02 7.236e+02, threshold=6.446e+02, percent-clipped=0.0 2023-04-16 18:33:03,054 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27118.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:33:25,356 INFO [train.py:893] (1/4) Epoch 11, batch 150, loss[loss=0.2169, simple_loss=0.2665, pruned_loss=0.08361, over 13439.00 frames. ], tot_loss[loss=0.2182, simple_loss=0.2643, pruned_loss=0.08607, over 1407381.99 frames. ], batch size: 95, lr: 1.22e-02, grad_scale: 16.0 2023-04-16 18:33:30,775 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27148.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:33:37,235 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.1392, 2.2480, 4.3418, 3.7522, 4.1496, 3.1695, 3.9113, 2.9569], device='cuda:1'), covar=tensor([0.1824, 0.1555, 0.0051, 0.0208, 0.0128, 0.0602, 0.0179, 0.0720], device='cuda:1'), in_proj_covar=tensor([0.0204, 0.0187, 0.0103, 0.0112, 0.0118, 0.0162, 0.0119, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 18:34:11,537 INFO [train.py:893] (1/4) Epoch 11, batch 200, loss[loss=0.2432, simple_loss=0.2919, pruned_loss=0.09721, over 13382.00 frames. ], tot_loss[loss=0.2209, simple_loss=0.2675, pruned_loss=0.08717, over 1682321.51 frames. ], batch size: 109, lr: 1.22e-02, grad_scale: 16.0 2023-04-16 18:34:26,395 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.87 vs. limit=5.0 2023-04-16 18:34:27,152 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27209.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:34:27,624 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.283e+02 3.174e+02 3.920e+02 4.819e+02 9.986e+02, threshold=7.840e+02, percent-clipped=7.0 2023-04-16 18:34:45,727 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1292, 4.0577, 3.2818, 3.8117, 3.1480, 2.2475, 3.9548, 2.3477], device='cuda:1'), covar=tensor([0.0593, 0.0339, 0.0396, 0.0244, 0.0744, 0.1730, 0.0924, 0.1291], device='cuda:1'), in_proj_covar=tensor([0.0122, 0.0113, 0.0110, 0.0097, 0.0133, 0.0163, 0.0124, 0.0129], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 18:34:57,058 INFO [train.py:893] (1/4) Epoch 11, batch 250, loss[loss=0.1802, simple_loss=0.2415, pruned_loss=0.05943, over 13527.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2663, pruned_loss=0.08669, over 1888903.22 frames. ], batch size: 72, lr: 1.22e-02, grad_scale: 16.0 2023-04-16 18:35:09,718 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-04-16 18:35:13,580 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27260.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 18:35:41,630 INFO [train.py:893] (1/4) Epoch 11, batch 300, loss[loss=0.2331, simple_loss=0.2873, pruned_loss=0.0895, over 13513.00 frames. ], tot_loss[loss=0.2195, simple_loss=0.2667, pruned_loss=0.08616, over 2063705.85 frames. ], batch size: 91, lr: 1.22e-02, grad_scale: 16.0 2023-04-16 18:35:53,894 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27306.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:35:56,884 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.999e+02 3.048e+02 3.431e+02 4.475e+02 6.700e+02, threshold=6.863e+02, percent-clipped=0.0 2023-04-16 18:36:03,386 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.4757, 3.4742, 2.4599, 3.1936, 2.5597, 1.5934, 3.5291, 1.7068], device='cuda:1'), covar=tensor([0.0682, 0.0382, 0.0574, 0.0310, 0.0767, 0.2070, 0.0562, 0.1520], device='cuda:1'), in_proj_covar=tensor([0.0122, 0.0113, 0.0111, 0.0097, 0.0132, 0.0164, 0.0124, 0.0130], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 18:36:19,784 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-04-16 18:36:26,305 INFO [train.py:893] (1/4) Epoch 11, batch 350, loss[loss=0.1907, simple_loss=0.2335, pruned_loss=0.07401, over 13377.00 frames. ], tot_loss[loss=0.2214, simple_loss=0.2679, pruned_loss=0.08743, over 2189599.00 frames. ], batch size: 62, lr: 1.22e-02, grad_scale: 16.0 2023-04-16 18:37:11,381 INFO [train.py:893] (1/4) Epoch 11, batch 400, loss[loss=0.2317, simple_loss=0.2739, pruned_loss=0.09477, over 13480.00 frames. ], tot_loss[loss=0.2217, simple_loss=0.2685, pruned_loss=0.08738, over 2293957.15 frames. ], batch size: 81, lr: 1.22e-02, grad_scale: 16.0 2023-04-16 18:37:29,239 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.964e+02 3.046e+02 3.470e+02 4.359e+02 1.054e+03, threshold=6.940e+02, percent-clipped=5.0 2023-04-16 18:37:34,339 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7985, 4.2338, 3.9466, 3.9133, 4.0490, 3.8143, 4.2698, 4.2496], device='cuda:1'), covar=tensor([0.0217, 0.0218, 0.0226, 0.0342, 0.0259, 0.0283, 0.0287, 0.0272], device='cuda:1'), in_proj_covar=tensor([0.0109, 0.0168, 0.0127, 0.0150, 0.0120, 0.0161, 0.0109, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 18:37:58,214 INFO [train.py:893] (1/4) Epoch 11, batch 450, loss[loss=0.2207, simple_loss=0.2698, pruned_loss=0.08579, over 13385.00 frames. ], tot_loss[loss=0.2241, simple_loss=0.2713, pruned_loss=0.08847, over 2377537.22 frames. ], batch size: 84, lr: 1.22e-02, grad_scale: 16.0 2023-04-16 18:38:23,446 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-16 18:38:23,905 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-04-16 18:38:42,129 INFO [train.py:893] (1/4) Epoch 11, batch 500, loss[loss=0.28, simple_loss=0.3162, pruned_loss=0.122, over 11724.00 frames. ], tot_loss[loss=0.2234, simple_loss=0.271, pruned_loss=0.0879, over 2439125.57 frames. ], batch size: 157, lr: 1.22e-02, grad_scale: 16.0 2023-04-16 18:38:53,779 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 2023-04-16 18:38:54,945 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=27504.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:38:57,064 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-04-16 18:38:59,643 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.020e+02 3.104e+02 3.811e+02 4.555e+02 9.896e+02, threshold=7.623e+02, percent-clipped=2.0 2023-04-16 18:39:28,704 INFO [train.py:893] (1/4) Epoch 11, batch 550, loss[loss=0.2151, simple_loss=0.2649, pruned_loss=0.08268, over 13505.00 frames. ], tot_loss[loss=0.2223, simple_loss=0.2703, pruned_loss=0.08712, over 2491511.55 frames. ], batch size: 70, lr: 1.21e-02, grad_scale: 16.0 2023-04-16 18:39:44,483 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27560.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 18:40:09,241 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2023-04-16 18:40:09,701 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0352, 3.8448, 4.0583, 2.6462, 4.4399, 4.1325, 4.0914, 4.2809], device='cuda:1'), covar=tensor([0.0250, 0.0131, 0.0125, 0.1024, 0.0149, 0.0220, 0.0175, 0.0111], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0039, 0.0061, 0.0086, 0.0076, 0.0075, 0.0061, 0.0054], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 18:40:13,415 INFO [train.py:893] (1/4) Epoch 11, batch 600, loss[loss=0.2244, simple_loss=0.278, pruned_loss=0.08541, over 13408.00 frames. ], tot_loss[loss=0.2207, simple_loss=0.2686, pruned_loss=0.08642, over 2527626.32 frames. ], batch size: 95, lr: 1.21e-02, grad_scale: 16.0 2023-04-16 18:40:27,679 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27606.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:40:29,316 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27608.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 18:40:30,677 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.019e+02 2.977e+02 3.539e+02 4.534e+02 8.649e+02, threshold=7.078e+02, percent-clipped=2.0 2023-04-16 18:40:59,623 INFO [train.py:893] (1/4) Epoch 11, batch 650, loss[loss=0.2145, simple_loss=0.2655, pruned_loss=0.08176, over 13522.00 frames. ], tot_loss[loss=0.2201, simple_loss=0.2682, pruned_loss=0.086, over 2558265.72 frames. ], batch size: 91, lr: 1.21e-02, grad_scale: 16.0 2023-04-16 18:41:10,239 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27654.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:41:44,839 INFO [train.py:893] (1/4) Epoch 11, batch 700, loss[loss=0.2165, simple_loss=0.2627, pruned_loss=0.08514, over 13034.00 frames. ], tot_loss[loss=0.2175, simple_loss=0.266, pruned_loss=0.08448, over 2578346.01 frames. ], batch size: 142, lr: 1.21e-02, grad_scale: 16.0 2023-04-16 18:42:00,290 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.022e+02 3.180e+02 3.711e+02 4.284e+02 7.518e+02, threshold=7.422e+02, percent-clipped=1.0 2023-04-16 18:42:00,838 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.05 vs. limit=5.0 2023-04-16 18:42:03,783 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=27714.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 18:42:05,608 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.5950, 3.6278, 2.9260, 3.2628, 3.0838, 1.9099, 3.6627, 1.8108], device='cuda:1'), covar=tensor([0.0674, 0.0410, 0.0463, 0.0392, 0.0616, 0.1960, 0.0782, 0.1529], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0109, 0.0109, 0.0095, 0.0129, 0.0161, 0.0122, 0.0125], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 18:42:29,598 INFO [train.py:893] (1/4) Epoch 11, batch 750, loss[loss=0.2533, simple_loss=0.2897, pruned_loss=0.1085, over 13537.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.266, pruned_loss=0.08486, over 2592920.90 frames. ], batch size: 85, lr: 1.21e-02, grad_scale: 16.0 2023-04-16 18:42:49,044 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0544, 2.5233, 2.1310, 3.9785, 4.5165, 3.3902, 4.3647, 4.1936], device='cuda:1'), covar=tensor([0.0094, 0.0789, 0.0998, 0.0098, 0.0067, 0.0482, 0.0087, 0.0070], device='cuda:1'), in_proj_covar=tensor([0.0062, 0.0081, 0.0089, 0.0068, 0.0051, 0.0073, 0.0046, 0.0061], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 18:42:59,907 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=27775.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 18:43:15,442 INFO [train.py:893] (1/4) Epoch 11, batch 800, loss[loss=0.2097, simple_loss=0.2416, pruned_loss=0.08895, over 12341.00 frames. ], tot_loss[loss=0.2185, simple_loss=0.2668, pruned_loss=0.08509, over 2603492.26 frames. ], batch size: 50, lr: 1.21e-02, grad_scale: 16.0 2023-04-16 18:43:26,200 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=27804.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:43:30,862 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.353e+02 3.099e+02 3.629e+02 4.867e+02 1.066e+03, threshold=7.258e+02, percent-clipped=2.0 2023-04-16 18:43:34,883 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4831, 4.3453, 4.6059, 4.5205, 4.8329, 4.3948, 4.8223, 4.8014], device='cuda:1'), covar=tensor([0.0447, 0.0634, 0.0661, 0.0545, 0.0651, 0.0819, 0.0548, 0.0512], device='cuda:1'), in_proj_covar=tensor([0.0226, 0.0250, 0.0235, 0.0180, 0.0343, 0.0283, 0.0213, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 18:43:59,876 INFO [train.py:893] (1/4) Epoch 11, batch 850, loss[loss=0.2198, simple_loss=0.2705, pruned_loss=0.08453, over 13505.00 frames. ], tot_loss[loss=0.2192, simple_loss=0.2675, pruned_loss=0.08542, over 2618262.96 frames. ], batch size: 91, lr: 1.21e-02, grad_scale: 16.0 2023-04-16 18:44:10,238 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=27852.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:44:29,424 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.5841, 3.3062, 2.7068, 2.9812, 2.7469, 1.7463, 3.3364, 1.8371], device='cuda:1'), covar=tensor([0.0583, 0.0563, 0.0362, 0.0419, 0.0650, 0.1916, 0.0806, 0.1322], device='cuda:1'), in_proj_covar=tensor([0.0123, 0.0110, 0.0110, 0.0097, 0.0132, 0.0166, 0.0123, 0.0127], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 18:44:45,323 INFO [train.py:893] (1/4) Epoch 11, batch 900, loss[loss=0.2192, simple_loss=0.2583, pruned_loss=0.09002, over 12015.00 frames. ], tot_loss[loss=0.2198, simple_loss=0.2682, pruned_loss=0.0857, over 2629466.84 frames. ], batch size: 157, lr: 1.21e-02, grad_scale: 16.0 2023-04-16 18:45:01,523 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.325e+02 3.181e+02 3.611e+02 4.274e+02 7.628e+02, threshold=7.222e+02, percent-clipped=1.0 2023-04-16 18:45:15,487 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-16 18:45:30,660 INFO [train.py:893] (1/4) Epoch 11, batch 950, loss[loss=0.2126, simple_loss=0.2652, pruned_loss=0.07999, over 13492.00 frames. ], tot_loss[loss=0.22, simple_loss=0.2679, pruned_loss=0.086, over 2640806.47 frames. ], batch size: 93, lr: 1.21e-02, grad_scale: 16.0 2023-04-16 18:45:35,378 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2023-04-16 18:45:41,667 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6296, 4.1327, 4.1371, 4.2358, 3.8531, 4.0104, 4.6130, 4.0764], device='cuda:1'), covar=tensor([0.0824, 0.1190, 0.2107, 0.2429, 0.1255, 0.1558, 0.0969, 0.1256], device='cuda:1'), in_proj_covar=tensor([0.0283, 0.0315, 0.0392, 0.0410, 0.0232, 0.0304, 0.0359, 0.0280], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 18:46:13,708 INFO [train.py:893] (1/4) Epoch 11, batch 1000, loss[loss=0.1984, simple_loss=0.249, pruned_loss=0.07394, over 13554.00 frames. ], tot_loss[loss=0.2178, simple_loss=0.2657, pruned_loss=0.08491, over 2646710.68 frames. ], batch size: 78, lr: 1.21e-02, grad_scale: 16.0 2023-04-16 18:46:18,849 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.3644, 4.8294, 4.7366, 4.8866, 4.5290, 4.7703, 5.3330, 4.8609], device='cuda:1'), covar=tensor([0.0758, 0.1096, 0.2690, 0.2322, 0.0822, 0.1588, 0.0760, 0.1002], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0313, 0.0393, 0.0407, 0.0231, 0.0303, 0.0357, 0.0278], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 18:46:34,491 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.038e+02 3.069e+02 3.661e+02 4.441e+02 9.390e+02, threshold=7.322e+02, percent-clipped=1.0 2023-04-16 18:46:48,159 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 2023-04-16 18:47:03,890 INFO [train.py:893] (1/4) Epoch 11, batch 1050, loss[loss=0.214, simple_loss=0.2535, pruned_loss=0.08726, over 13368.00 frames. ], tot_loss[loss=0.2147, simple_loss=0.2634, pruned_loss=0.08304, over 2653166.66 frames. ], batch size: 67, lr: 1.20e-02, grad_scale: 16.0 2023-04-16 18:47:29,423 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28070.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 18:47:47,429 INFO [train.py:893] (1/4) Epoch 11, batch 1100, loss[loss=0.2505, simple_loss=0.2888, pruned_loss=0.1061, over 13518.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2641, pruned_loss=0.08279, over 2649483.95 frames. ], batch size: 85, lr: 1.20e-02, grad_scale: 16.0 2023-04-16 18:48:04,937 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.261e+02 3.035e+02 3.453e+02 4.146e+02 7.146e+02, threshold=6.906e+02, percent-clipped=0.0 2023-04-16 18:48:34,321 INFO [train.py:893] (1/4) Epoch 11, batch 1150, loss[loss=0.2092, simple_loss=0.2634, pruned_loss=0.07751, over 13453.00 frames. ], tot_loss[loss=0.2134, simple_loss=0.2636, pruned_loss=0.08163, over 2654083.84 frames. ], batch size: 103, lr: 1.20e-02, grad_scale: 16.0 2023-04-16 18:49:18,350 INFO [train.py:893] (1/4) Epoch 11, batch 1200, loss[loss=0.2298, simple_loss=0.2804, pruned_loss=0.08957, over 13276.00 frames. ], tot_loss[loss=0.2137, simple_loss=0.264, pruned_loss=0.08173, over 2655055.95 frames. ], batch size: 124, lr: 1.20e-02, grad_scale: 16.0 2023-04-16 18:49:34,900 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.076e+02 2.988e+02 3.619e+02 4.401e+02 8.717e+02, threshold=7.238e+02, percent-clipped=6.0 2023-04-16 18:49:44,797 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-16 18:49:53,662 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28229.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:49:57,718 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-16 18:50:04,363 INFO [train.py:893] (1/4) Epoch 11, batch 1250, loss[loss=0.2039, simple_loss=0.2593, pruned_loss=0.07423, over 13522.00 frames. ], tot_loss[loss=0.214, simple_loss=0.2641, pruned_loss=0.08199, over 2655424.08 frames. ], batch size: 98, lr: 1.20e-02, grad_scale: 16.0 2023-04-16 18:50:17,469 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28255.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 18:50:24,714 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9173, 3.5623, 3.8824, 2.1767, 4.1466, 3.8750, 3.9278, 4.0211], device='cuda:1'), covar=tensor([0.0167, 0.0121, 0.0130, 0.1156, 0.0141, 0.0191, 0.0110, 0.0115], device='cuda:1'), in_proj_covar=tensor([0.0073, 0.0041, 0.0064, 0.0091, 0.0080, 0.0079, 0.0064, 0.0057], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 18:50:29,699 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2728, 2.4558, 2.2133, 4.1347, 4.6615, 3.5517, 4.5696, 4.2899], device='cuda:1'), covar=tensor([0.0078, 0.0870, 0.0944, 0.0086, 0.0060, 0.0366, 0.0067, 0.0067], device='cuda:1'), in_proj_covar=tensor([0.0062, 0.0080, 0.0088, 0.0068, 0.0051, 0.0072, 0.0045, 0.0061], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 18:50:45,836 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28287.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:50:48,925 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28290.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:50:50,185 INFO [train.py:893] (1/4) Epoch 11, batch 1300, loss[loss=0.2035, simple_loss=0.2514, pruned_loss=0.07774, over 13414.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2653, pruned_loss=0.0831, over 2655812.27 frames. ], batch size: 62, lr: 1.20e-02, grad_scale: 16.0 2023-04-16 18:51:05,260 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.287e+02 3.361e+02 3.950e+02 4.861e+02 7.192e+02, threshold=7.900e+02, percent-clipped=0.0 2023-04-16 18:51:11,881 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28316.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 18:51:15,920 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.90 vs. limit=5.0 2023-04-16 18:51:34,392 INFO [train.py:893] (1/4) Epoch 11, batch 1350, loss[loss=0.2123, simple_loss=0.2699, pruned_loss=0.07729, over 13365.00 frames. ], tot_loss[loss=0.216, simple_loss=0.2656, pruned_loss=0.08321, over 2654503.03 frames. ], batch size: 73, lr: 1.20e-02, grad_scale: 16.0 2023-04-16 18:51:40,355 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28348.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:51:59,731 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28370.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 18:52:04,380 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-04-16 18:52:20,669 INFO [train.py:893] (1/4) Epoch 11, batch 1400, loss[loss=0.2044, simple_loss=0.2589, pruned_loss=0.07496, over 13499.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2657, pruned_loss=0.08329, over 2657578.29 frames. ], batch size: 81, lr: 1.20e-02, grad_scale: 16.0 2023-04-16 18:52:26,130 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-04-16 18:52:36,463 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.913e+02 3.138e+02 3.548e+02 4.144e+02 7.337e+02, threshold=7.097e+02, percent-clipped=0.0 2023-04-16 18:52:39,858 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.0820, 4.8320, 5.1124, 4.8456, 5.3238, 4.8236, 5.4235, 5.3577], device='cuda:1'), covar=tensor([0.0298, 0.0527, 0.0495, 0.0527, 0.0463, 0.0686, 0.0329, 0.0376], device='cuda:1'), in_proj_covar=tensor([0.0217, 0.0240, 0.0231, 0.0175, 0.0335, 0.0277, 0.0204, 0.0231], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 18:52:40,012 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.2282, 4.2052, 3.5347, 2.9427, 3.1592, 2.6097, 4.4554, 2.5456], device='cuda:1'), covar=tensor([0.1301, 0.0283, 0.0675, 0.1536, 0.0665, 0.2747, 0.0171, 0.3349], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0270, 0.0267, 0.0282, 0.0226, 0.0289, 0.0186, 0.0332], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 18:52:45,465 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28418.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 18:53:06,670 INFO [train.py:893] (1/4) Epoch 11, batch 1450, loss[loss=0.2136, simple_loss=0.2673, pruned_loss=0.07997, over 13542.00 frames. ], tot_loss[loss=0.2158, simple_loss=0.2652, pruned_loss=0.08316, over 2659290.16 frames. ], batch size: 91, lr: 1.20e-02, grad_scale: 16.0 2023-04-16 18:53:19,606 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8341, 3.7946, 3.2541, 2.6278, 2.8113, 2.2414, 3.9453, 2.1736], device='cuda:1'), covar=tensor([0.1251, 0.0286, 0.0670, 0.1566, 0.0690, 0.2810, 0.0209, 0.3358], device='cuda:1'), in_proj_covar=tensor([0.0292, 0.0269, 0.0267, 0.0281, 0.0225, 0.0288, 0.0185, 0.0331], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 18:53:19,802 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-04-16 18:53:47,220 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28486.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:53:51,730 INFO [train.py:893] (1/4) Epoch 11, batch 1500, loss[loss=0.2001, simple_loss=0.2505, pruned_loss=0.07485, over 13544.00 frames. ], tot_loss[loss=0.2161, simple_loss=0.2657, pruned_loss=0.08324, over 2660178.71 frames. ], batch size: 72, lr: 1.19e-02, grad_scale: 16.0 2023-04-16 18:53:52,027 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28492.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:54:09,339 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.247e+02 3.197e+02 3.873e+02 4.387e+02 6.380e+02, threshold=7.745e+02, percent-clipped=0.0 2023-04-16 18:54:38,279 INFO [train.py:893] (1/4) Epoch 11, batch 1550, loss[loss=0.2154, simple_loss=0.2731, pruned_loss=0.07889, over 13218.00 frames. ], tot_loss[loss=0.2156, simple_loss=0.2655, pruned_loss=0.08284, over 2664635.54 frames. ], batch size: 132, lr: 1.19e-02, grad_scale: 16.0 2023-04-16 18:54:42,600 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28547.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:54:47,571 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28553.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:54:56,704 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5952, 3.9726, 3.7551, 4.3375, 2.2803, 3.2895, 4.0257, 2.1482], device='cuda:1'), covar=tensor([0.0091, 0.0333, 0.0625, 0.0377, 0.1471, 0.0761, 0.0505, 0.1936], device='cuda:1'), in_proj_covar=tensor([0.0098, 0.0167, 0.0190, 0.0194, 0.0177, 0.0187, 0.0169, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 18:55:12,874 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1018, 4.2861, 2.5889, 4.0875, 4.0691, 2.2629, 3.7297, 2.6626], device='cuda:1'), covar=tensor([0.0236, 0.0263, 0.1482, 0.0369, 0.0237, 0.1606, 0.0461, 0.1839], device='cuda:1'), in_proj_covar=tensor([0.0137, 0.0141, 0.0172, 0.0135, 0.0117, 0.0155, 0.0147, 0.0159], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 18:55:15,932 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28585.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:55:17,130 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 2023-04-16 18:55:18,484 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.8314, 2.5866, 2.2810, 1.5540, 1.4376, 2.2434, 2.3012, 2.8334], device='cuda:1'), covar=tensor([0.0737, 0.0239, 0.0735, 0.1550, 0.0332, 0.0347, 0.0625, 0.0253], device='cuda:1'), in_proj_covar=tensor([0.0130, 0.0103, 0.0097, 0.0174, 0.0091, 0.0120, 0.0128, 0.0100], device='cuda:1'), out_proj_covar=tensor([9.9547e-05, 7.8329e-05, 7.8154e-05, 1.3574e-04, 7.2753e-05, 9.1109e-05, 9.9420e-05, 7.4657e-05], device='cuda:1') 2023-04-16 18:55:21,455 INFO [train.py:893] (1/4) Epoch 11, batch 1600, loss[loss=0.2041, simple_loss=0.2639, pruned_loss=0.07212, over 13374.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.265, pruned_loss=0.08241, over 2662570.03 frames. ], batch size: 113, lr: 1.19e-02, grad_scale: 16.0 2023-04-16 18:55:39,665 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.973e+02 3.051e+02 3.528e+02 4.292e+02 7.736e+02, threshold=7.056e+02, percent-clipped=0.0 2023-04-16 18:55:40,657 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28611.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 18:56:08,801 INFO [train.py:893] (1/4) Epoch 11, batch 1650, loss[loss=0.1914, simple_loss=0.2487, pruned_loss=0.06709, over 13379.00 frames. ], tot_loss[loss=0.2148, simple_loss=0.2657, pruned_loss=0.08196, over 2659753.46 frames. ], batch size: 77, lr: 1.19e-02, grad_scale: 32.0 2023-04-16 18:56:09,918 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28643.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:56:30,356 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7246, 4.1314, 2.3712, 3.9355, 3.9270, 2.3226, 3.3208, 2.5691], device='cuda:1'), covar=tensor([0.0324, 0.0270, 0.1513, 0.0376, 0.0253, 0.1323, 0.0602, 0.1512], device='cuda:1'), in_proj_covar=tensor([0.0138, 0.0141, 0.0172, 0.0135, 0.0118, 0.0155, 0.0147, 0.0159], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 18:56:54,198 INFO [train.py:893] (1/4) Epoch 11, batch 1700, loss[loss=0.1891, simple_loss=0.2465, pruned_loss=0.06581, over 13501.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2656, pruned_loss=0.08152, over 2663990.41 frames. ], batch size: 81, lr: 1.19e-02, grad_scale: 32.0 2023-04-16 18:57:10,236 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.256e+02 2.920e+02 3.491e+02 4.240e+02 8.801e+02, threshold=6.981e+02, percent-clipped=2.0 2023-04-16 18:57:35,126 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.47 vs. limit=5.0 2023-04-16 18:57:39,414 INFO [train.py:893] (1/4) Epoch 11, batch 1750, loss[loss=0.2283, simple_loss=0.2765, pruned_loss=0.09011, over 13427.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2644, pruned_loss=0.08095, over 2668639.23 frames. ], batch size: 106, lr: 1.19e-02, grad_scale: 32.0 2023-04-16 18:58:25,770 INFO [train.py:893] (1/4) Epoch 11, batch 1800, loss[loss=0.2011, simple_loss=0.2574, pruned_loss=0.07242, over 13447.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2643, pruned_loss=0.08107, over 2661833.02 frames. ], batch size: 103, lr: 1.19e-02, grad_scale: 32.0 2023-04-16 18:58:40,788 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.909e+02 3.056e+02 3.632e+02 4.305e+02 5.743e+02, threshold=7.265e+02, percent-clipped=0.0 2023-04-16 18:58:57,995 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.06 vs. limit=5.0 2023-04-16 18:59:09,927 INFO [train.py:893] (1/4) Epoch 11, batch 1850, loss[loss=0.2079, simple_loss=0.2563, pruned_loss=0.07981, over 13274.00 frames. ], tot_loss[loss=0.2128, simple_loss=0.2637, pruned_loss=0.08098, over 2662561.71 frames. ], batch size: 124, lr: 1.19e-02, grad_scale: 32.0 2023-04-16 18:59:10,137 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28842.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:59:13,900 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-16 18:59:16,273 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=28848.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:59:28,544 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=28861.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:59:50,148 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28885.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 18:59:50,391 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-04-16 18:59:56,440 INFO [train.py:893] (1/4) Epoch 11, batch 1900, loss[loss=0.2045, simple_loss=0.2571, pruned_loss=0.07591, over 13477.00 frames. ], tot_loss[loss=0.2122, simple_loss=0.2631, pruned_loss=0.08068, over 2664045.64 frames. ], batch size: 79, lr: 1.19e-02, grad_scale: 32.0 2023-04-16 18:59:58,554 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.79 vs. limit=2.0 2023-04-16 19:00:02,433 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6556, 2.2328, 2.2212, 2.7387, 2.0876, 2.6670, 2.7453, 2.3994], device='cuda:1'), covar=tensor([0.0088, 0.0199, 0.0200, 0.0109, 0.0189, 0.0123, 0.0159, 0.0172], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0079, 0.0086, 0.0077, 0.0092, 0.0070, 0.0073, 0.0074], device='cuda:1'), out_proj_covar=tensor([7.8414e-05, 9.0188e-05, 9.9650e-05, 8.7782e-05, 1.0574e-04, 7.8244e-05, 8.3192e-05, 8.2767e-05], device='cuda:1') 2023-04-16 19:00:12,395 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.034e+02 3.088e+02 3.593e+02 4.419e+02 8.610e+02, threshold=7.186e+02, percent-clipped=4.0 2023-04-16 19:00:13,554 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28911.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 19:00:24,095 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=28922.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:00:33,230 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28933.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:00:41,274 INFO [train.py:893] (1/4) Epoch 11, batch 1950, loss[loss=0.2237, simple_loss=0.2751, pruned_loss=0.08609, over 13410.00 frames. ], tot_loss[loss=0.2114, simple_loss=0.2621, pruned_loss=0.0804, over 2662092.54 frames. ], batch size: 113, lr: 1.19e-02, grad_scale: 32.0 2023-04-16 19:00:42,402 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=28943.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:00:56,907 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28959.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 19:00:58,746 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-16 19:01:25,269 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=28991.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:01:25,901 INFO [train.py:893] (1/4) Epoch 11, batch 2000, loss[loss=0.2121, simple_loss=0.2725, pruned_loss=0.07586, over 13404.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2645, pruned_loss=0.0816, over 2662970.75 frames. ], batch size: 113, lr: 1.18e-02, grad_scale: 32.0 2023-04-16 19:01:32,622 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-16 19:01:35,547 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29002.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:01:43,241 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.955e+02 3.039e+02 3.558e+02 4.462e+02 6.294e+02, threshold=7.116e+02, percent-clipped=0.0 2023-04-16 19:01:53,767 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-04-16 19:02:09,697 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-04-16 19:02:12,482 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7555, 3.8684, 2.6722, 3.7319, 3.7424, 2.2020, 3.4192, 2.4574], device='cuda:1'), covar=tensor([0.0231, 0.0233, 0.1023, 0.0290, 0.0216, 0.1210, 0.0440, 0.1309], device='cuda:1'), in_proj_covar=tensor([0.0139, 0.0142, 0.0173, 0.0138, 0.0120, 0.0156, 0.0148, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 19:02:12,974 INFO [train.py:893] (1/4) Epoch 11, batch 2050, loss[loss=0.2403, simple_loss=0.2873, pruned_loss=0.09666, over 13243.00 frames. ], tot_loss[loss=0.2157, simple_loss=0.266, pruned_loss=0.0827, over 2658831.04 frames. ], batch size: 132, lr: 1.18e-02, grad_scale: 32.0 2023-04-16 19:02:28,223 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-04-16 19:02:30,620 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29063.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:02:49,952 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5893, 4.0503, 3.6654, 4.3771, 2.2691, 3.1146, 4.0608, 2.3662], device='cuda:1'), covar=tensor([0.0132, 0.0428, 0.0765, 0.0433, 0.1674, 0.1020, 0.0520, 0.1897], device='cuda:1'), in_proj_covar=tensor([0.0100, 0.0167, 0.0191, 0.0198, 0.0179, 0.0188, 0.0171, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 19:02:57,177 INFO [train.py:893] (1/4) Epoch 11, batch 2100, loss[loss=0.201, simple_loss=0.2639, pruned_loss=0.06903, over 13448.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2645, pruned_loss=0.08187, over 2659290.67 frames. ], batch size: 106, lr: 1.18e-02, grad_scale: 32.0 2023-04-16 19:03:14,862 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.887e+02 3.135e+02 3.626e+02 4.199e+02 7.888e+02, threshold=7.251e+02, percent-clipped=1.0 2023-04-16 19:03:43,824 INFO [train.py:893] (1/4) Epoch 11, batch 2150, loss[loss=0.2501, simple_loss=0.2923, pruned_loss=0.1039, over 11818.00 frames. ], tot_loss[loss=0.2135, simple_loss=0.2645, pruned_loss=0.08131, over 2658140.60 frames. ], batch size: 157, lr: 1.18e-02, grad_scale: 32.0 2023-04-16 19:03:44,047 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29142.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:03:48,947 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29148.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:03:58,185 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3578, 3.4294, 3.9847, 2.8664, 2.5522, 2.7166, 4.1383, 4.2982], device='cuda:1'), covar=tensor([0.1041, 0.1340, 0.0380, 0.1481, 0.1494, 0.1370, 0.0264, 0.0158], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0231, 0.0173, 0.0209, 0.0204, 0.0168, 0.0171, 0.0135], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 19:04:06,046 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3008, 2.7618, 2.2284, 4.0752, 4.7070, 3.4179, 4.6528, 4.2947], device='cuda:1'), covar=tensor([0.0092, 0.0851, 0.1051, 0.0112, 0.0062, 0.0479, 0.0065, 0.0075], device='cuda:1'), in_proj_covar=tensor([0.0062, 0.0081, 0.0088, 0.0068, 0.0052, 0.0072, 0.0045, 0.0062], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 19:04:07,655 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6198, 4.4841, 4.6887, 4.6154, 4.9759, 4.4625, 5.0159, 4.9381], device='cuda:1'), covar=tensor([0.0383, 0.0563, 0.0669, 0.0486, 0.0530, 0.0764, 0.0423, 0.0507], device='cuda:1'), in_proj_covar=tensor([0.0227, 0.0243, 0.0241, 0.0183, 0.0351, 0.0284, 0.0211, 0.0239], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 19:04:27,832 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29190.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:04:27,860 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.0742, 4.6183, 4.5588, 4.5224, 4.2696, 4.4499, 5.0546, 4.5584], device='cuda:1'), covar=tensor([0.0906, 0.1204, 0.2551, 0.3183, 0.0920, 0.1544, 0.1038, 0.1098], device='cuda:1'), in_proj_covar=tensor([0.0275, 0.0308, 0.0391, 0.0404, 0.0227, 0.0296, 0.0354, 0.0272], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 19:04:29,357 INFO [train.py:893] (1/4) Epoch 11, batch 2200, loss[loss=0.208, simple_loss=0.2597, pruned_loss=0.07814, over 13469.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2623, pruned_loss=0.08001, over 2653725.24 frames. ], batch size: 103, lr: 1.18e-02, grad_scale: 32.0 2023-04-16 19:04:34,228 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29196.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:04:45,925 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.987e+02 3.146e+02 3.718e+02 4.404e+02 7.530e+02, threshold=7.436e+02, percent-clipped=2.0 2023-04-16 19:04:51,870 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29217.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:05:10,415 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29236.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:05:15,027 INFO [train.py:893] (1/4) Epoch 11, batch 2250, loss[loss=0.1792, simple_loss=0.2251, pruned_loss=0.06664, over 13333.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.261, pruned_loss=0.07996, over 2651238.64 frames. ], batch size: 62, lr: 1.18e-02, grad_scale: 32.0 2023-04-16 19:05:47,407 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2554, 4.0130, 4.1561, 2.6390, 4.6243, 4.2868, 4.2131, 4.5907], device='cuda:1'), covar=tensor([0.0206, 0.0106, 0.0139, 0.1005, 0.0114, 0.0218, 0.0153, 0.0079], device='cuda:1'), in_proj_covar=tensor([0.0071, 0.0041, 0.0064, 0.0089, 0.0080, 0.0079, 0.0064, 0.0056], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 19:06:00,790 INFO [train.py:893] (1/4) Epoch 11, batch 2300, loss[loss=0.1975, simple_loss=0.2509, pruned_loss=0.07203, over 13408.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2602, pruned_loss=0.07946, over 2653321.05 frames. ], batch size: 62, lr: 1.18e-02, grad_scale: 32.0 2023-04-16 19:06:01,123 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9718, 2.4826, 2.4301, 2.9863, 2.3830, 3.0232, 2.8373, 2.8054], device='cuda:1'), covar=tensor([0.0073, 0.0148, 0.0137, 0.0108, 0.0140, 0.0095, 0.0169, 0.0112], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0079, 0.0086, 0.0077, 0.0090, 0.0070, 0.0074, 0.0074], device='cuda:1'), out_proj_covar=tensor([7.8791e-05, 8.8940e-05, 9.9082e-05, 8.8514e-05, 1.0337e-04, 7.7559e-05, 8.3536e-05, 8.3048e-05], device='cuda:1') 2023-04-16 19:06:06,060 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29297.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:06:16,522 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.915e+02 2.601e+02 3.338e+02 4.081e+02 5.584e+02, threshold=6.676e+02, percent-clipped=0.0 2023-04-16 19:06:36,202 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-04-16 19:06:45,619 INFO [train.py:893] (1/4) Epoch 11, batch 2350, loss[loss=0.2241, simple_loss=0.2674, pruned_loss=0.09046, over 13428.00 frames. ], tot_loss[loss=0.2089, simple_loss=0.2595, pruned_loss=0.07911, over 2656860.09 frames. ], batch size: 95, lr: 1.18e-02, grad_scale: 32.0 2023-04-16 19:07:01,854 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29358.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:07:09,930 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-16 19:07:32,340 INFO [train.py:893] (1/4) Epoch 11, batch 2400, loss[loss=0.2232, simple_loss=0.2795, pruned_loss=0.08347, over 13435.00 frames. ], tot_loss[loss=0.2082, simple_loss=0.2589, pruned_loss=0.07876, over 2654029.38 frames. ], batch size: 95, lr: 1.18e-02, grad_scale: 32.0 2023-04-16 19:07:48,291 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.129e+02 3.051e+02 3.687e+02 4.571e+02 9.088e+02, threshold=7.373e+02, percent-clipped=3.0 2023-04-16 19:08:17,321 INFO [train.py:893] (1/4) Epoch 11, batch 2450, loss[loss=0.2517, simple_loss=0.2843, pruned_loss=0.1095, over 13339.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.2596, pruned_loss=0.07945, over 2647510.14 frames. ], batch size: 118, lr: 1.18e-02, grad_scale: 32.0 2023-04-16 19:08:46,798 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5831, 2.5864, 3.1639, 4.1662, 3.7246, 4.1214, 3.4609, 2.4882], device='cuda:1'), covar=tensor([0.0329, 0.1067, 0.0655, 0.0056, 0.0251, 0.0048, 0.0498, 0.0927], device='cuda:1'), in_proj_covar=tensor([0.0109, 0.0154, 0.0158, 0.0084, 0.0109, 0.0081, 0.0160, 0.0144], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0001, 0.0003, 0.0002], device='cuda:1') 2023-04-16 19:09:03,807 INFO [train.py:893] (1/4) Epoch 11, batch 2500, loss[loss=0.2056, simple_loss=0.2515, pruned_loss=0.07989, over 13380.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.2596, pruned_loss=0.07945, over 2650872.99 frames. ], batch size: 73, lr: 1.17e-02, grad_scale: 32.0 2023-04-16 19:09:20,207 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.679e+02 3.058e+02 3.500e+02 4.137e+02 9.490e+02, threshold=7.000e+02, percent-clipped=2.0 2023-04-16 19:09:26,763 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29517.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:09:48,355 INFO [train.py:893] (1/4) Epoch 11, batch 2550, loss[loss=0.1784, simple_loss=0.2343, pruned_loss=0.06123, over 13333.00 frames. ], tot_loss[loss=0.2104, simple_loss=0.2608, pruned_loss=0.07996, over 2654908.89 frames. ], batch size: 67, lr: 1.17e-02, grad_scale: 32.0 2023-04-16 19:09:55,476 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-04-16 19:09:56,828 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29551.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:10:08,228 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29565.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:10:12,113 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-16 19:10:32,337 INFO [train.py:893] (1/4) Epoch 11, batch 2600, loss[loss=0.2288, simple_loss=0.2781, pruned_loss=0.08971, over 13365.00 frames. ], tot_loss[loss=0.2112, simple_loss=0.2612, pruned_loss=0.08054, over 2657219.03 frames. ], batch size: 118, lr: 1.17e-02, grad_scale: 32.0 2023-04-16 19:10:32,534 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29592.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:10:48,304 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.167e+02 3.181e+02 3.982e+02 4.759e+02 1.056e+03, threshold=7.963e+02, percent-clipped=3.0 2023-04-16 19:10:50,342 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29612.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:10:54,843 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3253, 3.6502, 3.4105, 3.9555, 2.1875, 2.9988, 3.7037, 2.1504], device='cuda:1'), covar=tensor([0.0118, 0.0504, 0.0780, 0.0597, 0.1604, 0.0921, 0.0643, 0.1940], device='cuda:1'), in_proj_covar=tensor([0.0099, 0.0168, 0.0189, 0.0198, 0.0177, 0.0187, 0.0172, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 19:11:13,234 INFO [train.py:893] (1/4) Epoch 11, batch 2650, loss[loss=0.2156, simple_loss=0.2694, pruned_loss=0.08088, over 13519.00 frames. ], tot_loss[loss=0.2121, simple_loss=0.2621, pruned_loss=0.08105, over 2660173.12 frames. ], batch size: 98, lr: 1.17e-02, grad_scale: 32.0 2023-04-16 19:11:22,135 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2023-04-16 19:11:25,451 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29658.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:11:28,438 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29662.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:12:11,515 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 19:12:19,418 INFO [train.py:893] (1/4) Epoch 12, batch 0, loss[loss=0.1617, simple_loss=0.2161, pruned_loss=0.05362, over 13381.00 frames. ], tot_loss[loss=0.1617, simple_loss=0.2161, pruned_loss=0.05362, over 13381.00 frames. ], batch size: 62, lr: 1.12e-02, grad_scale: 32.0 2023-04-16 19:12:19,419 INFO [train.py:918] (1/4) Computing validation loss 2023-04-16 19:12:26,162 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1477, 2.8515, 2.2899, 3.8944, 4.5105, 3.5216, 4.3554, 4.1390], device='cuda:1'), covar=tensor([0.0116, 0.0709, 0.0901, 0.0122, 0.0082, 0.0366, 0.0076, 0.0077], device='cuda:1'), in_proj_covar=tensor([0.0062, 0.0080, 0.0088, 0.0069, 0.0052, 0.0072, 0.0045, 0.0061], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 19:12:29,097 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.3304, 5.1915, 5.2068, 5.0573, 5.5698, 5.1276, 5.5404, 5.5559], device='cuda:1'), covar=tensor([0.0259, 0.0409, 0.0617, 0.0429, 0.0422, 0.0652, 0.0434, 0.0365], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0242, 0.0240, 0.0181, 0.0344, 0.0282, 0.0209, 0.0237], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 19:12:42,246 INFO [train.py:927] (1/4) Epoch 12, validation: loss=0.1552, simple_loss=0.2116, pruned_loss=0.04941, over 2446609.00 frames. 2023-04-16 19:12:42,248 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12829MB 2023-04-16 19:12:52,707 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7536, 2.5729, 2.2336, 2.9062, 2.2263, 2.9052, 2.7117, 2.5989], device='cuda:1'), covar=tensor([0.0099, 0.0164, 0.0185, 0.0142, 0.0178, 0.0118, 0.0231, 0.0151], device='cuda:1'), in_proj_covar=tensor([0.0075, 0.0081, 0.0089, 0.0078, 0.0093, 0.0072, 0.0077, 0.0076], device='cuda:1'), out_proj_covar=tensor([8.1349e-05, 9.2371e-05, 1.0343e-04, 8.8584e-05, 1.0690e-04, 7.9889e-05, 8.7058e-05, 8.4475e-05], device='cuda:1') 2023-04-16 19:12:56,720 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29706.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:12:59,565 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-04-16 19:12:59,727 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.306e+02 3.231e+02 3.816e+02 4.221e+02 8.534e+02, threshold=7.632e+02, percent-clipped=1.0 2023-04-16 19:13:06,357 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5583, 4.3815, 4.6121, 4.5312, 4.8320, 4.3627, 4.9214, 4.8505], device='cuda:1'), covar=tensor([0.0385, 0.0536, 0.0614, 0.0474, 0.0536, 0.0847, 0.0342, 0.0438], device='cuda:1'), in_proj_covar=tensor([0.0224, 0.0241, 0.0239, 0.0180, 0.0344, 0.0283, 0.0207, 0.0236], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 19:13:12,104 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29723.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 19:13:28,119 INFO [train.py:893] (1/4) Epoch 12, batch 50, loss[loss=0.2075, simple_loss=0.2525, pruned_loss=0.08121, over 13354.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2549, pruned_loss=0.07857, over 600330.11 frames. ], batch size: 73, lr: 1.12e-02, grad_scale: 32.0 2023-04-16 19:13:39,203 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-04-16 19:13:50,025 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-16 19:13:50,026 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-16 19:13:50,026 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-16 19:13:50,032 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-16 19:13:50,047 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-16 19:13:50,060 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-16 19:13:50,069 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-16 19:13:59,494 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.64 vs. limit=5.0 2023-04-16 19:14:13,986 INFO [train.py:893] (1/4) Epoch 12, batch 100, loss[loss=0.2301, simple_loss=0.275, pruned_loss=0.09265, over 13378.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.2575, pruned_loss=0.08094, over 1051283.75 frames. ], batch size: 113, lr: 1.12e-02, grad_scale: 32.0 2023-04-16 19:14:25,251 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=29802.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:14:32,111 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.956e+02 3.079e+02 3.520e+02 4.186e+02 7.081e+02, threshold=7.041e+02, percent-clipped=0.0 2023-04-16 19:15:01,244 INFO [train.py:893] (1/4) Epoch 12, batch 150, loss[loss=0.2025, simple_loss=0.2626, pruned_loss=0.07117, over 13469.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2607, pruned_loss=0.08207, over 1397919.71 frames. ], batch size: 79, lr: 1.12e-02, grad_scale: 32.0 2023-04-16 19:15:16,851 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1646, 4.5437, 3.2001, 4.4254, 4.3786, 2.8310, 3.8980, 2.9236], device='cuda:1'), covar=tensor([0.0283, 0.0232, 0.1033, 0.0285, 0.0220, 0.1135, 0.0548, 0.1445], device='cuda:1'), in_proj_covar=tensor([0.0137, 0.0140, 0.0170, 0.0139, 0.0118, 0.0153, 0.0147, 0.0157], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 19:15:19,968 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.8069, 1.9597, 3.6787, 3.5562, 3.4906, 2.8553, 3.3692, 2.5508], device='cuda:1'), covar=tensor([0.2037, 0.1682, 0.0157, 0.0203, 0.0266, 0.0718, 0.0259, 0.0945], device='cuda:1'), in_proj_covar=tensor([0.0203, 0.0188, 0.0106, 0.0111, 0.0120, 0.0162, 0.0119, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 19:15:21,684 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=29863.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:15:47,063 INFO [train.py:893] (1/4) Epoch 12, batch 200, loss[loss=0.2372, simple_loss=0.2681, pruned_loss=0.1031, over 13434.00 frames. ], tot_loss[loss=0.2146, simple_loss=0.2631, pruned_loss=0.08309, over 1677084.98 frames. ], batch size: 62, lr: 1.12e-02, grad_scale: 16.0 2023-04-16 19:15:48,003 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=29892.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:16:02,671 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=29907.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:16:05,677 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.117e+02 3.029e+02 3.563e+02 3.950e+02 7.099e+02, threshold=7.126e+02, percent-clipped=1.0 2023-04-16 19:16:33,378 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=29940.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:16:34,018 INFO [train.py:893] (1/4) Epoch 12, batch 250, loss[loss=0.2079, simple_loss=0.2628, pruned_loss=0.0765, over 13222.00 frames. ], tot_loss[loss=0.2142, simple_loss=0.2633, pruned_loss=0.08255, over 1897602.48 frames. ], batch size: 132, lr: 1.12e-02, grad_scale: 16.0 2023-04-16 19:16:46,357 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 2023-04-16 19:16:51,253 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-04-16 19:16:55,105 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3651, 4.6741, 4.4060, 4.2954, 4.4239, 4.8428, 4.6482, 4.4472], device='cuda:1'), covar=tensor([0.0266, 0.0264, 0.0290, 0.1088, 0.0287, 0.0225, 0.0283, 0.0254], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0128, 0.0139, 0.0236, 0.0142, 0.0156, 0.0140, 0.0100], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 19:17:10,434 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2022, 4.2895, 3.0093, 4.2394, 4.1655, 2.7329, 3.6290, 2.7627], device='cuda:1'), covar=tensor([0.0252, 0.0306, 0.1055, 0.0252, 0.0284, 0.1123, 0.0607, 0.1470], device='cuda:1'), in_proj_covar=tensor([0.0138, 0.0143, 0.0172, 0.0142, 0.0119, 0.0155, 0.0149, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 19:17:13,736 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-04-16 19:17:17,069 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7677, 2.5816, 2.1765, 1.5032, 1.4018, 1.9923, 1.9803, 2.8145], device='cuda:1'), covar=tensor([0.1029, 0.0380, 0.0883, 0.1753, 0.0321, 0.0384, 0.0869, 0.0251], device='cuda:1'), in_proj_covar=tensor([0.0136, 0.0108, 0.0097, 0.0179, 0.0092, 0.0125, 0.0135, 0.0107], device='cuda:1'), out_proj_covar=tensor([1.0356e-04, 8.2635e-05, 7.8172e-05, 1.3923e-04, 7.3351e-05, 9.4708e-05, 1.0337e-04, 7.9808e-05], device='cuda:1') 2023-04-16 19:17:20,729 INFO [train.py:893] (1/4) Epoch 12, batch 300, loss[loss=0.231, simple_loss=0.2688, pruned_loss=0.09664, over 13406.00 frames. ], tot_loss[loss=0.2141, simple_loss=0.2637, pruned_loss=0.08225, over 2069444.32 frames. ], batch size: 62, lr: 1.12e-02, grad_scale: 16.0 2023-04-16 19:17:40,958 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-04-16 19:17:43,026 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.413e+02 3.134e+02 3.533e+02 4.035e+02 9.509e+02, threshold=7.065e+02, percent-clipped=1.0 2023-04-16 19:17:49,998 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30018.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 19:17:56,020 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-04-16 19:18:11,224 INFO [train.py:893] (1/4) Epoch 12, batch 350, loss[loss=0.2124, simple_loss=0.26, pruned_loss=0.08243, over 13526.00 frames. ], tot_loss[loss=0.2154, simple_loss=0.2646, pruned_loss=0.08309, over 2198522.05 frames. ], batch size: 85, lr: 1.12e-02, grad_scale: 16.0 2023-04-16 19:18:58,017 INFO [train.py:893] (1/4) Epoch 12, batch 400, loss[loss=0.2169, simple_loss=0.2725, pruned_loss=0.08069, over 13521.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2646, pruned_loss=0.0828, over 2304096.55 frames. ], batch size: 91, lr: 1.11e-02, grad_scale: 16.0 2023-04-16 19:19:16,170 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.193e+02 3.172e+02 3.511e+02 4.299e+02 8.379e+02, threshold=7.022e+02, percent-clipped=2.0 2023-04-16 19:19:30,478 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30125.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:19:44,292 INFO [train.py:893] (1/4) Epoch 12, batch 450, loss[loss=0.2271, simple_loss=0.2748, pruned_loss=0.08973, over 13387.00 frames. ], tot_loss[loss=0.2166, simple_loss=0.2661, pruned_loss=0.08348, over 2383117.26 frames. ], batch size: 113, lr: 1.11e-02, grad_scale: 16.0 2023-04-16 19:20:00,054 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30158.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:20:07,940 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-04-16 19:20:08,997 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-16 19:20:26,279 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30186.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:20:30,077 INFO [train.py:893] (1/4) Epoch 12, batch 500, loss[loss=0.2283, simple_loss=0.2744, pruned_loss=0.09113, over 13520.00 frames. ], tot_loss[loss=0.2164, simple_loss=0.2661, pruned_loss=0.08334, over 2442716.18 frames. ], batch size: 91, lr: 1.11e-02, grad_scale: 16.0 2023-04-16 19:20:44,696 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30207.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:20:46,470 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30209.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:20:48,537 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.264e+02 3.071e+02 3.696e+02 4.723e+02 7.948e+02, threshold=7.391e+02, percent-clipped=1.0 2023-04-16 19:20:49,697 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.2440, 1.9811, 1.9466, 2.2488, 1.7348, 2.2583, 2.0422, 1.9106], device='cuda:1'), covar=tensor([0.0069, 0.0166, 0.0106, 0.0087, 0.0142, 0.0093, 0.0168, 0.0124], device='cuda:1'), in_proj_covar=tensor([0.0075, 0.0083, 0.0090, 0.0079, 0.0095, 0.0074, 0.0078, 0.0077], device='cuda:1'), out_proj_covar=tensor([8.1630e-05, 9.3239e-05, 1.0403e-04, 8.9196e-05, 1.0828e-04, 8.1335e-05, 8.8039e-05, 8.5760e-05], device='cuda:1') 2023-04-16 19:21:15,814 INFO [train.py:893] (1/4) Epoch 12, batch 550, loss[loss=0.1995, simple_loss=0.2568, pruned_loss=0.07112, over 13417.00 frames. ], tot_loss[loss=0.216, simple_loss=0.2661, pruned_loss=0.08297, over 2490453.01 frames. ], batch size: 95, lr: 1.11e-02, grad_scale: 16.0 2023-04-16 19:21:29,054 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30255.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:21:42,543 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30270.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:21:56,925 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-04-16 19:22:01,142 INFO [train.py:893] (1/4) Epoch 12, batch 600, loss[loss=0.2671, simple_loss=0.3022, pruned_loss=0.116, over 12054.00 frames. ], tot_loss[loss=0.2151, simple_loss=0.2649, pruned_loss=0.08267, over 2528063.01 frames. ], batch size: 158, lr: 1.11e-02, grad_scale: 16.0 2023-04-16 19:22:20,549 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.048e+02 3.005e+02 3.534e+02 4.580e+02 1.320e+03, threshold=7.068e+02, percent-clipped=5.0 2023-04-16 19:22:26,747 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30318.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 19:22:48,455 INFO [train.py:893] (1/4) Epoch 12, batch 650, loss[loss=0.2047, simple_loss=0.256, pruned_loss=0.07668, over 13384.00 frames. ], tot_loss[loss=0.2149, simple_loss=0.2646, pruned_loss=0.08258, over 2552972.72 frames. ], batch size: 118, lr: 1.11e-02, grad_scale: 16.0 2023-04-16 19:22:59,793 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-04-16 19:23:04,786 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30360.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:23:10,429 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30366.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:23:32,570 INFO [train.py:893] (1/4) Epoch 12, batch 700, loss[loss=0.2131, simple_loss=0.2469, pruned_loss=0.08966, over 12557.00 frames. ], tot_loss[loss=0.2129, simple_loss=0.2628, pruned_loss=0.0815, over 2571546.29 frames. ], batch size: 51, lr: 1.11e-02, grad_scale: 16.0 2023-04-16 19:23:52,289 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.015e+02 2.964e+02 3.447e+02 4.569e+02 1.039e+03, threshold=6.894e+02, percent-clipped=4.0 2023-04-16 19:23:53,511 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0238, 2.5426, 2.4747, 1.7080, 1.5193, 2.2735, 2.1737, 2.8799], device='cuda:1'), covar=tensor([0.0703, 0.0359, 0.0506, 0.1524, 0.0324, 0.0426, 0.0679, 0.0190], device='cuda:1'), in_proj_covar=tensor([0.0138, 0.0108, 0.0098, 0.0180, 0.0091, 0.0126, 0.0135, 0.0106], device='cuda:1'), out_proj_covar=tensor([1.0432e-04, 8.2640e-05, 7.8849e-05, 1.3916e-04, 7.3288e-05, 9.5932e-05, 1.0382e-04, 7.8957e-05], device='cuda:1') 2023-04-16 19:24:00,897 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30421.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:24:03,676 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2023-04-16 19:24:20,319 INFO [train.py:893] (1/4) Epoch 12, batch 750, loss[loss=0.2434, simple_loss=0.2851, pruned_loss=0.1008, over 13357.00 frames. ], tot_loss[loss=0.2132, simple_loss=0.2626, pruned_loss=0.08186, over 2585832.01 frames. ], batch size: 118, lr: 1.11e-02, grad_scale: 16.0 2023-04-16 19:24:34,869 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30458.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:24:46,088 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5989, 4.0233, 3.6329, 4.1871, 2.2819, 3.0258, 3.9385, 2.0065], device='cuda:1'), covar=tensor([0.0079, 0.0377, 0.0712, 0.0475, 0.1475, 0.0983, 0.0474, 0.2025], device='cuda:1'), in_proj_covar=tensor([0.0098, 0.0166, 0.0186, 0.0198, 0.0171, 0.0183, 0.0165, 0.0170], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 19:24:56,735 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30481.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:25:04,836 INFO [train.py:893] (1/4) Epoch 12, batch 800, loss[loss=0.2042, simple_loss=0.2647, pruned_loss=0.07188, over 13426.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2641, pruned_loss=0.08179, over 2605604.71 frames. ], batch size: 95, lr: 1.11e-02, grad_scale: 16.0 2023-04-16 19:25:20,078 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30506.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:25:23,177 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-04-16 19:25:24,122 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.034e+02 3.051e+02 3.650e+02 4.433e+02 6.512e+02, threshold=7.301e+02, percent-clipped=0.0 2023-04-16 19:25:51,402 INFO [train.py:893] (1/4) Epoch 12, batch 850, loss[loss=0.222, simple_loss=0.2661, pruned_loss=0.0889, over 13250.00 frames. ], tot_loss[loss=0.2143, simple_loss=0.2647, pruned_loss=0.0819, over 2619954.31 frames. ], batch size: 124, lr: 1.11e-02, grad_scale: 16.0 2023-04-16 19:26:13,891 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30565.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:26:37,263 INFO [train.py:893] (1/4) Epoch 12, batch 900, loss[loss=0.2033, simple_loss=0.257, pruned_loss=0.07481, over 13525.00 frames. ], tot_loss[loss=0.2139, simple_loss=0.2642, pruned_loss=0.08185, over 2626118.84 frames. ], batch size: 83, lr: 1.11e-02, grad_scale: 16.0 2023-04-16 19:26:54,220 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.806e+02 3.028e+02 3.474e+02 4.092e+02 7.864e+02, threshold=6.948e+02, percent-clipped=2.0 2023-04-16 19:27:07,865 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-16 19:27:21,879 INFO [train.py:893] (1/4) Epoch 12, batch 950, loss[loss=0.2019, simple_loss=0.2546, pruned_loss=0.07463, over 13520.00 frames. ], tot_loss[loss=0.2124, simple_loss=0.2622, pruned_loss=0.08124, over 2636310.07 frames. ], batch size: 76, lr: 1.10e-02, grad_scale: 16.0 2023-04-16 19:27:36,748 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30655.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:28:08,984 INFO [train.py:893] (1/4) Epoch 12, batch 1000, loss[loss=0.1881, simple_loss=0.2311, pruned_loss=0.0725, over 11804.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2593, pruned_loss=0.07976, over 2642113.77 frames. ], batch size: 48, lr: 1.10e-02, grad_scale: 16.0 2023-04-16 19:28:25,727 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.256e+02 3.039e+02 3.506e+02 4.139e+02 8.473e+02, threshold=7.013e+02, percent-clipped=1.0 2023-04-16 19:28:30,858 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=30716.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:28:30,989 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30716.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 19:28:31,978 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-04-16 19:28:52,240 INFO [train.py:893] (1/4) Epoch 12, batch 1050, loss[loss=0.1746, simple_loss=0.2381, pruned_loss=0.05555, over 13522.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2576, pruned_loss=0.07829, over 2636745.93 frames. ], batch size: 76, lr: 1.10e-02, grad_scale: 16.0 2023-04-16 19:29:12,454 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3924, 3.7171, 3.4540, 4.0832, 2.0650, 2.9333, 3.8217, 2.0258], device='cuda:1'), covar=tensor([0.0132, 0.0452, 0.0726, 0.0461, 0.1670, 0.0956, 0.0495, 0.1976], device='cuda:1'), in_proj_covar=tensor([0.0099, 0.0168, 0.0186, 0.0199, 0.0174, 0.0185, 0.0167, 0.0170], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 19:29:30,050 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30781.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:29:38,069 INFO [train.py:893] (1/4) Epoch 12, batch 1100, loss[loss=0.2152, simple_loss=0.2602, pruned_loss=0.08509, over 13431.00 frames. ], tot_loss[loss=0.2074, simple_loss=0.2584, pruned_loss=0.07827, over 2644208.11 frames. ], batch size: 95, lr: 1.10e-02, grad_scale: 16.0 2023-04-16 19:29:38,726 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.94 vs. limit=5.0 2023-04-16 19:29:56,589 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.861e+02 2.954e+02 3.511e+02 4.090e+02 1.224e+03, threshold=7.022e+02, percent-clipped=4.0 2023-04-16 19:30:12,551 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30829.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:30:23,875 INFO [train.py:893] (1/4) Epoch 12, batch 1150, loss[loss=0.2111, simple_loss=0.2651, pruned_loss=0.07849, over 13383.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2577, pruned_loss=0.07729, over 2649675.43 frames. ], batch size: 113, lr: 1.10e-02, grad_scale: 16.0 2023-04-16 19:30:28,028 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.1428, 4.7098, 4.5428, 4.6098, 4.2435, 4.4953, 5.0988, 4.6026], device='cuda:1'), covar=tensor([0.0636, 0.0921, 0.2032, 0.2391, 0.0898, 0.1351, 0.0816, 0.1005], device='cuda:1'), in_proj_covar=tensor([0.0282, 0.0316, 0.0401, 0.0414, 0.0238, 0.0304, 0.0366, 0.0285], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 19:30:45,817 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=30865.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:31:10,368 INFO [train.py:893] (1/4) Epoch 12, batch 1200, loss[loss=0.2028, simple_loss=0.2596, pruned_loss=0.07304, over 13516.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2576, pruned_loss=0.07657, over 2655329.84 frames. ], batch size: 83, lr: 1.10e-02, grad_scale: 16.0 2023-04-16 19:31:22,172 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30903.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:31:30,205 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.083e+02 2.813e+02 3.265e+02 3.803e+02 5.981e+02, threshold=6.529e+02, percent-clipped=0.0 2023-04-16 19:31:32,065 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=30913.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:31:39,311 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-16 19:31:42,190 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30925.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:31:50,871 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-16 19:31:51,122 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=30934.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:31:58,141 INFO [train.py:893] (1/4) Epoch 12, batch 1250, loss[loss=0.2275, simple_loss=0.2729, pruned_loss=0.09101, over 13480.00 frames. ], tot_loss[loss=0.2064, simple_loss=0.2582, pruned_loss=0.07731, over 2655820.45 frames. ], batch size: 93, lr: 1.10e-02, grad_scale: 16.0 2023-04-16 19:32:18,407 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30964.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:32:38,014 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30986.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:32:41,984 INFO [train.py:893] (1/4) Epoch 12, batch 1300, loss[loss=0.2383, simple_loss=0.2774, pruned_loss=0.09959, over 11918.00 frames. ], tot_loss[loss=0.2086, simple_loss=0.2603, pruned_loss=0.07841, over 2657223.61 frames. ], batch size: 157, lr: 1.10e-02, grad_scale: 16.0 2023-04-16 19:32:46,898 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=30995.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:33:01,085 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.190e+02 3.401e+02 4.070e+02 4.749e+02 7.610e+02, threshold=8.140e+02, percent-clipped=5.0 2023-04-16 19:33:01,285 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31011.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 19:33:01,398 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31011.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 19:33:05,372 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31016.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:33:28,353 INFO [train.py:893] (1/4) Epoch 12, batch 1350, loss[loss=0.2341, simple_loss=0.2769, pruned_loss=0.09564, over 13467.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2609, pruned_loss=0.07894, over 2654958.50 frames. ], batch size: 103, lr: 1.10e-02, grad_scale: 16.0 2023-04-16 19:33:43,134 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.65 vs. limit=2.0 2023-04-16 19:33:50,724 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31064.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:33:57,729 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31072.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 19:34:01,801 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3828, 4.7504, 4.4864, 4.5299, 4.4893, 4.9893, 4.6784, 4.6394], device='cuda:1'), covar=tensor([0.0284, 0.0300, 0.0238, 0.0948, 0.0249, 0.0194, 0.0242, 0.0199], device='cuda:1'), in_proj_covar=tensor([0.0129, 0.0131, 0.0142, 0.0239, 0.0144, 0.0160, 0.0141, 0.0100], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 19:34:14,693 INFO [train.py:893] (1/4) Epoch 12, batch 1400, loss[loss=0.2312, simple_loss=0.2787, pruned_loss=0.09188, over 13380.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.261, pruned_loss=0.07929, over 2653094.32 frames. ], batch size: 109, lr: 1.10e-02, grad_scale: 16.0 2023-04-16 19:34:22,380 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7123, 2.2934, 2.3186, 2.8036, 1.9718, 2.8024, 2.6680, 2.3590], device='cuda:1'), covar=tensor([0.0092, 0.0199, 0.0149, 0.0118, 0.0192, 0.0098, 0.0163, 0.0154], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0086, 0.0093, 0.0084, 0.0098, 0.0076, 0.0082, 0.0080], device='cuda:1'), out_proj_covar=tensor([8.4530e-05, 9.6313e-05, 1.0615e-04, 9.4730e-05, 1.1189e-04, 8.4182e-05, 9.2965e-05, 8.8550e-05], device='cuda:1') 2023-04-16 19:34:32,859 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.986e+02 2.961e+02 3.516e+02 3.951e+02 9.779e+02, threshold=7.031e+02, percent-clipped=1.0 2023-04-16 19:35:00,592 INFO [train.py:893] (1/4) Epoch 12, batch 1450, loss[loss=0.2257, simple_loss=0.2802, pruned_loss=0.08563, over 13450.00 frames. ], tot_loss[loss=0.2106, simple_loss=0.2616, pruned_loss=0.07987, over 2656711.21 frames. ], batch size: 103, lr: 1.10e-02, grad_scale: 16.0 2023-04-16 19:35:48,313 INFO [train.py:893] (1/4) Epoch 12, batch 1500, loss[loss=0.1938, simple_loss=0.2468, pruned_loss=0.07039, over 13370.00 frames. ], tot_loss[loss=0.2102, simple_loss=0.2614, pruned_loss=0.07955, over 2658688.34 frames. ], batch size: 73, lr: 1.10e-02, grad_scale: 16.0 2023-04-16 19:36:05,513 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.884e+02 3.015e+02 3.444e+02 4.195e+02 7.088e+02, threshold=6.887e+02, percent-clipped=1.0 2023-04-16 19:36:09,760 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0729, 2.5326, 1.9865, 3.9333, 4.4517, 3.4099, 4.3645, 4.1160], device='cuda:1'), covar=tensor([0.0097, 0.0833, 0.1044, 0.0106, 0.0065, 0.0444, 0.0076, 0.0072], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0081, 0.0088, 0.0070, 0.0053, 0.0073, 0.0046, 0.0062], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 19:36:20,998 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0685, 2.9138, 3.5592, 2.7869, 2.4568, 2.4709, 3.6883, 3.7830], device='cuda:1'), covar=tensor([0.1144, 0.1536, 0.0357, 0.1100, 0.1491, 0.1332, 0.0314, 0.0178], device='cuda:1'), in_proj_covar=tensor([0.0223, 0.0232, 0.0175, 0.0209, 0.0205, 0.0172, 0.0174, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 19:36:34,158 INFO [train.py:893] (1/4) Epoch 12, batch 1550, loss[loss=0.1658, simple_loss=0.2167, pruned_loss=0.05746, over 13119.00 frames. ], tot_loss[loss=0.21, simple_loss=0.261, pruned_loss=0.0795, over 2658610.94 frames. ], batch size: 58, lr: 1.09e-02, grad_scale: 16.0 2023-04-16 19:36:51,524 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31259.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:37:10,922 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31281.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:37:12,926 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-04-16 19:37:19,867 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31290.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:37:20,593 INFO [train.py:893] (1/4) Epoch 12, batch 1600, loss[loss=0.2291, simple_loss=0.2802, pruned_loss=0.089, over 13532.00 frames. ], tot_loss[loss=0.2097, simple_loss=0.261, pruned_loss=0.07924, over 2660913.98 frames. ], batch size: 83, lr: 1.09e-02, grad_scale: 16.0 2023-04-16 19:37:21,709 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8835, 3.8979, 2.7372, 3.7748, 3.7851, 2.2872, 3.3723, 2.6451], device='cuda:1'), covar=tensor([0.0231, 0.0191, 0.1139, 0.0357, 0.0222, 0.1258, 0.0555, 0.1327], device='cuda:1'), in_proj_covar=tensor([0.0141, 0.0146, 0.0172, 0.0146, 0.0122, 0.0157, 0.0151, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 19:37:38,545 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.949e+02 2.989e+02 3.671e+02 4.365e+02 8.305e+02, threshold=7.342e+02, percent-clipped=2.0 2023-04-16 19:37:38,772 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31311.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:38:06,493 INFO [train.py:893] (1/4) Epoch 12, batch 1650, loss[loss=0.2293, simple_loss=0.2857, pruned_loss=0.08643, over 13379.00 frames. ], tot_loss[loss=0.2101, simple_loss=0.2618, pruned_loss=0.07919, over 2658274.60 frames. ], batch size: 109, lr: 1.09e-02, grad_scale: 16.0 2023-04-16 19:38:22,888 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31359.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:38:29,441 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31367.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 19:38:45,188 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.94 vs. limit=2.0 2023-04-16 19:38:51,392 INFO [train.py:893] (1/4) Epoch 12, batch 1700, loss[loss=0.2047, simple_loss=0.2342, pruned_loss=0.08764, over 11266.00 frames. ], tot_loss[loss=0.21, simple_loss=0.262, pruned_loss=0.07895, over 2652468.67 frames. ], batch size: 46, lr: 1.09e-02, grad_scale: 16.0 2023-04-16 19:39:06,897 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1998, 4.3296, 2.8725, 4.1671, 4.1054, 2.6396, 3.7476, 2.8873], device='cuda:1'), covar=tensor([0.0245, 0.0233, 0.1148, 0.0248, 0.0211, 0.1192, 0.0435, 0.1284], device='cuda:1'), in_proj_covar=tensor([0.0141, 0.0147, 0.0174, 0.0147, 0.0122, 0.0157, 0.0151, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 19:39:09,985 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31409.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:39:11,406 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.968e+02 3.094e+02 3.544e+02 4.109e+02 6.529e+02, threshold=7.088e+02, percent-clipped=0.0 2023-04-16 19:39:39,658 INFO [train.py:893] (1/4) Epoch 12, batch 1750, loss[loss=0.1839, simple_loss=0.2381, pruned_loss=0.06486, over 13348.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.2599, pruned_loss=0.07752, over 2653645.98 frames. ], batch size: 73, lr: 1.09e-02, grad_scale: 16.0 2023-04-16 19:40:06,802 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31470.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:40:24,954 INFO [train.py:893] (1/4) Epoch 12, batch 1800, loss[loss=0.2076, simple_loss=0.2672, pruned_loss=0.07398, over 13489.00 frames. ], tot_loss[loss=0.2069, simple_loss=0.2594, pruned_loss=0.07719, over 2657307.66 frames. ], batch size: 93, lr: 1.09e-02, grad_scale: 16.0 2023-04-16 19:40:25,273 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31491.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 19:40:43,243 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.162e+02 2.911e+02 3.544e+02 4.163e+02 7.056e+02, threshold=7.087e+02, percent-clipped=0.0 2023-04-16 19:41:11,537 INFO [train.py:893] (1/4) Epoch 12, batch 1850, loss[loss=0.2052, simple_loss=0.2539, pruned_loss=0.07822, over 13517.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2586, pruned_loss=0.07681, over 2659705.01 frames. ], batch size: 70, lr: 1.09e-02, grad_scale: 16.0 2023-04-16 19:41:14,114 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-16 19:41:21,138 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31552.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 19:41:27,604 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31559.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:41:47,570 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31581.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:41:55,677 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31590.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:41:56,304 INFO [train.py:893] (1/4) Epoch 12, batch 1900, loss[loss=0.232, simple_loss=0.2747, pruned_loss=0.09463, over 11906.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2584, pruned_loss=0.07714, over 2657570.68 frames. ], batch size: 157, lr: 1.09e-02, grad_scale: 16.0 2023-04-16 19:42:11,518 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31607.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:42:14,659 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.895e+02 2.793e+02 3.446e+02 4.085e+02 7.140e+02, threshold=6.892e+02, percent-clipped=1.0 2023-04-16 19:42:31,979 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31629.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:42:39,456 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31638.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:42:41,734 INFO [train.py:893] (1/4) Epoch 12, batch 1950, loss[loss=0.1787, simple_loss=0.2153, pruned_loss=0.07103, over 11399.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.258, pruned_loss=0.07725, over 2659343.93 frames. ], batch size: 46, lr: 1.09e-02, grad_scale: 16.0 2023-04-16 19:43:04,131 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.10 vs. limit=5.0 2023-04-16 19:43:05,482 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=31667.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 19:43:28,473 INFO [train.py:893] (1/4) Epoch 12, batch 2000, loss[loss=0.2237, simple_loss=0.2804, pruned_loss=0.08349, over 13384.00 frames. ], tot_loss[loss=0.2076, simple_loss=0.2597, pruned_loss=0.07771, over 2662180.93 frames. ], batch size: 109, lr: 1.09e-02, grad_scale: 16.0 2023-04-16 19:43:33,295 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-16 19:43:39,325 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31704.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:43:45,688 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.068e+02 3.036e+02 3.598e+02 4.473e+02 7.037e+02, threshold=7.196e+02, percent-clipped=1.0 2023-04-16 19:43:49,165 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=31715.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 19:44:13,466 INFO [train.py:893] (1/4) Epoch 12, batch 2050, loss[loss=0.1805, simple_loss=0.2397, pruned_loss=0.06063, over 13569.00 frames. ], tot_loss[loss=0.2096, simple_loss=0.2615, pruned_loss=0.07878, over 2663074.81 frames. ], batch size: 78, lr: 1.09e-02, grad_scale: 16.0 2023-04-16 19:44:35,321 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31765.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:44:35,427 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31765.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:44:57,160 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31788.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:44:59,354 INFO [train.py:893] (1/4) Epoch 12, batch 2100, loss[loss=0.1902, simple_loss=0.251, pruned_loss=0.0647, over 13541.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2612, pruned_loss=0.07834, over 2664053.09 frames. ], batch size: 83, lr: 1.09e-02, grad_scale: 16.0 2023-04-16 19:45:01,420 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8430, 3.9462, 3.0550, 3.6872, 3.1958, 1.9850, 3.9219, 2.2481], device='cuda:1'), covar=tensor([0.0762, 0.0400, 0.0457, 0.0259, 0.0745, 0.2114, 0.0780, 0.1370], device='cuda:1'), in_proj_covar=tensor([0.0129, 0.0119, 0.0120, 0.0101, 0.0140, 0.0176, 0.0139, 0.0131], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 19:45:02,619 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-04-16 19:45:04,752 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0084, 4.1475, 3.3237, 2.8680, 2.8224, 2.4408, 4.3056, 2.3786], device='cuda:1'), covar=tensor([0.1426, 0.0339, 0.0810, 0.1691, 0.0800, 0.2949, 0.0210, 0.3659], device='cuda:1'), in_proj_covar=tensor([0.0294, 0.0271, 0.0275, 0.0288, 0.0229, 0.0293, 0.0186, 0.0339], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 19:45:18,211 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.630e+02 2.912e+02 3.561e+02 4.179e+02 8.521e+02, threshold=7.122e+02, percent-clipped=3.0 2023-04-16 19:45:22,531 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4893, 2.3258, 2.3657, 3.9270, 3.5674, 3.9748, 3.0503, 2.3214], device='cuda:1'), covar=tensor([0.0249, 0.1006, 0.0912, 0.0051, 0.0220, 0.0040, 0.0696, 0.0964], device='cuda:1'), in_proj_covar=tensor([0.0110, 0.0149, 0.0159, 0.0085, 0.0111, 0.0078, 0.0160, 0.0143], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0001, 0.0003, 0.0002], device='cuda:1') 2023-04-16 19:45:25,715 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9006, 1.8590, 3.6316, 3.5042, 3.4346, 2.7727, 3.3447, 2.5747], device='cuda:1'), covar=tensor([0.2088, 0.1642, 0.0126, 0.0176, 0.0177, 0.0732, 0.0209, 0.0863], device='cuda:1'), in_proj_covar=tensor([0.0202, 0.0183, 0.0105, 0.0113, 0.0119, 0.0162, 0.0119, 0.0170], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 19:45:29,079 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31823.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:45:45,575 INFO [train.py:893] (1/4) Epoch 12, batch 2150, loss[loss=0.1963, simple_loss=0.2479, pruned_loss=0.07239, over 13340.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2607, pruned_loss=0.07754, over 2666462.90 frames. ], batch size: 118, lr: 1.08e-02, grad_scale: 16.0 2023-04-16 19:45:51,538 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=31847.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 19:45:53,416 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31849.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:46:12,676 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6383, 4.4947, 4.7080, 4.5189, 4.9249, 4.4672, 4.9603, 4.9247], device='cuda:1'), covar=tensor([0.0374, 0.0573, 0.0695, 0.0562, 0.0607, 0.0822, 0.0451, 0.0539], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0250, 0.0250, 0.0184, 0.0351, 0.0285, 0.0216, 0.0246], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 19:46:21,041 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31878.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:46:22,078 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.53 vs. limit=5.0 2023-04-16 19:46:26,164 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31884.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:46:31,452 INFO [train.py:893] (1/4) Epoch 12, batch 2200, loss[loss=0.1828, simple_loss=0.2388, pruned_loss=0.06347, over 13342.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2595, pruned_loss=0.07629, over 2668084.91 frames. ], batch size: 67, lr: 1.08e-02, grad_scale: 32.0 2023-04-16 19:46:50,881 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.988e+02 2.906e+02 3.487e+02 4.056e+02 8.174e+02, threshold=6.975e+02, percent-clipped=2.0 2023-04-16 19:46:54,609 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31915.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:46:59,495 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.2754, 1.8938, 2.2825, 3.6497, 3.3317, 3.7136, 2.8016, 2.0809], device='cuda:1'), covar=tensor([0.0257, 0.1146, 0.0893, 0.0057, 0.0222, 0.0046, 0.0642, 0.0987], device='cuda:1'), in_proj_covar=tensor([0.0110, 0.0150, 0.0160, 0.0085, 0.0111, 0.0079, 0.0161, 0.0144], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0001, 0.0003, 0.0002], device='cuda:1') 2023-04-16 19:47:01,133 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8931, 4.0346, 3.8685, 3.3960, 3.9561, 4.2380, 4.1331, 3.9639], device='cuda:1'), covar=tensor([0.0321, 0.0385, 0.0419, 0.1920, 0.0335, 0.0399, 0.0345, 0.0626], device='cuda:1'), in_proj_covar=tensor([0.0132, 0.0134, 0.0146, 0.0245, 0.0146, 0.0163, 0.0143, 0.0102], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 19:47:03,798 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3761, 2.6887, 2.1350, 4.1899, 4.8068, 3.5032, 4.7272, 4.3494], device='cuda:1'), covar=tensor([0.0104, 0.0800, 0.1012, 0.0117, 0.0061, 0.0457, 0.0064, 0.0088], device='cuda:1'), in_proj_covar=tensor([0.0064, 0.0082, 0.0089, 0.0070, 0.0054, 0.0074, 0.0047, 0.0064], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 19:47:16,798 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31939.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:47:18,208 INFO [train.py:893] (1/4) Epoch 12, batch 2250, loss[loss=0.1972, simple_loss=0.2466, pruned_loss=0.0739, over 13521.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2582, pruned_loss=0.07612, over 2667523.32 frames. ], batch size: 72, lr: 1.08e-02, grad_scale: 32.0 2023-04-16 19:47:26,880 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=31950.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:47:51,263 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=31976.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:48:00,995 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.7255, 4.5507, 4.7613, 4.6572, 5.0318, 4.5525, 5.0464, 4.9995], device='cuda:1'), covar=tensor([0.0345, 0.0528, 0.0701, 0.0491, 0.0491, 0.0773, 0.0422, 0.0487], device='cuda:1'), in_proj_covar=tensor([0.0230, 0.0251, 0.0248, 0.0183, 0.0352, 0.0285, 0.0217, 0.0246], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 19:48:04,691 INFO [train.py:893] (1/4) Epoch 12, batch 2300, loss[loss=0.2019, simple_loss=0.2565, pruned_loss=0.07371, over 13527.00 frames. ], tot_loss[loss=0.2044, simple_loss=0.2574, pruned_loss=0.07576, over 2664222.97 frames. ], batch size: 83, lr: 1.08e-02, grad_scale: 32.0 2023-04-16 19:48:21,477 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1732, 4.7557, 4.4411, 4.4821, 4.5394, 4.3853, 4.8000, 4.7845], device='cuda:1'), covar=tensor([0.0244, 0.0196, 0.0209, 0.0270, 0.0241, 0.0249, 0.0275, 0.0218], device='cuda:1'), in_proj_covar=tensor([0.0114, 0.0172, 0.0133, 0.0156, 0.0120, 0.0170, 0.0114, 0.0141], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 19:48:27,134 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.818e+02 2.839e+02 3.553e+02 4.109e+02 7.664e+02, threshold=7.105e+02, percent-clipped=2.0 2023-04-16 19:48:27,461 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=32011.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:48:54,938 INFO [train.py:893] (1/4) Epoch 12, batch 2350, loss[loss=0.2026, simple_loss=0.2554, pruned_loss=0.0749, over 11688.00 frames. ], tot_loss[loss=0.205, simple_loss=0.258, pruned_loss=0.07603, over 2662163.67 frames. ], batch size: 157, lr: 1.08e-02, grad_scale: 32.0 2023-04-16 19:48:57,130 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3214, 4.5582, 3.8245, 3.0788, 3.2132, 2.6094, 4.6968, 2.7280], device='cuda:1'), covar=tensor([0.1341, 0.0257, 0.0642, 0.1590, 0.0618, 0.2913, 0.0140, 0.3047], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0275, 0.0279, 0.0294, 0.0232, 0.0299, 0.0188, 0.0344], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 19:49:13,264 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32060.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:49:13,411 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9547, 2.5307, 2.1896, 1.7462, 1.4325, 2.2154, 2.1152, 2.7872], device='cuda:1'), covar=tensor([0.0839, 0.0338, 0.0879, 0.1559, 0.0403, 0.0424, 0.0796, 0.0278], device='cuda:1'), in_proj_covar=tensor([0.0138, 0.0112, 0.0101, 0.0181, 0.0095, 0.0131, 0.0139, 0.0108], device='cuda:1'), out_proj_covar=tensor([1.0521e-04, 8.4964e-05, 8.1064e-05, 1.4032e-04, 7.5064e-05, 9.9278e-05, 1.0745e-04, 8.0842e-05], device='cuda:1') 2023-04-16 19:49:15,724 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-16 19:49:17,460 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32065.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:49:19,330 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.2914, 4.4506, 3.6550, 2.9934, 3.0827, 2.5620, 4.5588, 2.5656], device='cuda:1'), covar=tensor([0.1267, 0.0227, 0.0681, 0.1500, 0.0606, 0.2740, 0.0160, 0.3209], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0274, 0.0279, 0.0295, 0.0232, 0.0299, 0.0187, 0.0344], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 19:49:41,338 INFO [train.py:893] (1/4) Epoch 12, batch 2400, loss[loss=0.1949, simple_loss=0.2491, pruned_loss=0.07035, over 13457.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2569, pruned_loss=0.07545, over 2665223.59 frames. ], batch size: 100, lr: 1.08e-02, grad_scale: 32.0 2023-04-16 19:49:59,437 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.098e+02 2.793e+02 3.359e+02 3.984e+02 7.428e+02, threshold=6.718e+02, percent-clipped=1.0 2023-04-16 19:50:01,218 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32113.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:50:27,472 INFO [train.py:893] (1/4) Epoch 12, batch 2450, loss[loss=0.2431, simple_loss=0.2856, pruned_loss=0.1003, over 11807.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2574, pruned_loss=0.07585, over 2667102.05 frames. ], batch size: 157, lr: 1.08e-02, grad_scale: 32.0 2023-04-16 19:50:30,120 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32144.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:50:33,450 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32147.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 19:50:45,184 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6119, 2.6839, 2.7795, 4.0818, 3.6723, 4.1511, 3.3297, 2.5934], device='cuda:1'), covar=tensor([0.0232, 0.0855, 0.0766, 0.0048, 0.0214, 0.0039, 0.0590, 0.0823], device='cuda:1'), in_proj_covar=tensor([0.0109, 0.0147, 0.0156, 0.0084, 0.0108, 0.0078, 0.0159, 0.0140], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0001, 0.0003, 0.0002], device='cuda:1') 2023-04-16 19:51:03,341 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32179.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:51:14,864 INFO [train.py:893] (1/4) Epoch 12, batch 2500, loss[loss=0.2176, simple_loss=0.265, pruned_loss=0.0851, over 13221.00 frames. ], tot_loss[loss=0.2035, simple_loss=0.2566, pruned_loss=0.07515, over 2666602.38 frames. ], batch size: 132, lr: 1.08e-02, grad_scale: 32.0 2023-04-16 19:51:18,508 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32195.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 19:51:33,612 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.938e+02 2.861e+02 3.286e+02 4.044e+02 6.156e+02, threshold=6.571e+02, percent-clipped=0.0 2023-04-16 19:51:55,407 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32234.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:52:01,054 INFO [train.py:893] (1/4) Epoch 12, batch 2550, loss[loss=0.1531, simple_loss=0.2111, pruned_loss=0.0475, over 13133.00 frames. ], tot_loss[loss=0.2028, simple_loss=0.2563, pruned_loss=0.07463, over 2671923.47 frames. ], batch size: 58, lr: 1.08e-02, grad_scale: 32.0 2023-04-16 19:52:15,711 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-04-16 19:52:23,423 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 2023-04-16 19:52:26,054 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-16 19:52:29,377 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32271.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:52:32,834 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4620, 3.4512, 4.0554, 3.0163, 2.7629, 2.8033, 4.2726, 4.4022], device='cuda:1'), covar=tensor([0.1005, 0.1290, 0.0314, 0.1341, 0.1394, 0.1241, 0.0199, 0.0192], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0234, 0.0172, 0.0206, 0.0202, 0.0167, 0.0171, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 19:52:48,282 INFO [train.py:893] (1/4) Epoch 12, batch 2600, loss[loss=0.2052, simple_loss=0.2562, pruned_loss=0.0771, over 13478.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2568, pruned_loss=0.07519, over 2670514.80 frames. ], batch size: 100, lr: 1.08e-02, grad_scale: 32.0 2023-04-16 19:52:54,438 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8803, 3.9397, 3.3307, 2.7489, 2.7979, 2.3806, 4.1049, 2.3038], device='cuda:1'), covar=tensor([0.1271, 0.0263, 0.0686, 0.1444, 0.0665, 0.2600, 0.0190, 0.3357], device='cuda:1'), in_proj_covar=tensor([0.0294, 0.0267, 0.0274, 0.0290, 0.0229, 0.0294, 0.0184, 0.0337], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 19:53:02,235 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=32306.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:53:07,069 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.200e+02 2.795e+02 3.210e+02 3.908e+02 6.210e+02, threshold=6.421e+02, percent-clipped=0.0 2023-04-16 19:53:11,118 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3104, 2.6237, 2.2341, 4.1565, 4.7700, 3.5564, 4.6578, 4.3935], device='cuda:1'), covar=tensor([0.0103, 0.0800, 0.0981, 0.0110, 0.0075, 0.0423, 0.0070, 0.0082], device='cuda:1'), in_proj_covar=tensor([0.0063, 0.0080, 0.0087, 0.0069, 0.0054, 0.0072, 0.0046, 0.0062], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 19:53:30,384 INFO [train.py:893] (1/4) Epoch 12, batch 2650, loss[loss=0.1724, simple_loss=0.2295, pruned_loss=0.05761, over 13522.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2574, pruned_loss=0.07606, over 2667189.09 frames. ], batch size: 70, lr: 1.08e-02, grad_scale: 32.0 2023-04-16 19:53:44,868 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32360.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:53:54,089 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.66 vs. limit=5.0 2023-04-16 19:54:28,463 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 19:54:37,909 INFO [train.py:893] (1/4) Epoch 13, batch 0, loss[loss=0.2038, simple_loss=0.2592, pruned_loss=0.07422, over 13445.00 frames. ], tot_loss[loss=0.2038, simple_loss=0.2592, pruned_loss=0.07422, over 13445.00 frames. ], batch size: 103, lr: 1.03e-02, grad_scale: 16.0 2023-04-16 19:54:37,909 INFO [train.py:918] (1/4) Computing validation loss 2023-04-16 19:55:00,635 INFO [train.py:927] (1/4) Epoch 13, validation: loss=0.1517, simple_loss=0.209, pruned_loss=0.04718, over 2446609.00 frames. 2023-04-16 19:55:00,636 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12829MB 2023-04-16 19:55:17,715 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32408.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:55:20,767 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.013e+02 3.015e+02 3.552e+02 4.254e+02 7.750e+02, threshold=7.105e+02, percent-clipped=5.0 2023-04-16 19:55:47,342 INFO [train.py:893] (1/4) Epoch 13, batch 50, loss[loss=0.198, simple_loss=0.2445, pruned_loss=0.07581, over 13386.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2536, pruned_loss=0.0761, over 600149.09 frames. ], batch size: 62, lr: 1.03e-02, grad_scale: 16.0 2023-04-16 19:55:50,970 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32444.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:56:02,504 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.7922, 4.1502, 3.8429, 4.5533, 2.3922, 3.2692, 4.3517, 2.5054], device='cuda:1'), covar=tensor([0.0109, 0.0409, 0.0658, 0.0467, 0.1431, 0.0850, 0.0323, 0.1630], device='cuda:1'), in_proj_covar=tensor([0.0102, 0.0173, 0.0193, 0.0206, 0.0178, 0.0190, 0.0172, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 19:56:10,482 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-16 19:56:10,482 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-16 19:56:10,483 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-16 19:56:10,489 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-16 19:56:10,497 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-16 19:56:11,296 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-16 19:56:11,324 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-16 19:56:23,200 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32479.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:56:33,351 INFO [train.py:893] (1/4) Epoch 13, batch 100, loss[loss=0.2307, simple_loss=0.2777, pruned_loss=0.09182, over 13449.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2563, pruned_loss=0.07918, over 1055247.04 frames. ], batch size: 103, lr: 1.03e-02, grad_scale: 16.0 2023-04-16 19:56:35,942 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32492.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:56:53,988 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.249e+02 2.922e+02 3.454e+02 3.991e+02 1.041e+03, threshold=6.908e+02, percent-clipped=2.0 2023-04-16 19:57:07,965 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32527.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:57:14,729 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32534.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:57:19,382 INFO [train.py:893] (1/4) Epoch 13, batch 150, loss[loss=0.2001, simple_loss=0.2537, pruned_loss=0.07329, over 13165.00 frames. ], tot_loss[loss=0.2093, simple_loss=0.2585, pruned_loss=0.07999, over 1410140.58 frames. ], batch size: 132, lr: 1.03e-02, grad_scale: 16.0 2023-04-16 19:57:39,611 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-04-16 19:57:48,427 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32571.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:57:58,988 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32582.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:58:06,370 INFO [train.py:893] (1/4) Epoch 13, batch 200, loss[loss=0.2209, simple_loss=0.2538, pruned_loss=0.09405, over 13417.00 frames. ], tot_loss[loss=0.2116, simple_loss=0.261, pruned_loss=0.08111, over 1684121.00 frames. ], batch size: 65, lr: 1.03e-02, grad_scale: 16.0 2023-04-16 19:58:11,062 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.81 vs. limit=5.0 2023-04-16 19:58:20,952 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=32606.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:58:21,067 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5414, 3.4201, 4.5905, 4.3825, 4.4653, 4.0465, 4.3388, 3.7498], device='cuda:1'), covar=tensor([0.0946, 0.0756, 0.0052, 0.0149, 0.0115, 0.0290, 0.0109, 0.0453], device='cuda:1'), in_proj_covar=tensor([0.0207, 0.0184, 0.0108, 0.0117, 0.0122, 0.0166, 0.0121, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 19:58:27,220 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.025e+02 3.162e+02 3.651e+02 4.238e+02 1.178e+03, threshold=7.301e+02, percent-clipped=2.0 2023-04-16 19:58:31,827 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.60 vs. limit=5.0 2023-04-16 19:58:33,898 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32619.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:58:52,785 INFO [train.py:893] (1/4) Epoch 13, batch 250, loss[loss=0.2, simple_loss=0.2571, pruned_loss=0.07146, over 13371.00 frames. ], tot_loss[loss=0.2105, simple_loss=0.2606, pruned_loss=0.0802, over 1899307.26 frames. ], batch size: 118, lr: 1.03e-02, grad_scale: 16.0 2023-04-16 19:59:06,377 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=32654.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 19:59:40,136 INFO [train.py:893] (1/4) Epoch 13, batch 300, loss[loss=0.2022, simple_loss=0.2555, pruned_loss=0.07447, over 13528.00 frames. ], tot_loss[loss=0.2098, simple_loss=0.2608, pruned_loss=0.0794, over 2069186.94 frames. ], batch size: 70, lr: 1.03e-02, grad_scale: 16.0 2023-04-16 19:59:59,900 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.826e+02 2.901e+02 3.465e+02 4.067e+02 1.042e+03, threshold=6.930e+02, percent-clipped=2.0 2023-04-16 20:00:26,609 INFO [train.py:893] (1/4) Epoch 13, batch 350, loss[loss=0.1823, simple_loss=0.2429, pruned_loss=0.06089, over 13584.00 frames. ], tot_loss[loss=0.2095, simple_loss=0.2608, pruned_loss=0.07914, over 2198640.69 frames. ], batch size: 89, lr: 1.03e-02, grad_scale: 16.0 2023-04-16 20:00:33,715 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9313, 3.7082, 3.8535, 2.2518, 4.2633, 3.9232, 3.9654, 4.2136], device='cuda:1'), covar=tensor([0.0240, 0.0164, 0.0158, 0.1183, 0.0137, 0.0280, 0.0148, 0.0086], device='cuda:1'), in_proj_covar=tensor([0.0075, 0.0043, 0.0067, 0.0092, 0.0082, 0.0084, 0.0067, 0.0059], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 20:00:36,237 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8304, 2.3693, 1.6784, 3.5626, 4.1450, 3.0713, 4.0454, 3.8451], device='cuda:1'), covar=tensor([0.0108, 0.0973, 0.1145, 0.0132, 0.0068, 0.0489, 0.0090, 0.0099], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0084, 0.0090, 0.0071, 0.0055, 0.0073, 0.0047, 0.0065], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 20:00:59,214 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7715, 2.6521, 2.8099, 4.1848, 3.8285, 4.2422, 3.1555, 2.6413], device='cuda:1'), covar=tensor([0.0269, 0.0998, 0.0975, 0.0046, 0.0210, 0.0037, 0.0742, 0.1014], device='cuda:1'), in_proj_covar=tensor([0.0112, 0.0153, 0.0162, 0.0086, 0.0111, 0.0080, 0.0163, 0.0146], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0001, 0.0003, 0.0003], device='cuda:1') 2023-04-16 20:01:11,550 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8629, 4.2356, 4.0031, 3.9397, 3.9690, 4.3062, 4.1009, 3.8419], device='cuda:1'), covar=tensor([0.0279, 0.0222, 0.0265, 0.0829, 0.0281, 0.0217, 0.0263, 0.0460], device='cuda:1'), in_proj_covar=tensor([0.0133, 0.0134, 0.0145, 0.0243, 0.0148, 0.0162, 0.0142, 0.0101], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 20:01:12,941 INFO [train.py:893] (1/4) Epoch 13, batch 400, loss[loss=0.2115, simple_loss=0.2621, pruned_loss=0.08046, over 13463.00 frames. ], tot_loss[loss=0.2092, simple_loss=0.2606, pruned_loss=0.07884, over 2303283.60 frames. ], batch size: 103, lr: 1.03e-02, grad_scale: 16.0 2023-04-16 20:01:33,953 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.169e+02 2.983e+02 3.508e+02 4.207e+02 1.024e+03, threshold=7.015e+02, percent-clipped=2.0 2023-04-16 20:01:59,756 INFO [train.py:893] (1/4) Epoch 13, batch 450, loss[loss=0.2013, simple_loss=0.2556, pruned_loss=0.07353, over 13525.00 frames. ], tot_loss[loss=0.2099, simple_loss=0.2613, pruned_loss=0.07924, over 2376670.86 frames. ], batch size: 76, lr: 1.03e-02, grad_scale: 16.0 2023-04-16 20:02:02,568 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3925, 3.3162, 3.9152, 2.9311, 2.6243, 2.7102, 4.1591, 4.3029], device='cuda:1'), covar=tensor([0.1018, 0.1404, 0.0353, 0.1361, 0.1559, 0.1444, 0.0268, 0.0236], device='cuda:1'), in_proj_covar=tensor([0.0226, 0.0237, 0.0175, 0.0209, 0.0207, 0.0170, 0.0175, 0.0148], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 20:02:26,234 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-16 20:02:46,453 INFO [train.py:893] (1/4) Epoch 13, batch 500, loss[loss=0.1957, simple_loss=0.2468, pruned_loss=0.07229, over 13554.00 frames. ], tot_loss[loss=0.2094, simple_loss=0.2613, pruned_loss=0.07873, over 2443691.97 frames. ], batch size: 72, lr: 1.03e-02, grad_scale: 16.0 2023-04-16 20:03:07,864 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.049e+02 2.979e+02 3.537e+02 4.144e+02 6.731e+02, threshold=7.074e+02, percent-clipped=0.0 2023-04-16 20:03:33,886 INFO [train.py:893] (1/4) Epoch 13, batch 550, loss[loss=0.1797, simple_loss=0.2404, pruned_loss=0.05948, over 13545.00 frames. ], tot_loss[loss=0.209, simple_loss=0.2609, pruned_loss=0.07856, over 2490206.60 frames. ], batch size: 87, lr: 1.02e-02, grad_scale: 16.0 2023-04-16 20:03:47,773 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.86 vs. limit=2.0 2023-04-16 20:03:49,959 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6715, 2.5119, 3.0207, 4.1903, 3.7845, 4.2540, 3.5273, 2.8456], device='cuda:1'), covar=tensor([0.0352, 0.1066, 0.0737, 0.0046, 0.0224, 0.0034, 0.0503, 0.0802], device='cuda:1'), in_proj_covar=tensor([0.0112, 0.0154, 0.0161, 0.0086, 0.0111, 0.0080, 0.0162, 0.0146], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0001, 0.0003, 0.0003], device='cuda:1') 2023-04-16 20:03:59,030 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=32966.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:04:20,143 INFO [train.py:893] (1/4) Epoch 13, batch 600, loss[loss=0.1855, simple_loss=0.2457, pruned_loss=0.06268, over 13494.00 frames. ], tot_loss[loss=0.2071, simple_loss=0.2591, pruned_loss=0.07751, over 2531689.04 frames. ], batch size: 93, lr: 1.02e-02, grad_scale: 16.0 2023-04-16 20:04:41,170 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.223e+02 2.895e+02 3.369e+02 4.031e+02 6.824e+02, threshold=6.738e+02, percent-clipped=0.0 2023-04-16 20:04:55,568 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33027.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:05:07,988 INFO [train.py:893] (1/4) Epoch 13, batch 650, loss[loss=0.196, simple_loss=0.2432, pruned_loss=0.07444, over 12002.00 frames. ], tot_loss[loss=0.2059, simple_loss=0.258, pruned_loss=0.0769, over 2553689.64 frames. ], batch size: 157, lr: 1.02e-02, grad_scale: 16.0 2023-04-16 20:05:34,756 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33069.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:05:44,788 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8967, 2.5706, 2.3570, 2.9141, 2.2662, 3.1084, 2.8196, 2.5384], device='cuda:1'), covar=tensor([0.0100, 0.0147, 0.0171, 0.0163, 0.0205, 0.0104, 0.0233, 0.0169], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0085, 0.0095, 0.0084, 0.0100, 0.0079, 0.0083, 0.0082], device='cuda:1'), out_proj_covar=tensor([8.3897e-05, 9.4488e-05, 1.0779e-04, 9.3655e-05, 1.1311e-04, 8.6293e-05, 9.2727e-05, 8.9686e-05], device='cuda:1') 2023-04-16 20:05:54,198 INFO [train.py:893] (1/4) Epoch 13, batch 700, loss[loss=0.2117, simple_loss=0.26, pruned_loss=0.0817, over 13516.00 frames. ], tot_loss[loss=0.2053, simple_loss=0.2574, pruned_loss=0.07662, over 2578633.84 frames. ], batch size: 85, lr: 1.02e-02, grad_scale: 16.0 2023-04-16 20:06:10,384 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0064, 3.8191, 3.0122, 3.6245, 3.1980, 2.1058, 3.8764, 2.0482], device='cuda:1'), covar=tensor([0.0694, 0.0711, 0.0494, 0.0282, 0.0719, 0.1897, 0.0848, 0.1415], device='cuda:1'), in_proj_covar=tensor([0.0130, 0.0121, 0.0120, 0.0101, 0.0139, 0.0175, 0.0141, 0.0132], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 20:06:14,877 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.165e+02 2.977e+02 3.487e+02 4.255e+02 9.376e+02, threshold=6.973e+02, percent-clipped=2.0 2023-04-16 20:06:31,820 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33130.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:06:40,698 INFO [train.py:893] (1/4) Epoch 13, batch 750, loss[loss=0.2267, simple_loss=0.2593, pruned_loss=0.09707, over 13144.00 frames. ], tot_loss[loss=0.2063, simple_loss=0.2578, pruned_loss=0.07736, over 2595766.45 frames. ], batch size: 58, lr: 1.02e-02, grad_scale: 16.0 2023-04-16 20:06:54,937 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33155.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:07:27,558 INFO [train.py:893] (1/4) Epoch 13, batch 800, loss[loss=0.2072, simple_loss=0.2512, pruned_loss=0.08153, over 13374.00 frames. ], tot_loss[loss=0.2075, simple_loss=0.259, pruned_loss=0.078, over 2612941.46 frames. ], batch size: 62, lr: 1.02e-02, grad_scale: 16.0 2023-04-16 20:07:48,342 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.106e+02 3.031e+02 3.657e+02 4.273e+02 9.126e+02, threshold=7.314e+02, percent-clipped=2.0 2023-04-16 20:07:52,088 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33216.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:08:13,690 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9567, 1.7692, 3.6350, 3.5201, 3.4845, 2.6418, 3.2738, 2.6236], device='cuda:1'), covar=tensor([0.1827, 0.1601, 0.0091, 0.0151, 0.0199, 0.0821, 0.0195, 0.0752], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0185, 0.0109, 0.0116, 0.0122, 0.0169, 0.0124, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 20:08:14,169 INFO [train.py:893] (1/4) Epoch 13, batch 850, loss[loss=0.2149, simple_loss=0.2585, pruned_loss=0.08567, over 13262.00 frames. ], tot_loss[loss=0.2079, simple_loss=0.2597, pruned_loss=0.07799, over 2619722.54 frames. ], batch size: 124, lr: 1.02e-02, grad_scale: 16.0 2023-04-16 20:08:55,461 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8624, 2.5872, 1.8790, 3.7309, 4.1791, 3.1927, 4.0880, 3.8525], device='cuda:1'), covar=tensor([0.0091, 0.0787, 0.0958, 0.0089, 0.0045, 0.0435, 0.0072, 0.0081], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0082, 0.0089, 0.0070, 0.0054, 0.0074, 0.0047, 0.0064], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 20:08:55,513 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7244, 2.3228, 2.1407, 1.4797, 1.6608, 1.9179, 2.0321, 2.4555], device='cuda:1'), covar=tensor([0.0734, 0.0227, 0.0563, 0.1356, 0.0175, 0.0269, 0.0518, 0.0197], device='cuda:1'), in_proj_covar=tensor([0.0140, 0.0111, 0.0101, 0.0184, 0.0095, 0.0131, 0.0138, 0.0110], device='cuda:1'), out_proj_covar=tensor([1.0647e-04, 8.4459e-05, 8.1249e-05, 1.4176e-04, 7.4460e-05, 9.9147e-05, 1.0655e-04, 8.2514e-05], device='cuda:1') 2023-04-16 20:09:00,995 INFO [train.py:893] (1/4) Epoch 13, batch 900, loss[loss=0.2207, simple_loss=0.2691, pruned_loss=0.08608, over 13349.00 frames. ], tot_loss[loss=0.2088, simple_loss=0.2601, pruned_loss=0.07874, over 2631390.63 frames. ], batch size: 118, lr: 1.02e-02, grad_scale: 16.0 2023-04-16 20:09:06,252 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7246, 2.3190, 2.1787, 2.8286, 1.9807, 2.8247, 2.6635, 2.2953], device='cuda:1'), covar=tensor([0.0081, 0.0180, 0.0146, 0.0116, 0.0187, 0.0105, 0.0181, 0.0172], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0086, 0.0093, 0.0085, 0.0098, 0.0078, 0.0082, 0.0080], device='cuda:1'), out_proj_covar=tensor([8.3563e-05, 9.5589e-05, 1.0632e-04, 9.3974e-05, 1.1076e-04, 8.5135e-05, 9.1770e-05, 8.7879e-05], device='cuda:1') 2023-04-16 20:09:21,683 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.055e+02 2.966e+02 3.370e+02 4.159e+02 7.301e+02, threshold=6.740e+02, percent-clipped=0.0 2023-04-16 20:09:31,108 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33322.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:09:32,696 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-16 20:09:48,323 INFO [train.py:893] (1/4) Epoch 13, batch 950, loss[loss=0.2294, simple_loss=0.2679, pruned_loss=0.09546, over 13354.00 frames. ], tot_loss[loss=0.2091, simple_loss=0.2596, pruned_loss=0.07929, over 2641389.24 frames. ], batch size: 67, lr: 1.02e-02, grad_scale: 16.0 2023-04-16 20:10:25,882 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.2726, 4.7790, 4.6760, 4.7405, 4.4901, 4.5374, 5.2329, 4.8203], device='cuda:1'), covar=tensor([0.0734, 0.0957, 0.2192, 0.2779, 0.1045, 0.1740, 0.0834, 0.1001], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0323, 0.0416, 0.0428, 0.0252, 0.0320, 0.0377, 0.0296], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 20:10:34,632 INFO [train.py:893] (1/4) Epoch 13, batch 1000, loss[loss=0.1733, simple_loss=0.2356, pruned_loss=0.05549, over 13454.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2574, pruned_loss=0.07848, over 2645588.02 frames. ], batch size: 79, lr: 1.02e-02, grad_scale: 16.0 2023-04-16 20:10:55,537 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.940e+02 2.886e+02 3.416e+02 4.096e+02 6.022e+02, threshold=6.831e+02, percent-clipped=0.0 2023-04-16 20:11:07,510 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33425.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:11:11,463 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.5194, 3.2575, 2.5215, 2.9510, 2.8172, 1.9077, 3.3748, 1.8851], device='cuda:1'), covar=tensor([0.0619, 0.0713, 0.0519, 0.0399, 0.0702, 0.2000, 0.0842, 0.1298], device='cuda:1'), in_proj_covar=tensor([0.0131, 0.0122, 0.0121, 0.0101, 0.0141, 0.0178, 0.0142, 0.0133], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 20:11:21,171 INFO [train.py:893] (1/4) Epoch 13, batch 1050, loss[loss=0.1652, simple_loss=0.2281, pruned_loss=0.05112, over 13368.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2555, pruned_loss=0.07674, over 2650142.53 frames. ], batch size: 73, lr: 1.02e-02, grad_scale: 16.0 2023-04-16 20:11:52,252 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33472.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:12:08,479 INFO [train.py:893] (1/4) Epoch 13, batch 1100, loss[loss=0.1925, simple_loss=0.2565, pruned_loss=0.06429, over 13344.00 frames. ], tot_loss[loss=0.2036, simple_loss=0.2555, pruned_loss=0.07589, over 2654547.90 frames. ], batch size: 118, lr: 1.02e-02, grad_scale: 16.0 2023-04-16 20:12:28,312 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33511.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:12:28,913 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.702e+02 2.943e+02 3.547e+02 4.473e+02 8.261e+02, threshold=7.094e+02, percent-clipped=4.0 2023-04-16 20:12:49,621 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33533.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:12:55,307 INFO [train.py:893] (1/4) Epoch 13, batch 1150, loss[loss=0.2178, simple_loss=0.2671, pruned_loss=0.08421, over 13569.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2556, pruned_loss=0.07552, over 2653570.16 frames. ], batch size: 89, lr: 1.02e-02, grad_scale: 16.0 2023-04-16 20:13:03,092 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6004, 4.1641, 4.1256, 4.1543, 3.9200, 4.0182, 4.5970, 4.1486], device='cuda:1'), covar=tensor([0.0789, 0.1270, 0.2114, 0.2455, 0.1053, 0.1423, 0.0989, 0.1088], device='cuda:1'), in_proj_covar=tensor([0.0284, 0.0315, 0.0402, 0.0410, 0.0245, 0.0310, 0.0368, 0.0287], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 20:13:43,266 INFO [train.py:893] (1/4) Epoch 13, batch 1200, loss[loss=0.1911, simple_loss=0.2481, pruned_loss=0.06701, over 13331.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2564, pruned_loss=0.07545, over 2648308.54 frames. ], batch size: 73, lr: 1.02e-02, grad_scale: 16.0 2023-04-16 20:13:49,573 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5374, 2.5301, 2.8093, 4.1878, 3.6220, 4.1956, 3.2123, 2.5044], device='cuda:1'), covar=tensor([0.0304, 0.1036, 0.0782, 0.0045, 0.0274, 0.0040, 0.0738, 0.0983], device='cuda:1'), in_proj_covar=tensor([0.0112, 0.0152, 0.0159, 0.0084, 0.0113, 0.0080, 0.0164, 0.0145], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0001, 0.0003, 0.0003], device='cuda:1') 2023-04-16 20:13:53,624 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2023-04-16 20:14:03,748 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.965e+02 2.832e+02 3.223e+02 3.875e+02 8.978e+02, threshold=6.447e+02, percent-clipped=1.0 2023-04-16 20:14:09,579 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-16 20:14:13,966 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33622.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:14:22,926 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-16 20:14:29,586 INFO [train.py:893] (1/4) Epoch 13, batch 1250, loss[loss=0.2126, simple_loss=0.2678, pruned_loss=0.07867, over 13371.00 frames. ], tot_loss[loss=0.204, simple_loss=0.2564, pruned_loss=0.07582, over 2649130.71 frames. ], batch size: 113, lr: 1.01e-02, grad_scale: 16.0 2023-04-16 20:14:57,798 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33670.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:15:04,591 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33677.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:15:16,452 INFO [train.py:893] (1/4) Epoch 13, batch 1300, loss[loss=0.1841, simple_loss=0.2456, pruned_loss=0.0613, over 13493.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2581, pruned_loss=0.07652, over 2653751.86 frames. ], batch size: 81, lr: 1.01e-02, grad_scale: 16.0 2023-04-16 20:15:37,998 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.940e+02 3.082e+02 3.569e+02 4.063e+02 6.502e+02, threshold=7.138e+02, percent-clipped=1.0 2023-04-16 20:15:49,979 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33725.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:15:54,146 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8409, 4.3303, 4.1224, 4.0960, 4.1279, 3.9801, 4.4156, 4.4770], device='cuda:1'), covar=tensor([0.0243, 0.0260, 0.0206, 0.0346, 0.0349, 0.0287, 0.0280, 0.0231], device='cuda:1'), in_proj_covar=tensor([0.0113, 0.0173, 0.0132, 0.0155, 0.0123, 0.0168, 0.0116, 0.0144], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 20:16:01,568 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=33738.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 20:16:03,653 INFO [train.py:893] (1/4) Epoch 13, batch 1350, loss[loss=0.2031, simple_loss=0.2504, pruned_loss=0.07786, over 13517.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2595, pruned_loss=0.07745, over 2651502.31 frames. ], batch size: 76, lr: 1.01e-02, grad_scale: 16.0 2023-04-16 20:16:10,064 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-16 20:16:34,615 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33773.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:16:46,723 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-04-16 20:16:50,505 INFO [train.py:893] (1/4) Epoch 13, batch 1400, loss[loss=0.1813, simple_loss=0.2409, pruned_loss=0.06085, over 13503.00 frames. ], tot_loss[loss=0.2062, simple_loss=0.2583, pruned_loss=0.07703, over 2654063.53 frames. ], batch size: 81, lr: 1.01e-02, grad_scale: 16.0 2023-04-16 20:16:56,236 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-04-16 20:17:10,958 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=33811.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:17:11,548 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.842e+02 3.141e+02 3.812e+02 4.224e+02 6.150e+02, threshold=7.624e+02, percent-clipped=0.0 2023-04-16 20:17:26,067 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=33828.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:17:38,098 INFO [train.py:893] (1/4) Epoch 13, batch 1450, loss[loss=0.2021, simple_loss=0.2526, pruned_loss=0.07577, over 13344.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2577, pruned_loss=0.07709, over 2651486.95 frames. ], batch size: 118, lr: 1.01e-02, grad_scale: 16.0 2023-04-16 20:17:53,443 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-04-16 20:17:55,489 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=33859.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:18:01,675 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2380, 4.1091, 4.2254, 2.7873, 4.6664, 4.2731, 4.3130, 4.5391], device='cuda:1'), covar=tensor([0.0207, 0.0116, 0.0123, 0.0935, 0.0115, 0.0233, 0.0129, 0.0083], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0044, 0.0069, 0.0094, 0.0085, 0.0088, 0.0069, 0.0062], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 20:18:14,021 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.3016, 4.7714, 4.7186, 4.7628, 4.4026, 4.6638, 5.2289, 4.6993], device='cuda:1'), covar=tensor([0.0834, 0.1286, 0.2355, 0.2765, 0.1002, 0.1619, 0.0984, 0.1389], device='cuda:1'), in_proj_covar=tensor([0.0290, 0.0320, 0.0405, 0.0419, 0.0249, 0.0314, 0.0371, 0.0291], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 20:18:14,208 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5294, 3.8130, 3.4696, 4.1539, 2.2185, 2.8968, 3.8671, 2.1430], device='cuda:1'), covar=tensor([0.0099, 0.0439, 0.0774, 0.0549, 0.1496, 0.1043, 0.0491, 0.1914], device='cuda:1'), in_proj_covar=tensor([0.0103, 0.0173, 0.0193, 0.0211, 0.0177, 0.0190, 0.0173, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 20:18:24,076 INFO [train.py:893] (1/4) Epoch 13, batch 1500, loss[loss=0.2224, simple_loss=0.2804, pruned_loss=0.08216, over 13589.00 frames. ], tot_loss[loss=0.2056, simple_loss=0.258, pruned_loss=0.07661, over 2650773.93 frames. ], batch size: 89, lr: 1.01e-02, grad_scale: 16.0 2023-04-16 20:18:38,002 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.67 vs. limit=2.0 2023-04-16 20:18:45,825 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.120e+02 3.008e+02 3.356e+02 4.011e+02 6.551e+02, threshold=6.712e+02, percent-clipped=0.0 2023-04-16 20:18:46,112 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4038, 4.6516, 4.5273, 4.5407, 4.4911, 4.9600, 4.6358, 4.5980], device='cuda:1'), covar=tensor([0.0340, 0.0319, 0.0290, 0.0985, 0.0357, 0.0265, 0.0328, 0.0213], device='cuda:1'), in_proj_covar=tensor([0.0135, 0.0134, 0.0148, 0.0242, 0.0148, 0.0165, 0.0143, 0.0103], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 20:19:06,370 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-04-16 20:19:12,094 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=33939.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:19:12,628 INFO [train.py:893] (1/4) Epoch 13, batch 1550, loss[loss=0.2083, simple_loss=0.2658, pruned_loss=0.07542, over 13426.00 frames. ], tot_loss[loss=0.2046, simple_loss=0.2572, pruned_loss=0.076, over 2650012.05 frames. ], batch size: 95, lr: 1.01e-02, grad_scale: 16.0 2023-04-16 20:19:13,756 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.8880, 4.6650, 4.9656, 4.7670, 5.1938, 4.7050, 5.2313, 5.2122], device='cuda:1'), covar=tensor([0.0394, 0.0567, 0.0677, 0.0526, 0.0511, 0.0738, 0.0429, 0.0461], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0259, 0.0257, 0.0189, 0.0369, 0.0297, 0.0228, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0005, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 20:19:16,296 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0590, 4.2414, 2.8947, 4.1445, 4.0584, 2.5660, 3.8061, 2.8191], device='cuda:1'), covar=tensor([0.0254, 0.0230, 0.1101, 0.0308, 0.0255, 0.1168, 0.0385, 0.1343], device='cuda:1'), in_proj_covar=tensor([0.0146, 0.0150, 0.0172, 0.0155, 0.0123, 0.0156, 0.0150, 0.0159], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 20:19:52,829 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=2.73 vs. limit=5.0 2023-04-16 20:19:58,103 INFO [train.py:893] (1/4) Epoch 13, batch 1600, loss[loss=0.187, simple_loss=0.2359, pruned_loss=0.06904, over 13424.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2583, pruned_loss=0.07634, over 2651391.72 frames. ], batch size: 65, lr: 1.01e-02, grad_scale: 16.0 2023-04-16 20:20:12,446 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34000.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:20:22,931 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.987e+02 3.172e+02 3.831e+02 4.874e+02 1.162e+03, threshold=7.661e+02, percent-clipped=7.0 2023-04-16 20:20:43,738 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34033.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 20:20:49,358 INFO [train.py:893] (1/4) Epoch 13, batch 1650, loss[loss=0.2008, simple_loss=0.2473, pruned_loss=0.07713, over 13139.00 frames. ], tot_loss[loss=0.2052, simple_loss=0.2588, pruned_loss=0.07577, over 2653847.70 frames. ], batch size: 58, lr: 1.01e-02, grad_scale: 16.0 2023-04-16 20:20:52,934 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34044.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:21:36,487 INFO [train.py:893] (1/4) Epoch 13, batch 1700, loss[loss=0.1867, simple_loss=0.2456, pruned_loss=0.06386, over 13434.00 frames. ], tot_loss[loss=0.2043, simple_loss=0.2586, pruned_loss=0.07498, over 2660620.64 frames. ], batch size: 95, lr: 1.01e-02, grad_scale: 16.0 2023-04-16 20:21:50,133 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34105.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:21:55,755 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.016e+02 2.745e+02 3.237e+02 3.831e+02 8.000e+02, threshold=6.474e+02, percent-clipped=1.0 2023-04-16 20:22:12,385 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34128.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:22:22,243 INFO [train.py:893] (1/4) Epoch 13, batch 1750, loss[loss=0.2093, simple_loss=0.2686, pruned_loss=0.07506, over 13507.00 frames. ], tot_loss[loss=0.2039, simple_loss=0.2578, pruned_loss=0.07497, over 2657840.93 frames. ], batch size: 98, lr: 1.01e-02, grad_scale: 16.0 2023-04-16 20:22:26,745 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34145.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:22:53,149 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.3815, 2.1982, 2.0689, 2.4073, 1.8703, 2.5182, 2.4035, 2.0723], device='cuda:1'), covar=tensor([0.0074, 0.0168, 0.0133, 0.0130, 0.0192, 0.0117, 0.0161, 0.0161], device='cuda:1'), in_proj_covar=tensor([0.0081, 0.0087, 0.0095, 0.0087, 0.0101, 0.0080, 0.0084, 0.0082], device='cuda:1'), out_proj_covar=tensor([8.6772e-05, 9.6234e-05, 1.0753e-04, 9.6698e-05, 1.1363e-04, 8.7110e-05, 9.4161e-05, 8.8828e-05], device='cuda:1') 2023-04-16 20:22:55,618 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34176.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:23:01,162 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1716, 4.0215, 4.1303, 2.7541, 4.5547, 4.2463, 4.3123, 4.5182], device='cuda:1'), covar=tensor([0.0266, 0.0126, 0.0200, 0.1002, 0.0157, 0.0238, 0.0148, 0.0091], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0045, 0.0070, 0.0095, 0.0087, 0.0089, 0.0069, 0.0062], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 20:23:09,232 INFO [train.py:893] (1/4) Epoch 13, batch 1800, loss[loss=0.2088, simple_loss=0.2575, pruned_loss=0.08004, over 13198.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2563, pruned_loss=0.07399, over 2662806.51 frames. ], batch size: 132, lr: 1.01e-02, grad_scale: 16.0 2023-04-16 20:23:24,300 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34206.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:23:30,365 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.991e+02 2.794e+02 3.346e+02 4.036e+02 7.861e+02, threshold=6.691e+02, percent-clipped=3.0 2023-04-16 20:23:55,756 INFO [train.py:893] (1/4) Epoch 13, batch 1850, loss[loss=0.1701, simple_loss=0.2282, pruned_loss=0.05596, over 13474.00 frames. ], tot_loss[loss=0.2015, simple_loss=0.2557, pruned_loss=0.07372, over 2658401.93 frames. ], batch size: 100, lr: 1.01e-02, grad_scale: 16.0 2023-04-16 20:24:00,696 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-16 20:24:06,048 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9237, 3.8537, 3.3046, 3.6745, 3.1709, 2.0311, 3.8616, 2.0373], device='cuda:1'), covar=tensor([0.0739, 0.0528, 0.0436, 0.0286, 0.0771, 0.2175, 0.0878, 0.1569], device='cuda:1'), in_proj_covar=tensor([0.0134, 0.0123, 0.0121, 0.0102, 0.0144, 0.0180, 0.0145, 0.0134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 20:24:41,996 INFO [train.py:893] (1/4) Epoch 13, batch 1900, loss[loss=0.187, simple_loss=0.2405, pruned_loss=0.06674, over 13343.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2547, pruned_loss=0.07356, over 2660842.26 frames. ], batch size: 67, lr: 1.00e-02, grad_scale: 16.0 2023-04-16 20:24:46,439 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34295.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:25:01,819 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.076e+02 2.981e+02 3.408e+02 4.099e+02 6.680e+02, threshold=6.816e+02, percent-clipped=0.0 2023-04-16 20:25:20,517 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34333.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:25:28,203 INFO [train.py:893] (1/4) Epoch 13, batch 1950, loss[loss=0.193, simple_loss=0.2487, pruned_loss=0.06864, over 13497.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2538, pruned_loss=0.07294, over 2662598.43 frames. ], batch size: 81, lr: 1.00e-02, grad_scale: 16.0 2023-04-16 20:25:44,844 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9809, 4.1217, 3.3643, 2.7618, 2.9070, 2.5754, 4.3431, 2.4779], device='cuda:1'), covar=tensor([0.1366, 0.0294, 0.0791, 0.1659, 0.0646, 0.2457, 0.0175, 0.3128], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0276, 0.0281, 0.0297, 0.0234, 0.0299, 0.0192, 0.0344], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 20:26:06,018 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34381.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:26:13,278 INFO [train.py:893] (1/4) Epoch 13, batch 2000, loss[loss=0.221, simple_loss=0.2746, pruned_loss=0.08368, over 13437.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2571, pruned_loss=0.07482, over 2662337.36 frames. ], batch size: 95, lr: 1.00e-02, grad_scale: 32.0 2023-04-16 20:26:20,698 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-16 20:26:24,518 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34400.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:26:27,090 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7396, 2.5553, 2.2614, 1.5847, 1.4243, 2.2779, 2.1245, 2.7327], device='cuda:1'), covar=tensor([0.0918, 0.0316, 0.0599, 0.1562, 0.0239, 0.0398, 0.0710, 0.0259], device='cuda:1'), in_proj_covar=tensor([0.0144, 0.0115, 0.0104, 0.0188, 0.0095, 0.0135, 0.0142, 0.0113], device='cuda:1'), out_proj_covar=tensor([1.0933e-04, 8.7224e-05, 8.2734e-05, 1.4434e-04, 7.4134e-05, 1.0183e-04, 1.0935e-04, 8.4360e-05], device='cuda:1') 2023-04-16 20:26:35,283 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.214e+02 3.106e+02 3.552e+02 4.151e+02 6.678e+02, threshold=7.105e+02, percent-clipped=0.0 2023-04-16 20:27:01,999 INFO [train.py:893] (1/4) Epoch 13, batch 2050, loss[loss=0.215, simple_loss=0.2614, pruned_loss=0.08432, over 13219.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.259, pruned_loss=0.07606, over 2662581.52 frames. ], batch size: 132, lr: 1.00e-02, grad_scale: 32.0 2023-04-16 20:27:35,599 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5811, 3.6508, 4.2365, 3.0966, 2.8654, 2.9816, 4.4844, 4.5657], device='cuda:1'), covar=tensor([0.1120, 0.1364, 0.0318, 0.1566, 0.1552, 0.1313, 0.0239, 0.0247], device='cuda:1'), in_proj_covar=tensor([0.0227, 0.0241, 0.0175, 0.0212, 0.0207, 0.0171, 0.0179, 0.0149], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 20:27:47,839 INFO [train.py:893] (1/4) Epoch 13, batch 2100, loss[loss=0.1994, simple_loss=0.2517, pruned_loss=0.07354, over 13493.00 frames. ], tot_loss[loss=0.2042, simple_loss=0.2577, pruned_loss=0.07537, over 2660824.50 frames. ], batch size: 70, lr: 1.00e-02, grad_scale: 32.0 2023-04-16 20:27:59,472 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34501.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:28:08,484 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.948e+02 2.768e+02 3.313e+02 3.820e+02 1.138e+03, threshold=6.626e+02, percent-clipped=1.0 2023-04-16 20:28:14,965 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.5110, 2.3068, 2.2140, 2.5716, 1.9199, 2.6907, 2.6003, 2.1766], device='cuda:1'), covar=tensor([0.0096, 0.0159, 0.0140, 0.0118, 0.0207, 0.0117, 0.0192, 0.0195], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0088, 0.0097, 0.0089, 0.0103, 0.0080, 0.0086, 0.0084], device='cuda:1'), out_proj_covar=tensor([8.8781e-05, 9.7081e-05, 1.1058e-04, 9.8484e-05, 1.1580e-04, 8.7308e-05, 9.5778e-05, 9.2081e-05], device='cuda:1') 2023-04-16 20:28:35,139 INFO [train.py:893] (1/4) Epoch 13, batch 2150, loss[loss=0.1932, simple_loss=0.252, pruned_loss=0.06718, over 13234.00 frames. ], tot_loss[loss=0.2026, simple_loss=0.257, pruned_loss=0.07411, over 2661784.81 frames. ], batch size: 132, lr: 1.00e-02, grad_scale: 32.0 2023-04-16 20:29:22,371 INFO [train.py:893] (1/4) Epoch 13, batch 2200, loss[loss=0.1827, simple_loss=0.2415, pruned_loss=0.0619, over 13528.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2556, pruned_loss=0.07311, over 2659361.15 frames. ], batch size: 76, lr: 1.00e-02, grad_scale: 32.0 2023-04-16 20:29:25,183 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34593.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:29:27,678 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34595.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:29:30,224 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2552, 2.4835, 1.7981, 4.0326, 4.5930, 3.3893, 4.5123, 4.2576], device='cuda:1'), covar=tensor([0.0077, 0.0915, 0.1102, 0.0093, 0.0058, 0.0435, 0.0069, 0.0063], device='cuda:1'), in_proj_covar=tensor([0.0065, 0.0082, 0.0090, 0.0070, 0.0056, 0.0075, 0.0049, 0.0065], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 20:29:41,971 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.890e+02 2.826e+02 3.177e+02 3.694e+02 6.051e+02, threshold=6.354e+02, percent-clipped=0.0 2023-04-16 20:29:52,904 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34622.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:29:56,270 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34626.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:30:08,582 INFO [train.py:893] (1/4) Epoch 13, batch 2250, loss[loss=0.1712, simple_loss=0.2289, pruned_loss=0.05677, over 13519.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2544, pruned_loss=0.07312, over 2661354.17 frames. ], batch size: 72, lr: 1.00e-02, grad_scale: 32.0 2023-04-16 20:30:11,257 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34643.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:30:23,144 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34654.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:30:25,919 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.42 vs. limit=5.0 2023-04-16 20:30:49,914 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34683.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 20:30:54,111 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34687.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:30:56,297 INFO [train.py:893] (1/4) Epoch 13, batch 2300, loss[loss=0.21, simple_loss=0.2699, pruned_loss=0.0751, over 13337.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2537, pruned_loss=0.07252, over 2661434.85 frames. ], batch size: 118, lr: 9.99e-03, grad_scale: 16.0 2023-04-16 20:31:05,789 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34700.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:31:14,720 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=34710.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 20:31:17,616 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.020e+02 2.763e+02 3.206e+02 3.873e+02 6.964e+02, threshold=6.413e+02, percent-clipped=1.0 2023-04-16 20:31:22,862 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2100, 4.4103, 4.0867, 4.2264, 4.2665, 4.6307, 4.3556, 4.2816], device='cuda:1'), covar=tensor([0.0271, 0.0263, 0.0352, 0.0935, 0.0324, 0.0260, 0.0310, 0.0302], device='cuda:1'), in_proj_covar=tensor([0.0136, 0.0134, 0.0150, 0.0243, 0.0149, 0.0167, 0.0144, 0.0103], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 20:31:42,628 INFO [train.py:893] (1/4) Epoch 13, batch 2350, loss[loss=0.1461, simple_loss=0.1999, pruned_loss=0.04612, over 7765.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2528, pruned_loss=0.0721, over 2658602.87 frames. ], batch size: 31, lr: 9.99e-03, grad_scale: 16.0 2023-04-16 20:31:50,103 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34748.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:31:54,948 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.0711, 4.9034, 5.1097, 4.8997, 5.3920, 4.9283, 5.4625, 5.3844], device='cuda:1'), covar=tensor([0.0366, 0.0490, 0.0572, 0.0573, 0.0449, 0.0699, 0.0370, 0.0444], device='cuda:1'), in_proj_covar=tensor([0.0242, 0.0260, 0.0258, 0.0191, 0.0370, 0.0298, 0.0226, 0.0258], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0005, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 20:31:56,820 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6071, 3.9792, 3.7553, 4.3151, 2.5827, 3.1706, 4.0850, 2.1097], device='cuda:1'), covar=tensor([0.0082, 0.0453, 0.0742, 0.0480, 0.1434, 0.0983, 0.0542, 0.2040], device='cuda:1'), in_proj_covar=tensor([0.0104, 0.0174, 0.0194, 0.0212, 0.0177, 0.0191, 0.0173, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 20:32:06,567 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-16 20:32:11,840 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=34771.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 20:32:29,628 INFO [train.py:893] (1/4) Epoch 13, batch 2400, loss[loss=0.1784, simple_loss=0.2332, pruned_loss=0.06175, over 13444.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2525, pruned_loss=0.07235, over 2656201.17 frames. ], batch size: 65, lr: 9.98e-03, grad_scale: 16.0 2023-04-16 20:32:39,375 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=34801.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:32:39,692 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.18 vs. limit=5.0 2023-04-16 20:32:47,670 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.08 vs. limit=5.0 2023-04-16 20:32:47,893 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.77 vs. limit=5.0 2023-04-16 20:32:51,406 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.128e+02 2.703e+02 3.191e+02 3.711e+02 8.271e+02, threshold=6.381e+02, percent-clipped=2.0 2023-04-16 20:33:05,137 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1285, 4.3528, 4.1091, 4.1935, 4.1556, 4.5837, 4.3704, 4.2849], device='cuda:1'), covar=tensor([0.0391, 0.0283, 0.0352, 0.0850, 0.0365, 0.0254, 0.0331, 0.0353], device='cuda:1'), in_proj_covar=tensor([0.0136, 0.0135, 0.0150, 0.0243, 0.0150, 0.0166, 0.0144, 0.0104], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 20:33:16,431 INFO [train.py:893] (1/4) Epoch 13, batch 2450, loss[loss=0.18, simple_loss=0.2408, pruned_loss=0.05957, over 13522.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2523, pruned_loss=0.07236, over 2654415.38 frames. ], batch size: 72, lr: 9.97e-03, grad_scale: 16.0 2023-04-16 20:33:17,535 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.2015, 2.1522, 2.3887, 3.7637, 3.3014, 3.8393, 2.8808, 2.1777], device='cuda:1'), covar=tensor([0.0328, 0.0998, 0.0870, 0.0056, 0.0315, 0.0040, 0.0681, 0.1020], device='cuda:1'), in_proj_covar=tensor([0.0113, 0.0150, 0.0159, 0.0086, 0.0113, 0.0081, 0.0163, 0.0145], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0001, 0.0003, 0.0003], device='cuda:1') 2023-04-16 20:33:24,812 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=34849.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:33:39,365 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-04-16 20:34:02,365 INFO [train.py:893] (1/4) Epoch 13, batch 2500, loss[loss=0.2015, simple_loss=0.2552, pruned_loss=0.07393, over 13421.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2529, pruned_loss=0.07266, over 2655793.36 frames. ], batch size: 95, lr: 9.96e-03, grad_scale: 16.0 2023-04-16 20:34:24,105 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.999e+02 2.919e+02 3.279e+02 3.793e+02 5.690e+02, threshold=6.558e+02, percent-clipped=0.0 2023-04-16 20:34:49,061 INFO [train.py:893] (1/4) Epoch 13, batch 2550, loss[loss=0.2034, simple_loss=0.2638, pruned_loss=0.07151, over 13495.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2523, pruned_loss=0.07221, over 2659170.84 frames. ], batch size: 93, lr: 9.96e-03, grad_scale: 16.0 2023-04-16 20:34:58,786 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34949.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:35:03,856 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0934, 4.5121, 4.2592, 4.2941, 4.2067, 4.0873, 4.5879, 4.6103], device='cuda:1'), covar=tensor([0.0205, 0.0241, 0.0240, 0.0350, 0.0386, 0.0310, 0.0280, 0.0248], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0179, 0.0137, 0.0162, 0.0128, 0.0176, 0.0119, 0.0149], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 20:35:15,853 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-16 20:35:26,057 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34978.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 20:35:29,388 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=34982.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:35:35,930 INFO [train.py:893] (1/4) Epoch 13, batch 2600, loss[loss=0.2123, simple_loss=0.2651, pruned_loss=0.07976, over 13529.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2529, pruned_loss=0.0728, over 2660068.14 frames. ], batch size: 85, lr: 9.95e-03, grad_scale: 16.0 2023-04-16 20:35:58,855 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.271e+02 3.103e+02 3.615e+02 4.155e+02 9.868e+02, threshold=7.230e+02, percent-clipped=6.0 2023-04-16 20:36:11,072 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.8461, 5.3421, 5.4464, 5.5062, 5.2002, 5.1541, 5.8937, 5.4747], device='cuda:1'), covar=tensor([0.0732, 0.0959, 0.1501, 0.2109, 0.0762, 0.1417, 0.0730, 0.0797], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0326, 0.0415, 0.0427, 0.0252, 0.0318, 0.0375, 0.0302], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 20:36:18,261 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.68 vs. limit=5.0 2023-04-16 20:36:19,372 INFO [train.py:893] (1/4) Epoch 13, batch 2650, loss[loss=0.2272, simple_loss=0.2743, pruned_loss=0.09006, over 13084.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2539, pruned_loss=0.07378, over 2653431.18 frames. ], batch size: 142, lr: 9.94e-03, grad_scale: 16.0 2023-04-16 20:36:24,592 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 2023-04-16 20:36:35,192 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9160, 4.0583, 3.3652, 2.8345, 2.7417, 2.4038, 4.2469, 2.3609], device='cuda:1'), covar=tensor([0.1475, 0.0314, 0.0782, 0.1646, 0.0728, 0.2922, 0.0193, 0.3494], device='cuda:1'), in_proj_covar=tensor([0.0294, 0.0271, 0.0277, 0.0292, 0.0232, 0.0297, 0.0189, 0.0340], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 20:36:40,112 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35066.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 20:37:18,350 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 20:37:28,634 INFO [train.py:893] (1/4) Epoch 14, batch 0, loss[loss=0.2513, simple_loss=0.2839, pruned_loss=0.1094, over 13448.00 frames. ], tot_loss[loss=0.2513, simple_loss=0.2839, pruned_loss=0.1094, over 13448.00 frames. ], batch size: 100, lr: 9.58e-03, grad_scale: 16.0 2023-04-16 20:37:28,634 INFO [train.py:918] (1/4) Computing validation loss 2023-04-16 20:37:36,023 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1610, 4.0182, 4.2463, 2.8568, 4.5618, 4.2355, 4.1790, 4.4665], device='cuda:1'), covar=tensor([0.0255, 0.0160, 0.0144, 0.0937, 0.0128, 0.0227, 0.0144, 0.0102], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0046, 0.0069, 0.0094, 0.0085, 0.0088, 0.0069, 0.0062], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 20:37:36,977 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.7078, 4.1859, 4.3424, 4.3885, 4.2579, 4.1976, 4.7128, 4.2815], device='cuda:1'), covar=tensor([0.0847, 0.1265, 0.2034, 0.2470, 0.0966, 0.1592, 0.1022, 0.1185], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0330, 0.0418, 0.0427, 0.0253, 0.0319, 0.0378, 0.0303], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 20:37:39,002 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.2342, 4.7129, 4.7927, 4.7374, 4.6166, 4.7483, 5.2046, 4.7732], device='cuda:1'), covar=tensor([0.0830, 0.1231, 0.2189, 0.2837, 0.0914, 0.1581, 0.0963, 0.1054], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0330, 0.0418, 0.0427, 0.0253, 0.0319, 0.0378, 0.0303], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 20:37:50,575 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7675, 3.5509, 3.7500, 2.6168, 3.8877, 3.7844, 3.7787, 3.8844], device='cuda:1'), covar=tensor([0.0214, 0.0142, 0.0143, 0.0947, 0.0132, 0.0190, 0.0101, 0.0106], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0046, 0.0069, 0.0094, 0.0085, 0.0088, 0.0069, 0.0062], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 20:37:51,599 INFO [train.py:927] (1/4) Epoch 14, validation: loss=0.1497, simple_loss=0.207, pruned_loss=0.04622, over 2446609.00 frames. 2023-04-16 20:37:51,600 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12829MB 2023-04-16 20:37:54,425 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35092.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:37:54,490 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.6615, 2.4574, 2.1954, 1.5488, 1.3263, 2.1007, 2.1039, 2.6899], device='cuda:1'), covar=tensor([0.1049, 0.0376, 0.0734, 0.1806, 0.0226, 0.0411, 0.0745, 0.0337], device='cuda:1'), in_proj_covar=tensor([0.0148, 0.0119, 0.0105, 0.0190, 0.0096, 0.0135, 0.0146, 0.0116], device='cuda:1'), out_proj_covar=tensor([1.1225e-04, 8.9865e-05, 8.3769e-05, 1.4588e-04, 7.5156e-05, 1.0251e-04, 1.1220e-04, 8.6779e-05], device='cuda:1') 2023-04-16 20:38:08,073 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 2023-04-16 20:38:13,111 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.840e+02 2.859e+02 3.403e+02 4.075e+02 8.833e+02, threshold=6.807e+02, percent-clipped=1.0 2023-04-16 20:38:37,354 INFO [train.py:893] (1/4) Epoch 14, batch 50, loss[loss=0.1909, simple_loss=0.2474, pruned_loss=0.06723, over 13545.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2484, pruned_loss=0.07307, over 603116.48 frames. ], batch size: 98, lr: 9.57e-03, grad_scale: 16.0 2023-04-16 20:38:51,666 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35153.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:38:57,608 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4213, 3.7693, 3.5499, 4.1180, 2.0837, 3.0780, 3.9290, 2.1077], device='cuda:1'), covar=tensor([0.0119, 0.0451, 0.0712, 0.0537, 0.1667, 0.0935, 0.0548, 0.1915], device='cuda:1'), in_proj_covar=tensor([0.0106, 0.0175, 0.0195, 0.0214, 0.0177, 0.0190, 0.0173, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 20:39:01,965 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-16 20:39:01,966 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-16 20:39:01,966 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-16 20:39:01,974 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-16 20:39:01,989 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-16 20:39:02,716 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-16 20:39:02,747 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-16 20:39:08,034 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3978, 3.3486, 3.9655, 2.8668, 2.6521, 2.8063, 4.1891, 4.3865], device='cuda:1'), covar=tensor([0.1118, 0.1422, 0.0422, 0.1506, 0.1480, 0.1359, 0.0263, 0.0215], device='cuda:1'), in_proj_covar=tensor([0.0226, 0.0244, 0.0177, 0.0212, 0.0207, 0.0172, 0.0179, 0.0149], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 20:39:24,426 INFO [train.py:893] (1/4) Epoch 14, batch 100, loss[loss=0.2204, simple_loss=0.2652, pruned_loss=0.08784, over 13056.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2507, pruned_loss=0.07458, over 1061097.48 frames. ], batch size: 142, lr: 9.56e-03, grad_scale: 16.0 2023-04-16 20:39:28,111 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.2130, 2.4829, 4.2963, 3.9248, 4.0482, 3.4157, 3.9365, 3.0498], device='cuda:1'), covar=tensor([0.1870, 0.1220, 0.0072, 0.0220, 0.0157, 0.0493, 0.0192, 0.0678], device='cuda:1'), in_proj_covar=tensor([0.0204, 0.0178, 0.0107, 0.0117, 0.0121, 0.0169, 0.0128, 0.0170], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 20:39:46,298 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.998e+02 3.037e+02 3.469e+02 4.318e+02 9.058e+02, threshold=6.938e+02, percent-clipped=1.0 2023-04-16 20:40:03,033 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7663, 3.6015, 3.7156, 2.1938, 3.9463, 3.8453, 3.7902, 3.9469], device='cuda:1'), covar=tensor([0.0193, 0.0138, 0.0130, 0.1160, 0.0129, 0.0189, 0.0111, 0.0098], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0045, 0.0069, 0.0094, 0.0085, 0.0088, 0.0068, 0.0062], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 20:40:11,163 INFO [train.py:893] (1/4) Epoch 14, batch 150, loss[loss=0.2159, simple_loss=0.2792, pruned_loss=0.07623, over 13429.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2554, pruned_loss=0.07717, over 1414634.55 frames. ], batch size: 95, lr: 9.56e-03, grad_scale: 16.0 2023-04-16 20:40:21,417 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35249.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:40:47,749 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35278.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 20:40:52,023 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35282.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:40:58,491 INFO [train.py:893] (1/4) Epoch 14, batch 200, loss[loss=0.233, simple_loss=0.2851, pruned_loss=0.09044, over 13480.00 frames. ], tot_loss[loss=0.2068, simple_loss=0.2574, pruned_loss=0.07814, over 1686465.77 frames. ], batch size: 81, lr: 9.55e-03, grad_scale: 16.0 2023-04-16 20:41:05,503 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35297.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:41:20,379 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.003e+02 2.883e+02 3.259e+02 4.058e+02 6.702e+02, threshold=6.518e+02, percent-clipped=0.0 2023-04-16 20:41:33,090 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35326.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:41:33,180 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9187, 4.1861, 3.9863, 3.9696, 3.9841, 4.3425, 4.1446, 3.9428], device='cuda:1'), covar=tensor([0.0286, 0.0308, 0.0311, 0.0963, 0.0268, 0.0238, 0.0266, 0.0460], device='cuda:1'), in_proj_covar=tensor([0.0137, 0.0136, 0.0151, 0.0247, 0.0152, 0.0168, 0.0147, 0.0104], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 20:41:36,482 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35330.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:41:44,604 INFO [train.py:893] (1/4) Epoch 14, batch 250, loss[loss=0.2083, simple_loss=0.2649, pruned_loss=0.07585, over 13481.00 frames. ], tot_loss[loss=0.2061, simple_loss=0.2573, pruned_loss=0.07747, over 1899719.39 frames. ], batch size: 79, lr: 9.54e-03, grad_scale: 16.0 2023-04-16 20:41:51,501 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35345.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 20:42:10,645 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35366.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 20:42:18,087 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35374.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:42:31,805 INFO [train.py:893] (1/4) Epoch 14, batch 300, loss[loss=0.2339, simple_loss=0.2858, pruned_loss=0.09102, over 13485.00 frames. ], tot_loss[loss=0.2067, simple_loss=0.2582, pruned_loss=0.07763, over 2067999.42 frames. ], batch size: 70, lr: 9.54e-03, grad_scale: 16.0 2023-04-16 20:42:49,145 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35406.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 20:42:49,349 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.77 vs. limit=2.0 2023-04-16 20:42:54,695 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.158e+02 3.002e+02 3.392e+02 4.204e+02 7.144e+02, threshold=6.784e+02, percent-clipped=2.0 2023-04-16 20:42:56,405 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35414.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 20:43:15,059 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35435.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:43:19,558 INFO [train.py:893] (1/4) Epoch 14, batch 350, loss[loss=0.1973, simple_loss=0.2465, pruned_loss=0.07404, over 13517.00 frames. ], tot_loss[loss=0.2077, simple_loss=0.2594, pruned_loss=0.07801, over 2200778.14 frames. ], batch size: 76, lr: 9.53e-03, grad_scale: 16.0 2023-04-16 20:43:20,709 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7612, 3.7143, 2.8045, 3.2583, 2.9975, 1.8404, 3.6251, 1.8977], device='cuda:1'), covar=tensor([0.0730, 0.0363, 0.0500, 0.0334, 0.0660, 0.2090, 0.0871, 0.1393], device='cuda:1'), in_proj_covar=tensor([0.0131, 0.0123, 0.0121, 0.0102, 0.0138, 0.0174, 0.0144, 0.0131], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 20:43:27,841 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35448.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:43:57,035 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0070, 3.8241, 3.9734, 2.3964, 4.4003, 4.0636, 4.0084, 4.3610], device='cuda:1'), covar=tensor([0.0235, 0.0186, 0.0127, 0.1252, 0.0144, 0.0236, 0.0168, 0.0096], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0045, 0.0069, 0.0095, 0.0085, 0.0089, 0.0068, 0.0062], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 20:44:06,002 INFO [train.py:893] (1/4) Epoch 14, batch 400, loss[loss=0.1844, simple_loss=0.2499, pruned_loss=0.05949, over 13556.00 frames. ], tot_loss[loss=0.206, simple_loss=0.2584, pruned_loss=0.07684, over 2303495.46 frames. ], batch size: 89, lr: 9.52e-03, grad_scale: 16.0 2023-04-16 20:44:28,068 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.910e+02 3.027e+02 3.448e+02 3.927e+02 6.855e+02, threshold=6.896e+02, percent-clipped=1.0 2023-04-16 20:44:42,364 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.2794, 1.9992, 2.3464, 3.7772, 3.4212, 3.8018, 2.9142, 2.2438], device='cuda:1'), covar=tensor([0.0282, 0.1087, 0.0878, 0.0046, 0.0239, 0.0044, 0.0673, 0.0978], device='cuda:1'), in_proj_covar=tensor([0.0110, 0.0146, 0.0154, 0.0081, 0.0108, 0.0079, 0.0158, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0001, 0.0003, 0.0002], device='cuda:1') 2023-04-16 20:44:43,211 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35530.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:44:44,925 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7512, 3.6904, 4.3936, 3.1636, 2.9823, 3.0523, 4.6243, 4.7848], device='cuda:1'), covar=tensor([0.0981, 0.1356, 0.0296, 0.1394, 0.1376, 0.1235, 0.0260, 0.0185], device='cuda:1'), in_proj_covar=tensor([0.0225, 0.0244, 0.0176, 0.0212, 0.0208, 0.0171, 0.0178, 0.0148], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 20:44:52,038 INFO [train.py:893] (1/4) Epoch 14, batch 450, loss[loss=0.1765, simple_loss=0.2247, pruned_loss=0.06414, over 12426.00 frames. ], tot_loss[loss=0.2072, simple_loss=0.2597, pruned_loss=0.0774, over 2383581.81 frames. ], batch size: 51, lr: 9.52e-03, grad_scale: 16.0 2023-04-16 20:44:55,674 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6497, 2.4912, 2.5568, 4.3165, 3.8143, 4.3448, 3.2987, 2.5990], device='cuda:1'), covar=tensor([0.0273, 0.1101, 0.0947, 0.0032, 0.0202, 0.0038, 0.0649, 0.0970], device='cuda:1'), in_proj_covar=tensor([0.0111, 0.0147, 0.0156, 0.0082, 0.0109, 0.0080, 0.0159, 0.0144], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0001, 0.0003, 0.0003], device='cuda:1') 2023-04-16 20:45:16,913 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-16 20:45:26,822 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2609, 2.6727, 2.2259, 4.1362, 4.7411, 3.4631, 4.6135, 4.3702], device='cuda:1'), covar=tensor([0.0093, 0.0813, 0.0978, 0.0105, 0.0056, 0.0441, 0.0075, 0.0070], device='cuda:1'), in_proj_covar=tensor([0.0066, 0.0085, 0.0091, 0.0072, 0.0056, 0.0075, 0.0050, 0.0065], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 20:45:38,292 INFO [train.py:893] (1/4) Epoch 14, batch 500, loss[loss=0.2028, simple_loss=0.2588, pruned_loss=0.07339, over 13431.00 frames. ], tot_loss[loss=0.2058, simple_loss=0.2587, pruned_loss=0.07642, over 2448047.49 frames. ], batch size: 106, lr: 9.51e-03, grad_scale: 16.0 2023-04-16 20:45:40,147 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35591.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:46:01,679 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.031e+02 2.935e+02 3.425e+02 4.284e+02 7.521e+02, threshold=6.851e+02, percent-clipped=2.0 2023-04-16 20:46:26,155 INFO [train.py:893] (1/4) Epoch 14, batch 550, loss[loss=0.195, simple_loss=0.2531, pruned_loss=0.06844, over 13278.00 frames. ], tot_loss[loss=0.2045, simple_loss=0.2578, pruned_loss=0.07561, over 2495091.03 frames. ], batch size: 124, lr: 9.50e-03, grad_scale: 16.0 2023-04-16 20:46:46,111 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 2023-04-16 20:47:12,172 INFO [train.py:893] (1/4) Epoch 14, batch 600, loss[loss=0.1921, simple_loss=0.2367, pruned_loss=0.07377, over 13180.00 frames. ], tot_loss[loss=0.2041, simple_loss=0.2568, pruned_loss=0.07567, over 2526355.40 frames. ], batch size: 58, lr: 9.50e-03, grad_scale: 16.0 2023-04-16 20:47:16,580 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.6897, 2.5297, 2.2391, 1.6168, 1.4661, 2.2534, 2.1635, 2.8363], device='cuda:1'), covar=tensor([0.1139, 0.0354, 0.0808, 0.1936, 0.0300, 0.0524, 0.0861, 0.0217], device='cuda:1'), in_proj_covar=tensor([0.0147, 0.0116, 0.0104, 0.0187, 0.0096, 0.0135, 0.0143, 0.0113], device='cuda:1'), out_proj_covar=tensor([1.1161e-04, 8.7685e-05, 8.3245e-05, 1.4298e-04, 7.4968e-05, 1.0235e-04, 1.0989e-04, 8.4457e-05], device='cuda:1') 2023-04-16 20:47:24,582 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35701.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 20:47:28,110 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-16 20:47:35,217 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.988e+02 2.727e+02 3.202e+02 3.710e+02 6.936e+02, threshold=6.404e+02, percent-clipped=1.0 2023-04-16 20:47:51,291 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35730.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:48:00,072 INFO [train.py:893] (1/4) Epoch 14, batch 650, loss[loss=0.1913, simple_loss=0.2485, pruned_loss=0.06701, over 13404.00 frames. ], tot_loss[loss=0.2025, simple_loss=0.2555, pruned_loss=0.07476, over 2558566.95 frames. ], batch size: 113, lr: 9.49e-03, grad_scale: 16.0 2023-04-16 20:48:07,809 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=35748.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:48:30,042 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35771.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:48:46,709 INFO [train.py:893] (1/4) Epoch 14, batch 700, loss[loss=0.1996, simple_loss=0.2543, pruned_loss=0.07249, over 13421.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2539, pruned_loss=0.07396, over 2580354.90 frames. ], batch size: 88, lr: 9.48e-03, grad_scale: 16.0 2023-04-16 20:48:53,369 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=35796.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:49:07,405 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=35811.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:49:08,663 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.910e+02 3.025e+02 3.522e+02 4.201e+02 9.258e+02, threshold=7.044e+02, percent-clipped=2.0 2023-04-16 20:49:09,302 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2023-04-16 20:49:27,500 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35832.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:49:33,922 INFO [train.py:893] (1/4) Epoch 14, batch 750, loss[loss=0.2158, simple_loss=0.2642, pruned_loss=0.08373, over 13260.00 frames. ], tot_loss[loss=0.2014, simple_loss=0.2539, pruned_loss=0.07442, over 2597743.73 frames. ], batch size: 124, lr: 9.48e-03, grad_scale: 16.0 2023-04-16 20:49:39,891 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7682, 3.6144, 3.7448, 2.2061, 3.9725, 3.7357, 3.7462, 3.9147], device='cuda:1'), covar=tensor([0.0211, 0.0142, 0.0146, 0.1259, 0.0133, 0.0255, 0.0139, 0.0105], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0046, 0.0070, 0.0095, 0.0087, 0.0090, 0.0069, 0.0062], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 20:49:53,516 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2023-04-16 20:50:04,834 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=35872.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:50:17,045 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=35886.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:50:20,145 INFO [train.py:893] (1/4) Epoch 14, batch 800, loss[loss=0.2063, simple_loss=0.2588, pruned_loss=0.07691, over 13542.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2556, pruned_loss=0.07518, over 2609830.87 frames. ], batch size: 78, lr: 9.47e-03, grad_scale: 16.0 2023-04-16 20:50:41,653 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.031e+02 2.971e+02 3.594e+02 4.355e+02 9.749e+02, threshold=7.188e+02, percent-clipped=3.0 2023-04-16 20:50:50,991 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5428, 2.3211, 2.9819, 4.3337, 3.7072, 4.3313, 3.4200, 2.5809], device='cuda:1'), covar=tensor([0.0221, 0.1069, 0.0666, 0.0035, 0.0233, 0.0033, 0.0551, 0.0895], device='cuda:1'), in_proj_covar=tensor([0.0111, 0.0149, 0.0158, 0.0084, 0.0111, 0.0079, 0.0162, 0.0147], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0001, 0.0003, 0.0003], device='cuda:1') 2023-04-16 20:51:04,913 INFO [train.py:893] (1/4) Epoch 14, batch 850, loss[loss=0.2083, simple_loss=0.2589, pruned_loss=0.0788, over 13443.00 frames. ], tot_loss[loss=0.2047, simple_loss=0.2573, pruned_loss=0.07601, over 2617536.98 frames. ], batch size: 106, lr: 9.46e-03, grad_scale: 16.0 2023-04-16 20:51:20,817 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6991, 4.5852, 4.7764, 4.5923, 5.0282, 4.5363, 5.0263, 5.0287], device='cuda:1'), covar=tensor([0.0408, 0.0536, 0.0652, 0.0599, 0.0513, 0.0808, 0.0504, 0.0456], device='cuda:1'), in_proj_covar=tensor([0.0248, 0.0266, 0.0259, 0.0193, 0.0373, 0.0306, 0.0233, 0.0258], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0005, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 20:51:51,610 INFO [train.py:893] (1/4) Epoch 14, batch 900, loss[loss=0.1763, simple_loss=0.2253, pruned_loss=0.06362, over 13431.00 frames. ], tot_loss[loss=0.2054, simple_loss=0.2575, pruned_loss=0.07668, over 2625588.14 frames. ], batch size: 65, lr: 9.46e-03, grad_scale: 8.0 2023-04-16 20:52:06,915 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36001.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 20:52:19,158 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.226e+02 2.907e+02 3.358e+02 4.155e+02 9.448e+02, threshold=6.715e+02, percent-clipped=1.0 2023-04-16 20:52:25,872 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-16 20:52:33,440 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36030.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:52:42,246 INFO [train.py:893] (1/4) Epoch 14, batch 950, loss[loss=0.1807, simple_loss=0.2397, pruned_loss=0.06082, over 13537.00 frames. ], tot_loss[loss=0.2057, simple_loss=0.2569, pruned_loss=0.07726, over 2623957.78 frames. ], batch size: 76, lr: 9.45e-03, grad_scale: 8.0 2023-04-16 20:52:51,403 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36049.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 20:52:56,318 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4720, 4.7408, 4.4957, 4.5182, 4.5415, 4.9415, 4.7482, 4.6410], device='cuda:1'), covar=tensor([0.0296, 0.0339, 0.0307, 0.0926, 0.0290, 0.0233, 0.0299, 0.0188], device='cuda:1'), in_proj_covar=tensor([0.0138, 0.0137, 0.0152, 0.0250, 0.0153, 0.0168, 0.0150, 0.0103], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 20:52:58,380 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.77 vs. limit=2.0 2023-04-16 20:53:18,984 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36078.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:53:28,924 INFO [train.py:893] (1/4) Epoch 14, batch 1000, loss[loss=0.2204, simple_loss=0.2668, pruned_loss=0.08702, over 13243.00 frames. ], tot_loss[loss=0.2033, simple_loss=0.2548, pruned_loss=0.07589, over 2627859.43 frames. ], batch size: 124, lr: 9.44e-03, grad_scale: 8.0 2023-04-16 20:53:52,248 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.888e+02 2.953e+02 3.397e+02 4.421e+02 7.418e+02, threshold=6.794e+02, percent-clipped=2.0 2023-04-16 20:54:04,107 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36127.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:54:15,577 INFO [train.py:893] (1/4) Epoch 14, batch 1050, loss[loss=0.1771, simple_loss=0.2342, pruned_loss=0.05998, over 13549.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2526, pruned_loss=0.07398, over 2637911.18 frames. ], batch size: 83, lr: 9.44e-03, grad_scale: 8.0 2023-04-16 20:54:42,535 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36167.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:54:55,707 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36181.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:54:59,907 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36186.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:55:02,051 INFO [train.py:893] (1/4) Epoch 14, batch 1100, loss[loss=0.2073, simple_loss=0.2577, pruned_loss=0.07842, over 13029.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2531, pruned_loss=0.0738, over 2642487.22 frames. ], batch size: 142, lr: 9.43e-03, grad_scale: 8.0 2023-04-16 20:55:21,626 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8240, 2.3314, 2.2408, 2.8353, 2.1930, 2.9088, 2.6935, 2.3614], device='cuda:1'), covar=tensor([0.0085, 0.0183, 0.0155, 0.0131, 0.0211, 0.0110, 0.0192, 0.0163], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0089, 0.0097, 0.0091, 0.0105, 0.0081, 0.0086, 0.0085], device='cuda:1'), out_proj_covar=tensor([8.7876e-05, 9.8451e-05, 1.0948e-04, 1.0018e-04, 1.1768e-04, 8.7873e-05, 9.4681e-05, 9.2460e-05], device='cuda:1') 2023-04-16 20:55:26,409 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.646e+02 2.611e+02 3.323e+02 3.965e+02 5.525e+02, threshold=6.646e+02, percent-clipped=0.0 2023-04-16 20:55:45,657 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36234.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:55:49,263 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3619, 3.8747, 3.6416, 4.0255, 2.2138, 2.9958, 3.8879, 2.1377], device='cuda:1'), covar=tensor([0.0123, 0.0391, 0.0669, 0.0548, 0.1539, 0.0991, 0.0485, 0.1976], device='cuda:1'), in_proj_covar=tensor([0.0106, 0.0176, 0.0195, 0.0214, 0.0178, 0.0192, 0.0173, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 20:55:49,702 INFO [train.py:893] (1/4) Epoch 14, batch 1150, loss[loss=0.1966, simple_loss=0.2581, pruned_loss=0.06749, over 13526.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2527, pruned_loss=0.07305, over 2645629.32 frames. ], batch size: 91, lr: 9.43e-03, grad_scale: 8.0 2023-04-16 20:55:53,401 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36242.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:56:05,428 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9234, 1.9046, 3.7711, 3.6478, 3.6609, 2.8619, 3.4296, 2.8708], device='cuda:1'), covar=tensor([0.1977, 0.1546, 0.0109, 0.0145, 0.0164, 0.0740, 0.0238, 0.0719], device='cuda:1'), in_proj_covar=tensor([0.0208, 0.0182, 0.0110, 0.0117, 0.0121, 0.0173, 0.0130, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 20:56:36,373 INFO [train.py:893] (1/4) Epoch 14, batch 1200, loss[loss=0.2266, simple_loss=0.2773, pruned_loss=0.08801, over 13419.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2524, pruned_loss=0.07225, over 2645904.34 frames. ], batch size: 95, lr: 9.42e-03, grad_scale: 8.0 2023-04-16 20:56:59,636 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.943e+02 3.077e+02 3.534e+02 4.288e+02 1.109e+03, threshold=7.069e+02, percent-clipped=3.0 2023-04-16 20:57:01,443 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-16 20:57:04,220 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.9751, 4.8072, 5.0595, 4.8470, 5.3305, 4.8233, 5.3921, 5.3103], device='cuda:1'), covar=tensor([0.0386, 0.0666, 0.0658, 0.0511, 0.0546, 0.0868, 0.0450, 0.0542], device='cuda:1'), in_proj_covar=tensor([0.0245, 0.0263, 0.0257, 0.0192, 0.0368, 0.0301, 0.0231, 0.0256], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0005, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 20:57:14,248 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-16 20:57:23,279 INFO [train.py:893] (1/4) Epoch 14, batch 1250, loss[loss=0.1928, simple_loss=0.2545, pruned_loss=0.06556, over 13354.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2533, pruned_loss=0.07311, over 2646632.54 frames. ], batch size: 109, lr: 9.41e-03, grad_scale: 8.0 2023-04-16 20:57:34,125 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.64 vs. limit=2.0 2023-04-16 20:57:35,688 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1373, 3.9147, 3.2976, 3.7816, 2.9931, 2.2670, 3.9084, 2.0161], device='cuda:1'), covar=tensor([0.0589, 0.0386, 0.0375, 0.0226, 0.0749, 0.1812, 0.0813, 0.1499], device='cuda:1'), in_proj_covar=tensor([0.0131, 0.0124, 0.0120, 0.0103, 0.0140, 0.0176, 0.0146, 0.0132], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 20:58:10,726 INFO [train.py:893] (1/4) Epoch 14, batch 1300, loss[loss=0.195, simple_loss=0.2537, pruned_loss=0.06813, over 13478.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2549, pruned_loss=0.07362, over 2650443.01 frames. ], batch size: 79, lr: 9.41e-03, grad_scale: 8.0 2023-04-16 20:58:12,604 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3788, 4.1901, 4.4095, 4.3002, 4.6032, 4.2186, 4.6416, 4.5961], device='cuda:1'), covar=tensor([0.0384, 0.0629, 0.0624, 0.0542, 0.0585, 0.0762, 0.0439, 0.0455], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0260, 0.0255, 0.0192, 0.0366, 0.0297, 0.0230, 0.0254], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 20:58:34,169 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.990e+02 3.063e+02 3.435e+02 4.051e+02 8.324e+02, threshold=6.871e+02, percent-clipped=1.0 2023-04-16 20:58:47,698 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36427.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:58:57,487 INFO [train.py:893] (1/4) Epoch 14, batch 1350, loss[loss=0.1753, simple_loss=0.2283, pruned_loss=0.06119, over 13411.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2547, pruned_loss=0.07321, over 2653058.09 frames. ], batch size: 65, lr: 9.40e-03, grad_scale: 8.0 2023-04-16 20:59:16,240 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-04-16 20:59:17,633 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1671, 4.6853, 4.4499, 4.3990, 4.4540, 4.2590, 4.7067, 4.7494], device='cuda:1'), covar=tensor([0.0198, 0.0181, 0.0178, 0.0291, 0.0219, 0.0255, 0.0266, 0.0169], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0178, 0.0139, 0.0163, 0.0128, 0.0176, 0.0119, 0.0150], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 20:59:24,289 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36467.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:59:30,913 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36475.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 20:59:45,702 INFO [train.py:893] (1/4) Epoch 14, batch 1400, loss[loss=0.1998, simple_loss=0.2542, pruned_loss=0.07267, over 13388.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2539, pruned_loss=0.07276, over 2655363.62 frames. ], batch size: 113, lr: 9.39e-03, grad_scale: 8.0 2023-04-16 21:00:09,067 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.735e+02 2.745e+02 3.192e+02 3.647e+02 6.550e+02, threshold=6.384e+02, percent-clipped=0.0 2023-04-16 21:00:10,036 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36515.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:00:29,136 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36537.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:00:30,576 INFO [train.py:893] (1/4) Epoch 14, batch 1450, loss[loss=0.1795, simple_loss=0.2296, pruned_loss=0.06472, over 13165.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2535, pruned_loss=0.07285, over 2656998.81 frames. ], batch size: 58, lr: 9.39e-03, grad_scale: 8.0 2023-04-16 21:00:43,413 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1828, 3.9338, 4.1471, 2.6696, 4.5422, 4.2240, 4.2821, 4.5142], device='cuda:1'), covar=tensor([0.0238, 0.0154, 0.0135, 0.1079, 0.0154, 0.0264, 0.0132, 0.0091], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0045, 0.0070, 0.0094, 0.0087, 0.0090, 0.0069, 0.0062], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 21:01:17,371 INFO [train.py:893] (1/4) Epoch 14, batch 1500, loss[loss=0.2184, simple_loss=0.2757, pruned_loss=0.08057, over 13483.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2536, pruned_loss=0.07275, over 2658092.89 frames. ], batch size: 81, lr: 9.38e-03, grad_scale: 8.0 2023-04-16 21:01:32,348 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36605.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:01:35,835 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36608.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:01:39,129 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36612.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:01:40,470 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.986e+02 2.902e+02 3.279e+02 4.029e+02 7.502e+02, threshold=6.558e+02, percent-clipped=1.0 2023-04-16 21:02:04,667 INFO [train.py:893] (1/4) Epoch 14, batch 1550, loss[loss=0.1962, simple_loss=0.2576, pruned_loss=0.06739, over 13462.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2536, pruned_loss=0.07273, over 2659299.27 frames. ], batch size: 106, lr: 9.37e-03, grad_scale: 8.0 2023-04-16 21:02:07,731 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-04-16 21:02:28,538 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36666.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:02:33,410 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36669.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:02:36,808 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36673.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:02:45,019 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.3559, 3.0751, 2.8593, 2.1974, 2.0648, 2.6503, 2.8992, 3.3592], device='cuda:1'), covar=tensor([0.0866, 0.0257, 0.0575, 0.1219, 0.0586, 0.0459, 0.0531, 0.0221], device='cuda:1'), in_proj_covar=tensor([0.0152, 0.0119, 0.0106, 0.0194, 0.0100, 0.0140, 0.0148, 0.0115], device='cuda:1'), out_proj_covar=tensor([1.1453e-04, 8.9383e-05, 8.4928e-05, 1.4843e-04, 7.8121e-05, 1.0595e-04, 1.1397e-04, 8.5670e-05], device='cuda:1') 2023-04-16 21:02:48,336 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.5364, 5.0942, 4.9266, 5.0579, 4.8067, 4.8899, 5.5059, 5.0366], device='cuda:1'), covar=tensor([0.0655, 0.1044, 0.2222, 0.2144, 0.0827, 0.1396, 0.0788, 0.1085], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0337, 0.0430, 0.0435, 0.0257, 0.0321, 0.0384, 0.0306], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 21:02:50,545 INFO [train.py:893] (1/4) Epoch 14, batch 1600, loss[loss=0.209, simple_loss=0.2638, pruned_loss=0.07714, over 13240.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2544, pruned_loss=0.07306, over 2656582.87 frames. ], batch size: 132, lr: 9.37e-03, grad_scale: 8.0 2023-04-16 21:02:56,746 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9578, 4.0780, 2.7406, 3.8652, 3.8791, 2.5266, 3.5142, 2.5642], device='cuda:1'), covar=tensor([0.0290, 0.0296, 0.1249, 0.0467, 0.0290, 0.1301, 0.0562, 0.1683], device='cuda:1'), in_proj_covar=tensor([0.0148, 0.0154, 0.0173, 0.0167, 0.0127, 0.0157, 0.0151, 0.0158], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 21:03:01,631 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36700.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 21:03:14,050 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.896e+02 2.981e+02 3.490e+02 4.331e+02 9.127e+02, threshold=6.980e+02, percent-clipped=1.0 2023-04-16 21:03:24,665 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.47 vs. limit=5.0 2023-04-16 21:03:37,821 INFO [train.py:893] (1/4) Epoch 14, batch 1650, loss[loss=0.1909, simple_loss=0.2484, pruned_loss=0.0667, over 13358.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2549, pruned_loss=0.07224, over 2660559.43 frames. ], batch size: 84, lr: 9.36e-03, grad_scale: 8.0 2023-04-16 21:03:43,161 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8064, 4.1928, 3.9974, 3.9176, 4.0588, 3.8108, 4.2376, 4.2740], device='cuda:1'), covar=tensor([0.0256, 0.0267, 0.0253, 0.0399, 0.0326, 0.0326, 0.0327, 0.0273], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0183, 0.0142, 0.0166, 0.0130, 0.0179, 0.0120, 0.0152], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 21:03:57,643 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36761.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 21:04:22,437 INFO [train.py:893] (1/4) Epoch 14, batch 1700, loss[loss=0.1667, simple_loss=0.2254, pruned_loss=0.05402, over 13388.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.254, pruned_loss=0.07138, over 2659848.06 frames. ], batch size: 62, lr: 9.36e-03, grad_scale: 8.0 2023-04-16 21:04:32,976 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-04-16 21:04:38,579 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36805.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 21:04:45,891 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.849e+02 2.903e+02 3.363e+02 3.975e+02 5.959e+02, threshold=6.726e+02, percent-clipped=0.0 2023-04-16 21:05:08,690 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=36837.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:05:10,195 INFO [train.py:893] (1/4) Epoch 14, batch 1750, loss[loss=0.2045, simple_loss=0.2599, pruned_loss=0.07455, over 13420.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2532, pruned_loss=0.07103, over 2661850.30 frames. ], batch size: 95, lr: 9.35e-03, grad_scale: 8.0 2023-04-16 21:05:15,631 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4313, 2.1624, 2.4254, 3.9638, 3.6157, 3.9962, 2.9681, 2.1663], device='cuda:1'), covar=tensor([0.0273, 0.1120, 0.0926, 0.0054, 0.0228, 0.0050, 0.0722, 0.1223], device='cuda:1'), in_proj_covar=tensor([0.0110, 0.0145, 0.0157, 0.0082, 0.0110, 0.0079, 0.0162, 0.0144], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0001, 0.0003, 0.0003], device='cuda:1') 2023-04-16 21:05:36,551 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36866.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 21:05:53,874 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=36885.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:05:57,805 INFO [train.py:893] (1/4) Epoch 14, batch 1800, loss[loss=0.1785, simple_loss=0.241, pruned_loss=0.05802, over 13485.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2527, pruned_loss=0.0707, over 2656959.67 frames. ], batch size: 81, lr: 9.34e-03, grad_scale: 8.0 2023-04-16 21:05:59,199 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2023-04-16 21:06:13,306 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0642, 4.0257, 4.0270, 2.5786, 4.5316, 4.2300, 4.1472, 4.4954], device='cuda:1'), covar=tensor([0.0267, 0.0160, 0.0177, 0.1266, 0.0203, 0.0289, 0.0187, 0.0112], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0045, 0.0070, 0.0094, 0.0087, 0.0089, 0.0069, 0.0063], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 21:06:14,892 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.8373, 4.7323, 4.8810, 4.7815, 5.1889, 4.7350, 5.1882, 5.1972], device='cuda:1'), covar=tensor([0.0369, 0.0587, 0.0621, 0.0516, 0.0541, 0.0743, 0.0466, 0.0448], device='cuda:1'), in_proj_covar=tensor([0.0248, 0.0263, 0.0261, 0.0194, 0.0371, 0.0301, 0.0232, 0.0256], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0005, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 21:06:17,397 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36911.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:06:19,631 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.892e+02 2.727e+02 3.240e+02 3.998e+02 7.863e+02, threshold=6.481e+02, percent-clipped=1.0 2023-04-16 21:06:43,938 INFO [train.py:893] (1/4) Epoch 14, batch 1850, loss[loss=0.239, simple_loss=0.2901, pruned_loss=0.09388, over 13453.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2524, pruned_loss=0.07052, over 2661795.75 frames. ], batch size: 106, lr: 9.34e-03, grad_scale: 8.0 2023-04-16 21:06:44,897 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-16 21:07:02,614 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=36958.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:07:05,108 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36961.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:07:07,619 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36964.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:07:10,859 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=36968.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:07:14,390 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=36972.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:07:31,580 INFO [train.py:893] (1/4) Epoch 14, batch 1900, loss[loss=0.1955, simple_loss=0.2501, pruned_loss=0.07042, over 13517.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.2511, pruned_loss=0.07019, over 2661761.02 frames. ], batch size: 98, lr: 9.33e-03, grad_scale: 8.0 2023-04-16 21:07:36,073 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0420, 1.9855, 2.3457, 3.3167, 3.0884, 3.3496, 2.6534, 2.1494], device='cuda:1'), covar=tensor([0.0230, 0.0963, 0.0716, 0.0073, 0.0227, 0.0063, 0.0626, 0.0924], device='cuda:1'), in_proj_covar=tensor([0.0110, 0.0145, 0.0156, 0.0082, 0.0109, 0.0079, 0.0161, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0001, 0.0003, 0.0002], device='cuda:1') 2023-04-16 21:07:41,171 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.1204, 4.9653, 5.1222, 4.8773, 5.4726, 4.9530, 5.4829, 5.4783], device='cuda:1'), covar=tensor([0.0369, 0.0603, 0.0720, 0.0568, 0.0538, 0.0854, 0.0529, 0.0530], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0266, 0.0264, 0.0197, 0.0373, 0.0305, 0.0235, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0005, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 21:07:44,618 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37004.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:07:53,573 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.879e+02 2.818e+02 3.261e+02 3.804e+02 6.289e+02, threshold=6.522e+02, percent-clipped=0.0 2023-04-16 21:07:57,062 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0660, 1.9775, 3.8321, 3.7477, 3.7252, 3.0626, 3.5717, 2.9406], device='cuda:1'), covar=tensor([0.2089, 0.1533, 0.0096, 0.0210, 0.0254, 0.0640, 0.0220, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.0208, 0.0182, 0.0110, 0.0119, 0.0123, 0.0169, 0.0132, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 21:07:58,751 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37019.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:08:16,145 INFO [train.py:893] (1/4) Epoch 14, batch 1950, loss[loss=0.2322, simple_loss=0.2817, pruned_loss=0.09132, over 13497.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.2511, pruned_loss=0.07054, over 2662658.50 frames. ], batch size: 93, lr: 9.32e-03, grad_scale: 8.0 2023-04-16 21:08:33,825 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37056.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 21:08:41,380 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37065.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:08:55,465 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9181, 4.3851, 4.2531, 4.1786, 4.2016, 4.0409, 4.3878, 4.4594], device='cuda:1'), covar=tensor([0.0202, 0.0192, 0.0173, 0.0295, 0.0269, 0.0262, 0.0293, 0.0204], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0181, 0.0142, 0.0165, 0.0128, 0.0180, 0.0122, 0.0153], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 21:09:03,558 INFO [train.py:893] (1/4) Epoch 14, batch 2000, loss[loss=0.2799, simple_loss=0.3124, pruned_loss=0.1236, over 11650.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2533, pruned_loss=0.07175, over 2661463.54 frames. ], batch size: 157, lr: 9.32e-03, grad_scale: 8.0 2023-04-16 21:09:07,707 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-16 21:09:26,940 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.742e+02 2.839e+02 3.422e+02 4.168e+02 1.026e+03, threshold=6.843e+02, percent-clipped=3.0 2023-04-16 21:09:48,173 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5534, 4.3280, 4.5283, 4.4801, 4.7571, 4.2810, 4.7981, 4.7796], device='cuda:1'), covar=tensor([0.0375, 0.0577, 0.0636, 0.0496, 0.0559, 0.0840, 0.0509, 0.0490], device='cuda:1'), in_proj_covar=tensor([0.0251, 0.0267, 0.0264, 0.0198, 0.0375, 0.0306, 0.0236, 0.0260], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0005, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 21:09:49,570 INFO [train.py:893] (1/4) Epoch 14, batch 2050, loss[loss=0.2065, simple_loss=0.2619, pruned_loss=0.07552, over 13239.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2548, pruned_loss=0.07246, over 2659406.80 frames. ], batch size: 132, lr: 9.31e-03, grad_scale: 8.0 2023-04-16 21:10:09,956 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37161.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 21:10:21,856 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2491, 3.5089, 3.4057, 3.9118, 2.2339, 2.8295, 3.6424, 2.0029], device='cuda:1'), covar=tensor([0.0135, 0.0605, 0.0722, 0.0501, 0.1472, 0.1065, 0.0641, 0.1869], device='cuda:1'), in_proj_covar=tensor([0.0107, 0.0176, 0.0193, 0.0216, 0.0176, 0.0191, 0.0171, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 21:10:36,601 INFO [train.py:893] (1/4) Epoch 14, batch 2100, loss[loss=0.229, simple_loss=0.2807, pruned_loss=0.08862, over 13024.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.254, pruned_loss=0.07205, over 2660396.70 frames. ], batch size: 142, lr: 9.31e-03, grad_scale: 8.0 2023-04-16 21:10:51,259 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-04-16 21:11:00,506 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.834e+02 2.540e+02 2.961e+02 3.544e+02 6.483e+02, threshold=5.922e+02, percent-clipped=0.0 2023-04-16 21:11:15,271 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4805, 2.7420, 2.7974, 4.2982, 3.8439, 4.3231, 3.2671, 2.8148], device='cuda:1'), covar=tensor([0.0263, 0.0821, 0.0715, 0.0038, 0.0215, 0.0028, 0.0587, 0.0740], device='cuda:1'), in_proj_covar=tensor([0.0110, 0.0145, 0.0155, 0.0083, 0.0111, 0.0079, 0.0161, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0001, 0.0002, 0.0001, 0.0003, 0.0002], device='cuda:1') 2023-04-16 21:11:23,858 INFO [train.py:893] (1/4) Epoch 14, batch 2150, loss[loss=0.2221, simple_loss=0.2732, pruned_loss=0.08546, over 13335.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2543, pruned_loss=0.07152, over 2661478.78 frames. ], batch size: 118, lr: 9.30e-03, grad_scale: 8.0 2023-04-16 21:11:44,101 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37261.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:11:46,715 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37264.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:11:49,127 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37267.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:11:49,969 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37268.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:11:50,020 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37268.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:12:08,588 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.83 vs. limit=2.0 2023-04-16 21:12:08,911 INFO [train.py:893] (1/4) Epoch 14, batch 2200, loss[loss=0.1951, simple_loss=0.2491, pruned_loss=0.07056, over 13459.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2523, pruned_loss=0.07061, over 2663626.16 frames. ], batch size: 100, lr: 9.29e-03, grad_scale: 8.0 2023-04-16 21:12:11,705 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.2731, 3.2975, 4.0085, 2.8445, 2.6660, 2.7273, 4.1690, 4.3834], device='cuda:1'), covar=tensor([0.1171, 0.1577, 0.0354, 0.1469, 0.1455, 0.1341, 0.0277, 0.0181], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0245, 0.0178, 0.0211, 0.0205, 0.0173, 0.0183, 0.0150], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 21:12:11,797 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-04-16 21:12:28,762 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37309.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:12:31,394 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37312.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:12:32,932 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.839e+02 2.752e+02 3.188e+02 3.717e+02 5.607e+02, threshold=6.377e+02, percent-clipped=0.0 2023-04-16 21:12:33,132 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37314.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:12:34,166 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4529, 3.7758, 3.5447, 4.0779, 2.1961, 3.1162, 3.8535, 2.2455], device='cuda:1'), covar=tensor([0.0107, 0.0460, 0.0819, 0.0585, 0.1652, 0.0909, 0.0495, 0.2016], device='cuda:1'), in_proj_covar=tensor([0.0107, 0.0178, 0.0196, 0.0219, 0.0178, 0.0193, 0.0171, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 21:12:34,803 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37316.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:12:45,028 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-04-16 21:12:47,404 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37329.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:12:55,988 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.87 vs. limit=5.0 2023-04-16 21:12:56,960 INFO [train.py:893] (1/4) Epoch 14, batch 2250, loss[loss=0.22, simple_loss=0.2641, pruned_loss=0.08792, over 11823.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2522, pruned_loss=0.0708, over 2661804.25 frames. ], batch size: 157, lr: 9.29e-03, grad_scale: 8.0 2023-04-16 21:13:12,994 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37356.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 21:13:16,454 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37360.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:13:20,637 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37364.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:13:43,686 INFO [train.py:893] (1/4) Epoch 14, batch 2300, loss[loss=0.1885, simple_loss=0.2426, pruned_loss=0.0672, over 13406.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2518, pruned_loss=0.07073, over 2662122.58 frames. ], batch size: 65, lr: 9.28e-03, grad_scale: 8.0 2023-04-16 21:13:58,158 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37404.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 21:14:06,333 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.211e+02 2.889e+02 3.254e+02 3.813e+02 6.240e+02, threshold=6.509e+02, percent-clipped=0.0 2023-04-16 21:14:18,277 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=37425.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:14:30,526 INFO [train.py:893] (1/4) Epoch 14, batch 2350, loss[loss=0.2031, simple_loss=0.2546, pruned_loss=0.07586, over 13377.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2517, pruned_loss=0.07066, over 2664332.77 frames. ], batch size: 109, lr: 9.28e-03, grad_scale: 8.0 2023-04-16 21:14:45,897 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-04-16 21:14:51,985 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-16 21:14:52,174 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37461.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 21:15:16,018 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2023-04-16 21:15:17,094 INFO [train.py:893] (1/4) Epoch 14, batch 2400, loss[loss=0.16, simple_loss=0.2122, pruned_loss=0.05389, over 13173.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.2508, pruned_loss=0.07039, over 2664482.28 frames. ], batch size: 58, lr: 9.27e-03, grad_scale: 8.0 2023-04-16 21:15:36,051 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37509.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 21:15:38,668 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9288, 3.9093, 3.2019, 3.7391, 3.1687, 1.9730, 3.8604, 1.9233], device='cuda:1'), covar=tensor([0.0730, 0.0440, 0.0429, 0.0241, 0.0734, 0.2200, 0.0753, 0.1646], device='cuda:1'), in_proj_covar=tensor([0.0133, 0.0127, 0.0125, 0.0105, 0.0143, 0.0178, 0.0150, 0.0133], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 21:15:40,917 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.026e+02 2.653e+02 3.203e+02 3.692e+02 8.487e+02, threshold=6.405e+02, percent-clipped=2.0 2023-04-16 21:15:41,958 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.2886, 4.7828, 4.7752, 4.7739, 4.5641, 4.6275, 5.2539, 4.7579], device='cuda:1'), covar=tensor([0.0718, 0.1068, 0.2109, 0.2633, 0.0890, 0.1587, 0.0782, 0.1117], device='cuda:1'), in_proj_covar=tensor([0.0296, 0.0337, 0.0426, 0.0440, 0.0256, 0.0321, 0.0384, 0.0311], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 21:16:03,442 INFO [train.py:893] (1/4) Epoch 14, batch 2450, loss[loss=0.2084, simple_loss=0.2661, pruned_loss=0.07535, over 13430.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.2512, pruned_loss=0.07051, over 2668098.47 frames. ], batch size: 95, lr: 9.26e-03, grad_scale: 8.0 2023-04-16 21:16:30,194 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37567.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:16:37,704 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7873, 3.7761, 3.7131, 2.3485, 4.1360, 3.8718, 3.8865, 4.0292], device='cuda:1'), covar=tensor([0.0278, 0.0142, 0.0185, 0.1337, 0.0213, 0.0318, 0.0176, 0.0161], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0046, 0.0071, 0.0096, 0.0089, 0.0092, 0.0072, 0.0064], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 21:16:51,644 INFO [train.py:893] (1/4) Epoch 14, batch 2500, loss[loss=0.1658, simple_loss=0.227, pruned_loss=0.05231, over 13488.00 frames. ], tot_loss[loss=0.1956, simple_loss=0.2508, pruned_loss=0.07014, over 2669936.88 frames. ], batch size: 81, lr: 9.26e-03, grad_scale: 8.0 2023-04-16 21:17:14,572 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.069e+02 2.773e+02 3.225e+02 3.758e+02 6.643e+02, threshold=6.450e+02, percent-clipped=1.0 2023-04-16 21:17:14,849 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37614.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:17:15,622 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37615.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:17:24,953 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37624.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:17:38,654 INFO [train.py:893] (1/4) Epoch 14, batch 2550, loss[loss=0.2225, simple_loss=0.274, pruned_loss=0.08547, over 13233.00 frames. ], tot_loss[loss=0.1956, simple_loss=0.2508, pruned_loss=0.07018, over 2668762.96 frames. ], batch size: 124, lr: 9.25e-03, grad_scale: 8.0 2023-04-16 21:17:43,965 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8198, 4.1212, 3.9384, 3.9106, 3.9336, 4.2759, 4.0589, 3.9509], device='cuda:1'), covar=tensor([0.0311, 0.0244, 0.0269, 0.0874, 0.0307, 0.0209, 0.0271, 0.0432], device='cuda:1'), in_proj_covar=tensor([0.0140, 0.0138, 0.0153, 0.0252, 0.0156, 0.0171, 0.0151, 0.0105], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 21:17:58,130 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37660.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:17:59,628 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37662.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:18:01,195 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-16 21:18:22,192 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2942, 4.5470, 4.3018, 4.3165, 4.3928, 4.7566, 4.5373, 4.4296], device='cuda:1'), covar=tensor([0.0251, 0.0257, 0.0296, 0.0919, 0.0256, 0.0210, 0.0251, 0.0257], device='cuda:1'), in_proj_covar=tensor([0.0139, 0.0137, 0.0151, 0.0250, 0.0155, 0.0170, 0.0150, 0.0104], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 21:18:25,172 INFO [train.py:893] (1/4) Epoch 14, batch 2600, loss[loss=0.2123, simple_loss=0.2642, pruned_loss=0.08016, over 13526.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2512, pruned_loss=0.07094, over 2665893.42 frames. ], batch size: 91, lr: 9.25e-03, grad_scale: 8.0 2023-04-16 21:18:25,531 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.1372, 2.0858, 3.9094, 3.7302, 3.7026, 3.0278, 3.5114, 2.9632], device='cuda:1'), covar=tensor([0.1954, 0.1607, 0.0111, 0.0223, 0.0178, 0.0677, 0.0249, 0.0795], device='cuda:1'), in_proj_covar=tensor([0.0201, 0.0179, 0.0105, 0.0116, 0.0119, 0.0163, 0.0128, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 21:18:30,814 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1800, 4.4176, 3.5731, 3.0178, 3.1732, 2.6617, 4.5736, 2.5880], device='cuda:1'), covar=tensor([0.1425, 0.0258, 0.0825, 0.1628, 0.0690, 0.2760, 0.0187, 0.3460], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0279, 0.0287, 0.0303, 0.0238, 0.0303, 0.0195, 0.0347], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 21:18:42,897 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37708.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:18:47,446 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.401e+02 3.085e+02 3.718e+02 4.390e+02 1.242e+03, threshold=7.436e+02, percent-clipped=3.0 2023-04-16 21:18:52,003 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=37720.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:19:07,752 INFO [train.py:893] (1/4) Epoch 14, batch 2650, loss[loss=0.2114, simple_loss=0.2672, pruned_loss=0.07783, over 13445.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2525, pruned_loss=0.07189, over 2665825.84 frames. ], batch size: 103, lr: 9.24e-03, grad_scale: 8.0 2023-04-16 21:20:04,838 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 21:20:15,093 INFO [train.py:893] (1/4) Epoch 15, batch 0, loss[loss=0.1798, simple_loss=0.2426, pruned_loss=0.05854, over 13453.00 frames. ], tot_loss[loss=0.1798, simple_loss=0.2426, pruned_loss=0.05854, over 13453.00 frames. ], batch size: 79, lr: 8.92e-03, grad_scale: 8.0 2023-04-16 21:20:15,094 INFO [train.py:918] (1/4) Computing validation loss 2023-04-16 21:20:23,045 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7645, 3.5745, 2.8975, 3.3214, 2.8869, 2.0425, 3.4767, 1.7921], device='cuda:1'), covar=tensor([0.0639, 0.0461, 0.0407, 0.0390, 0.0674, 0.2023, 0.1087, 0.1567], device='cuda:1'), in_proj_covar=tensor([0.0135, 0.0129, 0.0127, 0.0106, 0.0144, 0.0180, 0.0154, 0.0135], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 21:20:37,763 INFO [train.py:927] (1/4) Epoch 15, validation: loss=0.1473, simple_loss=0.2051, pruned_loss=0.04473, over 2446609.00 frames. 2023-04-16 21:20:37,763 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12829MB 2023-04-16 21:20:58,292 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9677, 4.4310, 4.1435, 4.1917, 4.2332, 4.0071, 4.4806, 4.5096], device='cuda:1'), covar=tensor([0.0282, 0.0319, 0.0357, 0.0428, 0.0404, 0.0384, 0.0352, 0.0314], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0182, 0.0142, 0.0163, 0.0129, 0.0179, 0.0120, 0.0154], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 21:21:03,133 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.991e+02 2.725e+02 3.297e+02 3.929e+02 8.342e+02, threshold=6.594e+02, percent-clipped=2.0 2023-04-16 21:21:22,768 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-04-16 21:21:24,595 INFO [train.py:893] (1/4) Epoch 15, batch 50, loss[loss=0.1994, simple_loss=0.255, pruned_loss=0.07191, over 13533.00 frames. ], tot_loss[loss=0.1894, simple_loss=0.2436, pruned_loss=0.06758, over 606219.92 frames. ], batch size: 91, lr: 8.91e-03, grad_scale: 8.0 2023-04-16 21:21:43,329 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-04-16 21:21:48,726 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-16 21:21:48,726 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-16 21:21:48,726 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-16 21:21:48,732 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-16 21:21:48,740 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-16 21:21:48,761 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-16 21:21:48,771 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-16 21:22:11,255 INFO [train.py:893] (1/4) Epoch 15, batch 100, loss[loss=0.2258, simple_loss=0.2715, pruned_loss=0.09003, over 13581.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.249, pruned_loss=0.07185, over 1060492.76 frames. ], batch size: 89, lr: 8.91e-03, grad_scale: 8.0 2023-04-16 21:22:20,764 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0722, 4.3419, 4.0732, 4.0798, 4.1749, 4.4918, 4.3283, 4.1644], device='cuda:1'), covar=tensor([0.0271, 0.0285, 0.0297, 0.0872, 0.0252, 0.0218, 0.0251, 0.0330], device='cuda:1'), in_proj_covar=tensor([0.0138, 0.0136, 0.0149, 0.0245, 0.0153, 0.0167, 0.0147, 0.0103], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 21:22:36,142 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.741e+02 2.904e+02 3.473e+02 4.227e+02 6.603e+02, threshold=6.947e+02, percent-clipped=1.0 2023-04-16 21:22:46,272 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=37924.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:22:58,546 INFO [train.py:893] (1/4) Epoch 15, batch 150, loss[loss=0.188, simple_loss=0.2515, pruned_loss=0.06227, over 13460.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2513, pruned_loss=0.07273, over 1411709.54 frames. ], batch size: 100, lr: 8.90e-03, grad_scale: 8.0 2023-04-16 21:23:24,881 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0574, 2.2398, 1.9047, 3.8318, 4.3946, 3.2934, 4.2593, 4.0083], device='cuda:1'), covar=tensor([0.0113, 0.1108, 0.1246, 0.0136, 0.0110, 0.0476, 0.0120, 0.0117], device='cuda:1'), in_proj_covar=tensor([0.0067, 0.0084, 0.0092, 0.0073, 0.0059, 0.0076, 0.0050, 0.0067], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 21:23:31,301 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=37972.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:23:37,373 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=37979.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:23:46,330 INFO [train.py:893] (1/4) Epoch 15, batch 200, loss[loss=0.221, simple_loss=0.2716, pruned_loss=0.08517, over 13517.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2526, pruned_loss=0.0733, over 1679487.60 frames. ], batch size: 85, lr: 8.90e-03, grad_scale: 16.0 2023-04-16 21:24:13,369 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.082e+02 2.887e+02 3.344e+02 4.175e+02 8.023e+02, threshold=6.688e+02, percent-clipped=1.0 2023-04-16 21:24:20,115 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38020.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:24:36,574 INFO [train.py:893] (1/4) Epoch 15, batch 250, loss[loss=0.2006, simple_loss=0.2581, pruned_loss=0.07154, over 13366.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2545, pruned_loss=0.07408, over 1900667.92 frames. ], batch size: 109, lr: 8.89e-03, grad_scale: 16.0 2023-04-16 21:24:38,578 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38040.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:25:04,609 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=38068.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:25:22,406 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-04-16 21:25:22,632 INFO [train.py:893] (1/4) Epoch 15, batch 300, loss[loss=0.2074, simple_loss=0.2677, pruned_loss=0.0736, over 13538.00 frames. ], tot_loss[loss=0.2021, simple_loss=0.2554, pruned_loss=0.07437, over 2064836.84 frames. ], batch size: 70, lr: 8.88e-03, grad_scale: 16.0 2023-04-16 21:25:28,551 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7172, 2.4668, 1.9961, 3.5552, 3.9920, 3.0916, 3.9109, 3.6987], device='cuda:1'), covar=tensor([0.0070, 0.0882, 0.0953, 0.0094, 0.0058, 0.0430, 0.0079, 0.0088], device='cuda:1'), in_proj_covar=tensor([0.0067, 0.0085, 0.0092, 0.0073, 0.0059, 0.0076, 0.0051, 0.0068], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 21:25:47,314 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.869e+02 2.752e+02 3.316e+02 4.021e+02 5.948e+02, threshold=6.631e+02, percent-clipped=0.0 2023-04-16 21:26:08,992 INFO [train.py:893] (1/4) Epoch 15, batch 350, loss[loss=0.2011, simple_loss=0.2541, pruned_loss=0.07404, over 13361.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2547, pruned_loss=0.0745, over 2191929.90 frames. ], batch size: 73, lr: 8.88e-03, grad_scale: 16.0 2023-04-16 21:26:42,697 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-04-16 21:26:56,445 INFO [train.py:893] (1/4) Epoch 15, batch 400, loss[loss=0.1709, simple_loss=0.2319, pruned_loss=0.05499, over 13331.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2552, pruned_loss=0.07435, over 2300672.49 frames. ], batch size: 73, lr: 8.87e-03, grad_scale: 16.0 2023-04-16 21:27:21,335 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.975e+02 2.951e+02 3.498e+02 4.195e+02 7.332e+02, threshold=6.996e+02, percent-clipped=1.0 2023-04-16 21:27:44,052 INFO [train.py:893] (1/4) Epoch 15, batch 450, loss[loss=0.1977, simple_loss=0.2537, pruned_loss=0.07085, over 13537.00 frames. ], tot_loss[loss=0.2037, simple_loss=0.2569, pruned_loss=0.07526, over 2379754.23 frames. ], batch size: 72, lr: 8.87e-03, grad_scale: 16.0 2023-04-16 21:28:07,468 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-16 21:28:29,427 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1852, 4.1546, 3.1204, 3.8905, 3.1882, 2.1454, 4.0599, 2.0082], device='cuda:1'), covar=tensor([0.0584, 0.0340, 0.0433, 0.0229, 0.0614, 0.1880, 0.0804, 0.1421], device='cuda:1'), in_proj_covar=tensor([0.0134, 0.0128, 0.0126, 0.0107, 0.0143, 0.0180, 0.0154, 0.0135], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 21:28:31,550 INFO [train.py:893] (1/4) Epoch 15, batch 500, loss[loss=0.1913, simple_loss=0.2527, pruned_loss=0.06492, over 13381.00 frames. ], tot_loss[loss=0.2029, simple_loss=0.2563, pruned_loss=0.07471, over 2442242.08 frames. ], batch size: 113, lr: 8.86e-03, grad_scale: 16.0 2023-04-16 21:28:55,561 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.883e+02 2.767e+02 3.304e+02 3.926e+02 2.162e+03, threshold=6.608e+02, percent-clipped=2.0 2023-04-16 21:29:10,215 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-04-16 21:29:15,805 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38335.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:29:16,258 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.68 vs. limit=5.0 2023-04-16 21:29:18,099 INFO [train.py:893] (1/4) Epoch 15, batch 550, loss[loss=0.2106, simple_loss=0.2647, pruned_loss=0.07821, over 13457.00 frames. ], tot_loss[loss=0.2016, simple_loss=0.2556, pruned_loss=0.07384, over 2489837.62 frames. ], batch size: 103, lr: 8.86e-03, grad_scale: 16.0 2023-04-16 21:29:38,351 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38360.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:29:59,650 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6026, 2.2093, 2.2515, 2.7259, 2.0654, 2.6804, 2.6187, 2.2276], device='cuda:1'), covar=tensor([0.0094, 0.0197, 0.0155, 0.0143, 0.0194, 0.0109, 0.0171, 0.0174], device='cuda:1'), in_proj_covar=tensor([0.0084, 0.0092, 0.0100, 0.0094, 0.0107, 0.0084, 0.0087, 0.0085], device='cuda:1'), out_proj_covar=tensor([8.9390e-05, 1.0123e-04, 1.1131e-04, 1.0301e-04, 1.1946e-04, 9.0794e-05, 9.6199e-05, 9.2002e-05], device='cuda:1') 2023-04-16 21:30:04,324 INFO [train.py:893] (1/4) Epoch 15, batch 600, loss[loss=0.2126, simple_loss=0.2575, pruned_loss=0.08388, over 11907.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2543, pruned_loss=0.07343, over 2521887.91 frames. ], batch size: 157, lr: 8.85e-03, grad_scale: 16.0 2023-04-16 21:30:29,601 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.966e+02 2.757e+02 3.172e+02 3.903e+02 7.089e+02, threshold=6.344e+02, percent-clipped=2.0 2023-04-16 21:30:35,980 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38421.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:30:51,599 INFO [train.py:893] (1/4) Epoch 15, batch 650, loss[loss=0.1933, simple_loss=0.2509, pruned_loss=0.06787, over 13376.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2533, pruned_loss=0.07325, over 2551876.52 frames. ], batch size: 109, lr: 8.84e-03, grad_scale: 16.0 2023-04-16 21:31:08,330 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9530, 4.4054, 4.1325, 4.1275, 4.2378, 3.9727, 4.3884, 4.4579], device='cuda:1'), covar=tensor([0.0239, 0.0248, 0.0252, 0.0341, 0.0267, 0.0326, 0.0328, 0.0238], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0180, 0.0140, 0.0163, 0.0128, 0.0177, 0.0119, 0.0153], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 21:31:19,254 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.1476, 2.1840, 4.0070, 3.8661, 3.7845, 2.9945, 3.6768, 2.8495], device='cuda:1'), covar=tensor([0.2067, 0.1549, 0.0085, 0.0193, 0.0253, 0.0746, 0.0196, 0.0884], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0185, 0.0112, 0.0120, 0.0126, 0.0172, 0.0132, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 21:31:33,441 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.2665, 1.9972, 1.8139, 2.2658, 1.6606, 2.1903, 2.0616, 1.8573], device='cuda:1'), covar=tensor([0.0097, 0.0197, 0.0137, 0.0133, 0.0206, 0.0156, 0.0201, 0.0186], device='cuda:1'), in_proj_covar=tensor([0.0086, 0.0094, 0.0102, 0.0096, 0.0110, 0.0086, 0.0089, 0.0087], device='cuda:1'), out_proj_covar=tensor([9.1907e-05, 1.0334e-04, 1.1380e-04, 1.0499e-04, 1.2262e-04, 9.2438e-05, 9.7796e-05, 9.3843e-05], device='cuda:1') 2023-04-16 21:31:38,114 INFO [train.py:893] (1/4) Epoch 15, batch 700, loss[loss=0.2023, simple_loss=0.2587, pruned_loss=0.073, over 13458.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2523, pruned_loss=0.07254, over 2571553.17 frames. ], batch size: 106, lr: 8.84e-03, grad_scale: 16.0 2023-04-16 21:32:02,936 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.840e+02 2.763e+02 3.318e+02 3.949e+02 6.088e+02, threshold=6.635e+02, percent-clipped=0.0 2023-04-16 21:32:22,804 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8041, 3.8760, 2.8464, 3.5278, 3.0948, 2.1305, 3.7538, 1.9306], device='cuda:1'), covar=tensor([0.0816, 0.0375, 0.0571, 0.0358, 0.0725, 0.2273, 0.0917, 0.1508], device='cuda:1'), in_proj_covar=tensor([0.0135, 0.0130, 0.0127, 0.0107, 0.0144, 0.0180, 0.0156, 0.0135], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 21:32:25,738 INFO [train.py:893] (1/4) Epoch 15, batch 750, loss[loss=0.1972, simple_loss=0.2432, pruned_loss=0.07564, over 13387.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2518, pruned_loss=0.07269, over 2589130.39 frames. ], batch size: 67, lr: 8.83e-03, grad_scale: 16.0 2023-04-16 21:33:12,674 INFO [train.py:893] (1/4) Epoch 15, batch 800, loss[loss=0.2156, simple_loss=0.2642, pruned_loss=0.08348, over 13513.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2527, pruned_loss=0.07304, over 2603379.08 frames. ], batch size: 85, lr: 8.83e-03, grad_scale: 16.0 2023-04-16 21:33:16,197 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.5304, 5.0809, 5.0092, 5.1422, 4.7808, 4.9351, 5.5851, 4.9510], device='cuda:1'), covar=tensor([0.0707, 0.1147, 0.2167, 0.2474, 0.0877, 0.1506, 0.0743, 0.1195], device='cuda:1'), in_proj_covar=tensor([0.0297, 0.0341, 0.0431, 0.0446, 0.0264, 0.0329, 0.0390, 0.0313], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 21:33:26,551 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.95 vs. limit=5.0 2023-04-16 21:33:27,068 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3551, 2.9081, 2.3288, 4.1975, 4.8224, 3.6458, 4.7024, 4.3826], device='cuda:1'), covar=tensor([0.0103, 0.0822, 0.1038, 0.0112, 0.0070, 0.0423, 0.0084, 0.0092], device='cuda:1'), in_proj_covar=tensor([0.0068, 0.0086, 0.0093, 0.0074, 0.0059, 0.0077, 0.0051, 0.0069], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 21:33:36,765 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.167e+02 2.924e+02 3.517e+02 4.186e+02 6.673e+02, threshold=7.034e+02, percent-clipped=1.0 2023-04-16 21:33:45,462 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7467, 3.8680, 3.0061, 3.4483, 3.1364, 2.1634, 3.7838, 1.9288], device='cuda:1'), covar=tensor([0.0748, 0.0378, 0.0504, 0.0348, 0.0649, 0.1976, 0.0909, 0.1379], device='cuda:1'), in_proj_covar=tensor([0.0135, 0.0130, 0.0126, 0.0107, 0.0143, 0.0181, 0.0155, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 21:33:58,028 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=38635.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:34:00,217 INFO [train.py:893] (1/4) Epoch 15, batch 850, loss[loss=0.2074, simple_loss=0.2658, pruned_loss=0.07451, over 13471.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2543, pruned_loss=0.07353, over 2617903.52 frames. ], batch size: 103, lr: 8.82e-03, grad_scale: 16.0 2023-04-16 21:34:03,079 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1495, 4.4965, 4.2604, 4.2739, 4.2841, 4.7165, 4.4343, 4.4574], device='cuda:1'), covar=tensor([0.0307, 0.0303, 0.0320, 0.1101, 0.0285, 0.0242, 0.0286, 0.0246], device='cuda:1'), in_proj_covar=tensor([0.0140, 0.0138, 0.0152, 0.0248, 0.0154, 0.0170, 0.0150, 0.0106], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 21:34:42,324 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=38683.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:34:46,947 INFO [train.py:893] (1/4) Epoch 15, batch 900, loss[loss=0.1783, simple_loss=0.2279, pruned_loss=0.06436, over 13416.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2543, pruned_loss=0.07397, over 2619125.64 frames. ], batch size: 65, lr: 8.82e-03, grad_scale: 16.0 2023-04-16 21:35:04,796 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.7369, 3.9559, 3.7708, 4.3492, 2.3230, 3.3363, 4.1432, 2.2505], device='cuda:1'), covar=tensor([0.0100, 0.0464, 0.0667, 0.0688, 0.1498, 0.0837, 0.0415, 0.1857], device='cuda:1'), in_proj_covar=tensor([0.0109, 0.0177, 0.0195, 0.0218, 0.0178, 0.0191, 0.0172, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 21:35:10,823 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.917e+02 2.822e+02 3.453e+02 4.319e+02 8.057e+02, threshold=6.905e+02, percent-clipped=3.0 2023-04-16 21:35:12,772 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=38716.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:35:18,437 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-16 21:35:33,150 INFO [train.py:893] (1/4) Epoch 15, batch 950, loss[loss=0.2064, simple_loss=0.254, pruned_loss=0.0794, over 13362.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2535, pruned_loss=0.07423, over 2629134.94 frames. ], batch size: 73, lr: 8.81e-03, grad_scale: 16.0 2023-04-16 21:36:01,131 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.3023, 5.1839, 5.4323, 5.1022, 5.6776, 5.1630, 5.6783, 5.6815], device='cuda:1'), covar=tensor([0.0332, 0.0546, 0.0609, 0.0505, 0.0509, 0.0822, 0.0414, 0.0436], device='cuda:1'), in_proj_covar=tensor([0.0254, 0.0272, 0.0271, 0.0200, 0.0387, 0.0311, 0.0243, 0.0266], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0002, 0.0005, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 21:36:20,159 INFO [train.py:893] (1/4) Epoch 15, batch 1000, loss[loss=0.2212, simple_loss=0.2616, pruned_loss=0.09041, over 13363.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2514, pruned_loss=0.07332, over 2635199.36 frames. ], batch size: 118, lr: 8.80e-03, grad_scale: 16.0 2023-04-16 21:36:21,426 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-04-16 21:36:44,532 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.713e+02 2.946e+02 3.426e+02 3.965e+02 6.022e+02, threshold=6.852e+02, percent-clipped=0.0 2023-04-16 21:36:53,934 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=38823.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:37:06,384 INFO [train.py:893] (1/4) Epoch 15, batch 1050, loss[loss=0.1714, simple_loss=0.2225, pruned_loss=0.0601, over 13176.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.25, pruned_loss=0.07223, over 2642632.39 frames. ], batch size: 58, lr: 8.80e-03, grad_scale: 16.0 2023-04-16 21:37:07,523 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6099, 3.7838, 2.4446, 3.5243, 3.6207, 2.2870, 3.2618, 2.4548], device='cuda:1'), covar=tensor([0.0285, 0.0293, 0.1170, 0.0419, 0.0290, 0.1265, 0.0629, 0.1405], device='cuda:1'), in_proj_covar=tensor([0.0147, 0.0155, 0.0169, 0.0169, 0.0125, 0.0157, 0.0151, 0.0156], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 21:37:44,532 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.86 vs. limit=5.0 2023-04-16 21:37:50,047 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=38884.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:37:52,986 INFO [train.py:893] (1/4) Epoch 15, batch 1100, loss[loss=0.1919, simple_loss=0.255, pruned_loss=0.0644, over 13518.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.2495, pruned_loss=0.07106, over 2650726.06 frames. ], batch size: 91, lr: 8.79e-03, grad_scale: 16.0 2023-04-16 21:38:16,576 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.006e+02 2.873e+02 3.467e+02 4.296e+02 7.811e+02, threshold=6.933e+02, percent-clipped=2.0 2023-04-16 21:38:38,235 INFO [train.py:893] (1/4) Epoch 15, batch 1150, loss[loss=0.205, simple_loss=0.2566, pruned_loss=0.07668, over 13524.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.2501, pruned_loss=0.07061, over 2657841.50 frames. ], batch size: 83, lr: 8.79e-03, grad_scale: 16.0 2023-04-16 21:38:39,389 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0469, 1.8855, 3.7984, 3.7493, 3.7030, 2.9604, 3.4336, 2.7354], device='cuda:1'), covar=tensor([0.2017, 0.1546, 0.0105, 0.0271, 0.0184, 0.0663, 0.0234, 0.0806], device='cuda:1'), in_proj_covar=tensor([0.0205, 0.0179, 0.0110, 0.0119, 0.0122, 0.0169, 0.0130, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 21:39:03,827 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1065, 3.4226, 3.3273, 3.8378, 2.1519, 2.8017, 3.5768, 1.9588], device='cuda:1'), covar=tensor([0.0110, 0.0519, 0.0734, 0.0549, 0.1621, 0.1052, 0.0634, 0.1953], device='cuda:1'), in_proj_covar=tensor([0.0111, 0.0178, 0.0197, 0.0222, 0.0180, 0.0193, 0.0173, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 21:39:26,103 INFO [train.py:893] (1/4) Epoch 15, batch 1200, loss[loss=0.1851, simple_loss=0.2407, pruned_loss=0.06476, over 13524.00 frames. ], tot_loss[loss=0.195, simple_loss=0.2499, pruned_loss=0.07007, over 2655584.86 frames. ], batch size: 70, lr: 8.78e-03, grad_scale: 16.0 2023-04-16 21:39:50,724 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.045e+02 2.812e+02 3.202e+02 4.092e+02 7.055e+02, threshold=6.404e+02, percent-clipped=2.0 2023-04-16 21:39:52,738 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39016.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:39:55,066 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-16 21:40:05,951 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-16 21:40:12,596 INFO [train.py:893] (1/4) Epoch 15, batch 1250, loss[loss=0.2194, simple_loss=0.2684, pruned_loss=0.08513, over 13197.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2512, pruned_loss=0.07107, over 2653925.39 frames. ], batch size: 132, lr: 8.78e-03, grad_scale: 16.0 2023-04-16 21:40:36,673 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.5964, 3.5710, 2.8024, 3.2804, 2.8349, 1.9841, 3.5809, 1.9552], device='cuda:1'), covar=tensor([0.0741, 0.0700, 0.0508, 0.0376, 0.0744, 0.2375, 0.0986, 0.1599], device='cuda:1'), in_proj_covar=tensor([0.0134, 0.0130, 0.0126, 0.0108, 0.0145, 0.0181, 0.0154, 0.0137], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 21:40:37,448 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39064.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:40:50,769 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0133, 4.0537, 3.3312, 2.8818, 3.0113, 2.5054, 4.3038, 2.4839], device='cuda:1'), covar=tensor([0.1389, 0.0299, 0.0868, 0.1641, 0.0705, 0.2765, 0.0193, 0.3251], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0279, 0.0290, 0.0302, 0.0240, 0.0306, 0.0196, 0.0346], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 21:40:59,508 INFO [train.py:893] (1/4) Epoch 15, batch 1300, loss[loss=0.2079, simple_loss=0.2611, pruned_loss=0.07736, over 13394.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2534, pruned_loss=0.0722, over 2654692.96 frames. ], batch size: 113, lr: 8.77e-03, grad_scale: 16.0 2023-04-16 21:41:11,336 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.4149, 4.9249, 4.8374, 4.9766, 4.5816, 4.7266, 5.4349, 4.9100], device='cuda:1'), covar=tensor([0.0705, 0.1014, 0.1930, 0.2321, 0.0922, 0.1484, 0.0777, 0.1229], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0337, 0.0428, 0.0440, 0.0261, 0.0327, 0.0388, 0.0312], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 21:41:23,782 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.038e+02 2.996e+02 3.347e+02 3.864e+02 5.865e+02, threshold=6.694e+02, percent-clipped=0.0 2023-04-16 21:41:46,891 INFO [train.py:893] (1/4) Epoch 15, batch 1350, loss[loss=0.2091, simple_loss=0.2623, pruned_loss=0.07796, over 13440.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2542, pruned_loss=0.07262, over 2657548.74 frames. ], batch size: 106, lr: 8.77e-03, grad_scale: 16.0 2023-04-16 21:42:03,244 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.8019, 4.0733, 3.7013, 4.4808, 2.3699, 3.2877, 4.3412, 2.4716], device='cuda:1'), covar=tensor([0.0123, 0.0468, 0.0803, 0.0553, 0.1625, 0.0919, 0.0370, 0.1702], device='cuda:1'), in_proj_covar=tensor([0.0112, 0.0181, 0.0201, 0.0226, 0.0182, 0.0196, 0.0175, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 21:42:24,289 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39179.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:42:29,252 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39185.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:42:32,184 INFO [train.py:893] (1/4) Epoch 15, batch 1400, loss[loss=0.1967, simple_loss=0.25, pruned_loss=0.07165, over 13239.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2536, pruned_loss=0.07262, over 2653959.13 frames. ], batch size: 124, lr: 8.76e-03, grad_scale: 16.0 2023-04-16 21:42:37,436 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6302, 3.1160, 2.7008, 4.4305, 5.0284, 3.7208, 4.9483, 4.6648], device='cuda:1'), covar=tensor([0.0075, 0.0727, 0.0871, 0.0098, 0.0057, 0.0409, 0.0058, 0.0072], device='cuda:1'), in_proj_covar=tensor([0.0069, 0.0086, 0.0093, 0.0075, 0.0060, 0.0077, 0.0050, 0.0069], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 21:42:40,092 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-04-16 21:42:55,636 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.978e+02 2.883e+02 3.401e+02 4.315e+02 6.663e+02, threshold=6.801e+02, percent-clipped=0.0 2023-04-16 21:43:17,757 INFO [train.py:893] (1/4) Epoch 15, batch 1450, loss[loss=0.2196, simple_loss=0.2759, pruned_loss=0.08166, over 13452.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2527, pruned_loss=0.07228, over 2660623.22 frames. ], batch size: 103, lr: 8.75e-03, grad_scale: 16.0 2023-04-16 21:43:24,900 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39246.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:43:59,224 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39284.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:44:03,093 INFO [train.py:893] (1/4) Epoch 15, batch 1500, loss[loss=0.1703, simple_loss=0.2273, pruned_loss=0.05664, over 13532.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.252, pruned_loss=0.07191, over 2662259.63 frames. ], batch size: 72, lr: 8.75e-03, grad_scale: 16.0 2023-04-16 21:44:14,612 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 2023-04-16 21:44:14,712 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-04-16 21:44:27,200 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.904e+02 2.935e+02 3.502e+02 4.230e+02 9.051e+02, threshold=7.004e+02, percent-clipped=3.0 2023-04-16 21:44:43,433 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7411, 2.6211, 2.1665, 1.4983, 1.4755, 2.1982, 2.1963, 2.7775], device='cuda:1'), covar=tensor([0.0944, 0.0316, 0.0688, 0.1669, 0.0191, 0.0503, 0.0813, 0.0192], device='cuda:1'), in_proj_covar=tensor([0.0150, 0.0122, 0.0107, 0.0195, 0.0098, 0.0144, 0.0150, 0.0116], device='cuda:1'), out_proj_covar=tensor([1.1280e-04, 9.1843e-05, 8.5167e-05, 1.4781e-04, 7.5212e-05, 1.0878e-04, 1.1557e-04, 8.6508e-05], device='cuda:1') 2023-04-16 21:44:49,927 INFO [train.py:893] (1/4) Epoch 15, batch 1550, loss[loss=0.1966, simple_loss=0.2536, pruned_loss=0.06983, over 13322.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2514, pruned_loss=0.07151, over 2655939.37 frames. ], batch size: 118, lr: 8.74e-03, grad_scale: 16.0 2023-04-16 21:44:56,938 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39345.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:45:05,989 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-04-16 21:45:13,911 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-04-16 21:45:37,379 INFO [train.py:893] (1/4) Epoch 15, batch 1600, loss[loss=0.1872, simple_loss=0.2494, pruned_loss=0.06247, over 13546.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2514, pruned_loss=0.07098, over 2651261.05 frames. ], batch size: 76, lr: 8.74e-03, grad_scale: 16.0 2023-04-16 21:46:01,159 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.914e+02 2.757e+02 3.152e+02 3.828e+02 7.575e+02, threshold=6.304e+02, percent-clipped=2.0 2023-04-16 21:46:23,029 INFO [train.py:893] (1/4) Epoch 15, batch 1650, loss[loss=0.1882, simple_loss=0.2465, pruned_loss=0.06499, over 13514.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.252, pruned_loss=0.07074, over 2654260.24 frames. ], batch size: 87, lr: 8.73e-03, grad_scale: 16.0 2023-04-16 21:46:45,784 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39463.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:47:01,467 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39479.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:47:08,578 INFO [train.py:893] (1/4) Epoch 15, batch 1700, loss[loss=0.1661, simple_loss=0.2241, pruned_loss=0.05407, over 13379.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.2516, pruned_loss=0.07032, over 2653552.40 frames. ], batch size: 62, lr: 8.73e-03, grad_scale: 16.0 2023-04-16 21:47:24,234 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.62 vs. limit=5.0 2023-04-16 21:47:32,708 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.821e+02 2.928e+02 3.437e+02 4.043e+02 6.312e+02, threshold=6.874e+02, percent-clipped=1.0 2023-04-16 21:47:41,671 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39524.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:47:44,012 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39527.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:47:54,043 INFO [train.py:893] (1/4) Epoch 15, batch 1750, loss[loss=0.2084, simple_loss=0.255, pruned_loss=0.08084, over 11924.00 frames. ], tot_loss[loss=0.1945, simple_loss=0.2502, pruned_loss=0.06939, over 2649372.04 frames. ], batch size: 157, lr: 8.72e-03, grad_scale: 16.0 2023-04-16 21:47:57,599 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39541.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:48:40,253 INFO [train.py:893] (1/4) Epoch 15, batch 1800, loss[loss=0.1765, simple_loss=0.2393, pruned_loss=0.05685, over 13545.00 frames. ], tot_loss[loss=0.1941, simple_loss=0.2501, pruned_loss=0.06906, over 2653947.88 frames. ], batch size: 72, lr: 8.72e-03, grad_scale: 16.0 2023-04-16 21:49:04,641 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.825e+02 2.614e+02 3.171e+02 3.838e+02 5.993e+02, threshold=6.341e+02, percent-clipped=0.0 2023-04-16 21:49:07,839 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.53 vs. limit=5.0 2023-04-16 21:49:15,325 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5015, 3.4220, 4.0435, 2.9414, 2.7360, 2.8386, 4.2878, 4.3508], device='cuda:1'), covar=tensor([0.1045, 0.1520, 0.0358, 0.1479, 0.1381, 0.1330, 0.0253, 0.0211], device='cuda:1'), in_proj_covar=tensor([0.0228, 0.0246, 0.0180, 0.0213, 0.0206, 0.0174, 0.0186, 0.0154], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 21:49:18,224 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-04-16 21:49:28,273 INFO [train.py:893] (1/4) Epoch 15, batch 1850, loss[loss=0.2447, simple_loss=0.2897, pruned_loss=0.09986, over 13527.00 frames. ], tot_loss[loss=0.1939, simple_loss=0.2498, pruned_loss=0.06898, over 2654155.51 frames. ], batch size: 91, lr: 8.71e-03, grad_scale: 16.0 2023-04-16 21:49:29,357 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39639.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 21:49:30,118 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39640.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:49:30,865 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-16 21:50:14,115 INFO [train.py:893] (1/4) Epoch 15, batch 1900, loss[loss=0.1722, simple_loss=0.2362, pruned_loss=0.05411, over 13537.00 frames. ], tot_loss[loss=0.1931, simple_loss=0.249, pruned_loss=0.06861, over 2658075.53 frames. ], batch size: 98, lr: 8.71e-03, grad_scale: 16.0 2023-04-16 21:50:27,052 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39700.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 21:50:33,352 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3406, 2.5371, 2.5212, 3.9756, 3.5903, 4.0410, 3.0389, 2.3881], device='cuda:1'), covar=tensor([0.0352, 0.0890, 0.0833, 0.0065, 0.0256, 0.0043, 0.0678, 0.0911], device='cuda:1'), in_proj_covar=tensor([0.0115, 0.0146, 0.0159, 0.0087, 0.0112, 0.0085, 0.0162, 0.0144], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0001, 0.0003, 0.0003], device='cuda:1') 2023-04-16 21:50:38,875 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.021e+02 2.868e+02 3.320e+02 3.903e+02 8.982e+02, threshold=6.640e+02, percent-clipped=2.0 2023-04-16 21:51:01,192 INFO [train.py:893] (1/4) Epoch 15, batch 1950, loss[loss=0.181, simple_loss=0.2354, pruned_loss=0.06335, over 13549.00 frames. ], tot_loss[loss=0.193, simple_loss=0.2485, pruned_loss=0.06878, over 2658424.82 frames. ], batch size: 78, lr: 8.70e-03, grad_scale: 16.0 2023-04-16 21:51:29,313 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39767.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:51:33,625 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4957, 3.0745, 2.8512, 4.4148, 4.9558, 3.6092, 4.7838, 4.5353], device='cuda:1'), covar=tensor([0.0095, 0.0668, 0.0733, 0.0097, 0.0061, 0.0418, 0.0084, 0.0077], device='cuda:1'), in_proj_covar=tensor([0.0069, 0.0086, 0.0094, 0.0075, 0.0060, 0.0077, 0.0052, 0.0070], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 21:51:49,298 INFO [train.py:893] (1/4) Epoch 15, batch 2000, loss[loss=0.2122, simple_loss=0.2735, pruned_loss=0.07545, over 13422.00 frames. ], tot_loss[loss=0.1942, simple_loss=0.2498, pruned_loss=0.06928, over 2661979.27 frames. ], batch size: 95, lr: 8.69e-03, grad_scale: 16.0 2023-04-16 21:51:55,273 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-16 21:51:58,741 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.4235, 3.2426, 2.4640, 2.8093, 2.6617, 1.8338, 3.2695, 1.8799], device='cuda:1'), covar=tensor([0.0708, 0.0731, 0.0521, 0.0501, 0.0733, 0.2020, 0.1028, 0.1409], device='cuda:1'), in_proj_covar=tensor([0.0135, 0.0131, 0.0126, 0.0109, 0.0145, 0.0182, 0.0155, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 21:52:12,887 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.198e+02 2.857e+02 3.261e+02 3.989e+02 6.795e+02, threshold=6.522e+02, percent-clipped=2.0 2023-04-16 21:52:14,788 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=39815.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:52:16,525 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.2556, 1.9399, 3.7785, 3.7013, 3.7332, 2.9549, 3.4775, 2.8500], device='cuda:1'), covar=tensor([0.1963, 0.1618, 0.0117, 0.0198, 0.0216, 0.0698, 0.0249, 0.0804], device='cuda:1'), in_proj_covar=tensor([0.0210, 0.0182, 0.0112, 0.0119, 0.0122, 0.0171, 0.0131, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 21:52:19,781 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39819.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:52:27,330 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39828.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:52:35,375 INFO [train.py:893] (1/4) Epoch 15, batch 2050, loss[loss=0.1948, simple_loss=0.254, pruned_loss=0.06781, over 13507.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2529, pruned_loss=0.07133, over 2653898.23 frames. ], batch size: 81, lr: 8.69e-03, grad_scale: 16.0 2023-04-16 21:52:36,521 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9799, 2.3069, 2.0704, 3.7797, 4.3638, 3.1824, 4.2824, 3.9632], device='cuda:1'), covar=tensor([0.0134, 0.1128, 0.1143, 0.0146, 0.0117, 0.0520, 0.0114, 0.0123], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0087, 0.0095, 0.0076, 0.0061, 0.0078, 0.0052, 0.0071], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 21:52:38,863 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39841.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:53:09,598 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.67 vs. limit=2.0 2023-04-16 21:53:11,069 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=39876.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:53:23,124 INFO [train.py:893] (1/4) Epoch 15, batch 2100, loss[loss=0.1861, simple_loss=0.2457, pruned_loss=0.0632, over 13351.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2524, pruned_loss=0.07064, over 2658594.89 frames. ], batch size: 73, lr: 8.68e-03, grad_scale: 16.0 2023-04-16 21:53:24,227 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39889.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:53:29,519 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.67 vs. limit=2.0 2023-04-16 21:53:48,111 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.102e+02 2.822e+02 3.258e+02 3.975e+02 7.473e+02, threshold=6.516e+02, percent-clipped=2.0 2023-04-16 21:54:00,446 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0901, 2.8039, 2.5513, 1.8817, 1.7501, 2.6777, 2.6810, 3.0988], device='cuda:1'), covar=tensor([0.1033, 0.0359, 0.0678, 0.1756, 0.0440, 0.0683, 0.0694, 0.0277], device='cuda:1'), in_proj_covar=tensor([0.0150, 0.0123, 0.0106, 0.0195, 0.0100, 0.0144, 0.0151, 0.0118], device='cuda:1'), out_proj_covar=tensor([1.1263e-04, 9.2139e-05, 8.4639e-05, 1.4772e-04, 7.6557e-05, 1.0912e-04, 1.1591e-04, 8.7815e-05], device='cuda:1') 2023-04-16 21:54:09,167 INFO [train.py:893] (1/4) Epoch 15, batch 2150, loss[loss=0.21, simple_loss=0.2637, pruned_loss=0.07818, over 13535.00 frames. ], tot_loss[loss=0.195, simple_loss=0.2511, pruned_loss=0.06944, over 2659057.06 frames. ], batch size: 85, lr: 8.68e-03, grad_scale: 16.0 2023-04-16 21:54:11,053 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=39940.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:54:54,911 INFO [train.py:893] (1/4) Epoch 15, batch 2200, loss[loss=0.1926, simple_loss=0.2546, pruned_loss=0.06532, over 13540.00 frames. ], tot_loss[loss=0.1945, simple_loss=0.2509, pruned_loss=0.06909, over 2661015.87 frames. ], batch size: 98, lr: 8.67e-03, grad_scale: 32.0 2023-04-16 21:54:55,126 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=39988.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:55:01,950 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=39995.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 21:55:12,051 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9650, 3.7974, 3.0391, 3.5390, 2.9336, 2.2520, 3.8403, 2.1281], device='cuda:1'), covar=tensor([0.0708, 0.0539, 0.0494, 0.0338, 0.0751, 0.1968, 0.0779, 0.1423], device='cuda:1'), in_proj_covar=tensor([0.0136, 0.0132, 0.0127, 0.0111, 0.0147, 0.0183, 0.0156, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 21:55:12,074 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7755, 2.5340, 2.4123, 2.8662, 2.1658, 2.9501, 2.9653, 2.4049], device='cuda:1'), covar=tensor([0.0076, 0.0160, 0.0132, 0.0109, 0.0204, 0.0103, 0.0125, 0.0155], device='cuda:1'), in_proj_covar=tensor([0.0089, 0.0097, 0.0105, 0.0100, 0.0113, 0.0089, 0.0092, 0.0092], device='cuda:1'), out_proj_covar=tensor([9.4382e-05, 1.0509e-04, 1.1626e-04, 1.0983e-04, 1.2540e-04, 9.5864e-05, 1.0039e-04, 9.8177e-05], device='cuda:1') 2023-04-16 21:55:24,250 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.868e+02 2.683e+02 3.225e+02 3.898e+02 7.029e+02, threshold=6.450e+02, percent-clipped=1.0 2023-04-16 21:55:46,741 INFO [train.py:893] (1/4) Epoch 15, batch 2250, loss[loss=0.1988, simple_loss=0.2558, pruned_loss=0.07091, over 13466.00 frames. ], tot_loss[loss=0.1928, simple_loss=0.2492, pruned_loss=0.06819, over 2664082.28 frames. ], batch size: 103, lr: 8.67e-03, grad_scale: 32.0 2023-04-16 21:55:56,319 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9440, 4.1635, 3.2314, 2.8623, 3.0376, 2.4960, 4.2767, 2.4281], device='cuda:1'), covar=tensor([0.1521, 0.0312, 0.1014, 0.1817, 0.0710, 0.3137, 0.0209, 0.3878], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0279, 0.0293, 0.0307, 0.0245, 0.0311, 0.0201, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 21:56:32,587 INFO [train.py:893] (1/4) Epoch 15, batch 2300, loss[loss=0.1888, simple_loss=0.2436, pruned_loss=0.06695, over 13529.00 frames. ], tot_loss[loss=0.192, simple_loss=0.2485, pruned_loss=0.06774, over 2665988.20 frames. ], batch size: 83, lr: 8.66e-03, grad_scale: 32.0 2023-04-16 21:56:41,145 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7060, 3.3879, 3.6658, 2.3394, 3.8004, 3.6769, 3.6125, 3.7462], device='cuda:1'), covar=tensor([0.0198, 0.0138, 0.0126, 0.1084, 0.0114, 0.0196, 0.0120, 0.0117], device='cuda:1'), in_proj_covar=tensor([0.0083, 0.0048, 0.0074, 0.0099, 0.0091, 0.0095, 0.0073, 0.0067], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 21:56:56,942 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.950e+02 2.485e+02 3.043e+02 3.886e+02 6.974e+02, threshold=6.085e+02, percent-clipped=1.0 2023-04-16 21:57:01,352 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40119.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:57:05,549 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40123.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:57:09,918 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5032, 3.7530, 3.6534, 4.1905, 2.2530, 3.2313, 3.9161, 2.1598], device='cuda:1'), covar=tensor([0.0165, 0.0485, 0.0700, 0.0599, 0.1531, 0.0824, 0.0506, 0.1906], device='cuda:1'), in_proj_covar=tensor([0.0112, 0.0180, 0.0197, 0.0225, 0.0180, 0.0194, 0.0174, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 21:57:20,493 INFO [train.py:893] (1/4) Epoch 15, batch 2350, loss[loss=0.1782, simple_loss=0.2387, pruned_loss=0.05882, over 13531.00 frames. ], tot_loss[loss=0.1919, simple_loss=0.2483, pruned_loss=0.06771, over 2664967.42 frames. ], batch size: 70, lr: 8.66e-03, grad_scale: 32.0 2023-04-16 21:57:43,071 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-16 21:57:48,248 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40167.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:57:51,646 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40171.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:58:06,581 INFO [train.py:893] (1/4) Epoch 15, batch 2400, loss[loss=0.2166, simple_loss=0.2732, pruned_loss=0.08003, over 13516.00 frames. ], tot_loss[loss=0.1915, simple_loss=0.2478, pruned_loss=0.06755, over 2664536.62 frames. ], batch size: 85, lr: 8.65e-03, grad_scale: 32.0 2023-04-16 21:58:13,692 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40195.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:58:19,915 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.3448, 2.0244, 4.1471, 3.9180, 4.0742, 3.1414, 3.8085, 3.0185], device='cuda:1'), covar=tensor([0.1739, 0.1519, 0.0087, 0.0175, 0.0136, 0.0583, 0.0188, 0.0708], device='cuda:1'), in_proj_covar=tensor([0.0209, 0.0180, 0.0111, 0.0119, 0.0122, 0.0170, 0.0131, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 21:58:30,554 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.938e+02 2.754e+02 3.113e+02 3.687e+02 7.218e+02, threshold=6.226e+02, percent-clipped=3.0 2023-04-16 21:58:35,126 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40218.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:58:35,298 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-04-16 21:58:54,051 INFO [train.py:893] (1/4) Epoch 15, batch 2450, loss[loss=0.1781, simple_loss=0.2408, pruned_loss=0.05771, over 13456.00 frames. ], tot_loss[loss=0.1918, simple_loss=0.2478, pruned_loss=0.06792, over 2659761.25 frames. ], batch size: 79, lr: 8.65e-03, grad_scale: 32.0 2023-04-16 21:58:55,276 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9995, 1.7824, 3.8273, 3.7005, 3.6895, 2.8154, 3.5249, 2.7123], device='cuda:1'), covar=tensor([0.2241, 0.1811, 0.0102, 0.0186, 0.0185, 0.0800, 0.0230, 0.0924], device='cuda:1'), in_proj_covar=tensor([0.0208, 0.0180, 0.0110, 0.0118, 0.0122, 0.0169, 0.0131, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 21:58:57,951 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.97 vs. limit=5.0 2023-04-16 21:59:11,064 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40256.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:59:26,286 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40272.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:59:32,137 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40279.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 21:59:40,943 INFO [train.py:893] (1/4) Epoch 15, batch 2500, loss[loss=0.2123, simple_loss=0.2625, pruned_loss=0.08099, over 13433.00 frames. ], tot_loss[loss=0.192, simple_loss=0.248, pruned_loss=0.06794, over 2663176.67 frames. ], batch size: 106, lr: 8.64e-03, grad_scale: 32.0 2023-04-16 21:59:47,747 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40295.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 21:59:54,678 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8663, 3.6718, 3.8687, 2.4376, 4.1372, 3.8903, 3.9551, 4.0843], device='cuda:1'), covar=tensor([0.0217, 0.0130, 0.0123, 0.1062, 0.0116, 0.0201, 0.0115, 0.0092], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0048, 0.0073, 0.0098, 0.0091, 0.0094, 0.0073, 0.0067], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 22:00:05,312 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.732e+02 2.683e+02 3.082e+02 3.857e+02 6.124e+02, threshold=6.165e+02, percent-clipped=0.0 2023-04-16 22:00:21,492 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40330.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:00:23,968 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40333.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:00:27,911 INFO [train.py:893] (1/4) Epoch 15, batch 2550, loss[loss=0.1698, simple_loss=0.2292, pruned_loss=0.05525, over 13370.00 frames. ], tot_loss[loss=0.1914, simple_loss=0.2475, pruned_loss=0.06763, over 2666550.82 frames. ], batch size: 62, lr: 8.64e-03, grad_scale: 32.0 2023-04-16 22:00:28,225 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7272, 3.3776, 2.8061, 3.0757, 2.8673, 2.0978, 3.3983, 1.9149], device='cuda:1'), covar=tensor([0.0691, 0.0642, 0.0455, 0.0434, 0.0707, 0.1916, 0.1090, 0.1484], device='cuda:1'), in_proj_covar=tensor([0.0136, 0.0132, 0.0128, 0.0111, 0.0147, 0.0184, 0.0157, 0.0140], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 22:00:33,089 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40343.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 22:00:53,388 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-16 22:01:15,765 INFO [train.py:893] (1/4) Epoch 15, batch 2600, loss[loss=0.2029, simple_loss=0.2593, pruned_loss=0.07328, over 13523.00 frames. ], tot_loss[loss=0.1922, simple_loss=0.2479, pruned_loss=0.0682, over 2665868.17 frames. ], batch size: 98, lr: 8.63e-03, grad_scale: 32.0 2023-04-16 22:01:18,722 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40391.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:01:39,436 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.061e+02 2.820e+02 3.308e+02 3.713e+02 1.179e+03, threshold=6.616e+02, percent-clipped=5.0 2023-04-16 22:01:46,593 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40423.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:01:57,672 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.60 vs. limit=5.0 2023-04-16 22:01:58,690 INFO [train.py:893] (1/4) Epoch 15, batch 2650, loss[loss=0.2169, simple_loss=0.265, pruned_loss=0.08439, over 13389.00 frames. ], tot_loss[loss=0.1927, simple_loss=0.248, pruned_loss=0.06868, over 2652041.65 frames. ], batch size: 109, lr: 8.63e-03, grad_scale: 32.0 2023-04-16 22:02:24,410 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40471.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:02:24,503 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40471.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:02:57,511 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 22:03:07,417 INFO [train.py:893] (1/4) Epoch 16, batch 0, loss[loss=0.1925, simple_loss=0.2377, pruned_loss=0.07368, over 11881.00 frames. ], tot_loss[loss=0.1925, simple_loss=0.2377, pruned_loss=0.07368, over 11881.00 frames. ], batch size: 158, lr: 8.34e-03, grad_scale: 32.0 2023-04-16 22:03:07,417 INFO [train.py:918] (1/4) Computing validation loss 2023-04-16 22:03:24,970 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.3458, 2.1403, 2.1112, 2.4093, 1.9432, 2.4756, 2.3797, 1.9419], device='cuda:1'), covar=tensor([0.0091, 0.0188, 0.0125, 0.0136, 0.0199, 0.0116, 0.0184, 0.0185], device='cuda:1'), in_proj_covar=tensor([0.0090, 0.0099, 0.0106, 0.0100, 0.0114, 0.0089, 0.0092, 0.0091], device='cuda:1'), out_proj_covar=tensor([9.5390e-05, 1.0749e-04, 1.1806e-04, 1.0956e-04, 1.2664e-04, 9.5101e-05, 1.0094e-04, 9.7093e-05], device='cuda:1') 2023-04-16 22:03:26,940 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1939, 2.7930, 2.6616, 3.1159, 2.5896, 3.2970, 2.9926, 2.6614], device='cuda:1'), covar=tensor([0.0101, 0.0238, 0.0469, 0.0204, 0.0224, 0.0135, 0.0187, 0.0171], device='cuda:1'), in_proj_covar=tensor([0.0090, 0.0099, 0.0106, 0.0100, 0.0114, 0.0089, 0.0092, 0.0091], device='cuda:1'), out_proj_covar=tensor([9.5390e-05, 1.0749e-04, 1.1806e-04, 1.0956e-04, 1.2664e-04, 9.5101e-05, 1.0094e-04, 9.7093e-05], device='cuda:1') 2023-04-16 22:03:30,243 INFO [train.py:927] (1/4) Epoch 16, validation: loss=0.1448, simple_loss=0.2029, pruned_loss=0.04338, over 2446609.00 frames. 2023-04-16 22:03:30,244 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12829MB 2023-04-16 22:03:35,675 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40492.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:03:43,047 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3794, 2.1734, 2.6189, 3.9971, 3.5961, 4.0267, 3.1114, 2.1607], device='cuda:1'), covar=tensor([0.0287, 0.1092, 0.0833, 0.0049, 0.0249, 0.0048, 0.0658, 0.0988], device='cuda:1'), in_proj_covar=tensor([0.0112, 0.0146, 0.0158, 0.0088, 0.0112, 0.0084, 0.0162, 0.0144], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0001, 0.0003, 0.0003], device='cuda:1') 2023-04-16 22:03:52,174 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.9631, 4.8050, 5.0897, 4.8548, 5.3172, 4.8156, 5.3479, 5.3507], device='cuda:1'), covar=tensor([0.0432, 0.0603, 0.0582, 0.0559, 0.0534, 0.0885, 0.0425, 0.0446], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0273, 0.0272, 0.0205, 0.0391, 0.0316, 0.0244, 0.0268], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 22:03:55,234 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.945e+02 2.953e+02 3.389e+02 4.043e+02 6.293e+02, threshold=6.778e+02, percent-clipped=0.0 2023-04-16 22:04:00,536 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40519.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:04:18,041 INFO [train.py:893] (1/4) Epoch 16, batch 50, loss[loss=0.1829, simple_loss=0.2383, pruned_loss=0.06371, over 13489.00 frames. ], tot_loss[loss=0.1919, simple_loss=0.2443, pruned_loss=0.06979, over 593721.33 frames. ], batch size: 93, lr: 8.34e-03, grad_scale: 32.0 2023-04-16 22:04:30,742 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40551.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:04:32,500 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40553.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:04:42,104 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-16 22:04:42,105 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-16 22:04:42,105 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-16 22:04:42,113 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-16 22:04:42,132 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-16 22:04:42,158 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-16 22:04:42,168 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-16 22:04:52,262 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40574.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:05:03,727 INFO [train.py:893] (1/4) Epoch 16, batch 100, loss[loss=0.2024, simple_loss=0.2449, pruned_loss=0.07989, over 13378.00 frames. ], tot_loss[loss=0.1926, simple_loss=0.2453, pruned_loss=0.06995, over 1052740.88 frames. ], batch size: 62, lr: 8.33e-03, grad_scale: 32.0 2023-04-16 22:05:16,101 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 2023-04-16 22:05:28,494 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.136e+02 2.861e+02 3.473e+02 4.004e+02 9.328e+02, threshold=6.946e+02, percent-clipped=3.0 2023-04-16 22:05:41,998 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40628.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:05:50,828 INFO [train.py:893] (1/4) Epoch 16, batch 150, loss[loss=0.1987, simple_loss=0.2577, pruned_loss=0.0698, over 13457.00 frames. ], tot_loss[loss=0.1951, simple_loss=0.2478, pruned_loss=0.07118, over 1406680.03 frames. ], batch size: 79, lr: 8.33e-03, grad_scale: 32.0 2023-04-16 22:06:03,550 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40651.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:06:36,469 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.8831, 4.8495, 4.0946, 4.6981, 4.7320, 4.0737, 4.4881, 3.6982], device='cuda:1'), covar=tensor([0.0164, 0.0157, 0.0497, 0.0495, 0.0115, 0.0540, 0.0258, 0.0790], device='cuda:1'), in_proj_covar=tensor([0.0149, 0.0159, 0.0175, 0.0178, 0.0131, 0.0158, 0.0155, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 22:06:37,203 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40686.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:06:37,817 INFO [train.py:893] (1/4) Epoch 16, batch 200, loss[loss=0.2075, simple_loss=0.2589, pruned_loss=0.07805, over 13074.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2504, pruned_loss=0.07245, over 1672867.07 frames. ], batch size: 142, lr: 8.32e-03, grad_scale: 32.0 2023-04-16 22:06:49,097 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.74 vs. limit=2.0 2023-04-16 22:06:53,455 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 2023-04-16 22:07:01,484 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40712.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:07:02,785 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.944e+02 2.805e+02 3.442e+02 4.281e+02 7.612e+02, threshold=6.884e+02, percent-clipped=2.0 2023-04-16 22:07:11,081 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7763, 2.4456, 2.0865, 1.4415, 1.4863, 2.0616, 2.1321, 2.6124], device='cuda:1'), covar=tensor([0.0845, 0.0288, 0.0643, 0.1680, 0.0167, 0.0498, 0.0768, 0.0246], device='cuda:1'), in_proj_covar=tensor([0.0151, 0.0125, 0.0107, 0.0194, 0.0099, 0.0146, 0.0151, 0.0119], device='cuda:1'), out_proj_covar=tensor([1.1305e-04, 9.3752e-05, 8.5440e-05, 1.4743e-04, 7.5805e-05, 1.1057e-04, 1.1645e-04, 8.8649e-05], device='cuda:1') 2023-04-16 22:07:25,028 INFO [train.py:893] (1/4) Epoch 16, batch 250, loss[loss=0.1713, simple_loss=0.2275, pruned_loss=0.0576, over 13356.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.251, pruned_loss=0.07242, over 1877279.82 frames. ], batch size: 67, lr: 8.32e-03, grad_scale: 32.0 2023-04-16 22:07:54,651 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.2508, 3.1729, 3.7518, 2.7682, 2.5624, 2.5920, 4.0036, 4.1220], device='cuda:1'), covar=tensor([0.1310, 0.1704, 0.0452, 0.1731, 0.1597, 0.1644, 0.0325, 0.0248], device='cuda:1'), in_proj_covar=tensor([0.0229, 0.0248, 0.0181, 0.0214, 0.0208, 0.0174, 0.0188, 0.0156], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 22:08:08,915 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7630, 2.3723, 2.3722, 2.8113, 2.0941, 2.8723, 2.7098, 2.3704], device='cuda:1'), covar=tensor([0.0096, 0.0217, 0.0165, 0.0154, 0.0238, 0.0123, 0.0187, 0.0168], device='cuda:1'), in_proj_covar=tensor([0.0090, 0.0098, 0.0105, 0.0099, 0.0111, 0.0089, 0.0091, 0.0090], device='cuda:1'), out_proj_covar=tensor([9.5484e-05, 1.0662e-04, 1.1626e-04, 1.0853e-04, 1.2325e-04, 9.5489e-05, 9.8939e-05, 9.6407e-05], device='cuda:1') 2023-04-16 22:08:12,652 INFO [train.py:893] (1/4) Epoch 16, batch 300, loss[loss=0.2324, simple_loss=0.2772, pruned_loss=0.09376, over 13530.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2519, pruned_loss=0.0723, over 2052065.19 frames. ], batch size: 83, lr: 8.31e-03, grad_scale: 32.0 2023-04-16 22:08:38,220 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.958e+02 2.972e+02 3.495e+02 4.194e+02 5.977e+02, threshold=6.991e+02, percent-clipped=0.0 2023-04-16 22:08:59,501 INFO [train.py:893] (1/4) Epoch 16, batch 350, loss[loss=0.2222, simple_loss=0.2788, pruned_loss=0.08285, over 13442.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2522, pruned_loss=0.07237, over 2186198.56 frames. ], batch size: 95, lr: 8.31e-03, grad_scale: 32.0 2023-04-16 22:09:10,604 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=40848.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:09:13,193 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40851.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:09:16,693 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-04-16 22:09:25,896 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40865.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:09:35,029 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40874.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:09:47,329 INFO [train.py:893] (1/4) Epoch 16, batch 400, loss[loss=0.1883, simple_loss=0.2476, pruned_loss=0.06449, over 13504.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2519, pruned_loss=0.07152, over 2295002.75 frames. ], batch size: 81, lr: 8.30e-03, grad_scale: 32.0 2023-04-16 22:09:58,472 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40899.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:09:59,428 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40900.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:10:12,108 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.876e+02 2.703e+02 3.218e+02 3.995e+02 8.871e+02, threshold=6.435e+02, percent-clipped=1.0 2023-04-16 22:10:19,933 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40922.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:10:21,785 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9557, 2.7132, 2.5846, 2.9799, 2.3322, 3.0619, 2.8945, 2.6581], device='cuda:1'), covar=tensor([0.0067, 0.0148, 0.0133, 0.0130, 0.0197, 0.0112, 0.0164, 0.0141], device='cuda:1'), in_proj_covar=tensor([0.0090, 0.0097, 0.0104, 0.0099, 0.0111, 0.0089, 0.0090, 0.0090], device='cuda:1'), out_proj_covar=tensor([9.5729e-05, 1.0566e-04, 1.1572e-04, 1.0811e-04, 1.2367e-04, 9.5136e-05, 9.8117e-05, 9.6489e-05], device='cuda:1') 2023-04-16 22:10:23,517 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40926.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:10:25,191 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40928.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:10:34,479 INFO [train.py:893] (1/4) Epoch 16, batch 450, loss[loss=0.1934, simple_loss=0.2411, pruned_loss=0.07284, over 13360.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.254, pruned_loss=0.0725, over 2372140.01 frames. ], batch size: 73, lr: 8.30e-03, grad_scale: 32.0 2023-04-16 22:10:56,457 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=40961.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 22:10:59,385 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-16 22:11:10,934 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=40976.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:11:20,190 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=40986.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:11:20,743 INFO [train.py:893] (1/4) Epoch 16, batch 500, loss[loss=0.175, simple_loss=0.2391, pruned_loss=0.05541, over 13487.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2542, pruned_loss=0.07219, over 2440537.34 frames. ], batch size: 93, lr: 8.29e-03, grad_scale: 32.0 2023-04-16 22:11:30,061 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=40997.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:11:40,247 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41007.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:11:46,651 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.017e+02 2.721e+02 3.359e+02 3.870e+02 6.391e+02, threshold=6.718e+02, percent-clipped=0.0 2023-04-16 22:12:05,233 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41034.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:12:08,067 INFO [train.py:893] (1/4) Epoch 16, batch 550, loss[loss=0.2273, simple_loss=0.2788, pruned_loss=0.08795, over 13028.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.254, pruned_loss=0.07176, over 2490231.45 frames. ], batch size: 142, lr: 8.29e-03, grad_scale: 32.0 2023-04-16 22:12:10,904 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9599, 4.0680, 3.1986, 2.7150, 2.7893, 2.4932, 4.1319, 2.3660], device='cuda:1'), covar=tensor([0.1367, 0.0332, 0.0927, 0.1840, 0.0800, 0.2893, 0.0228, 0.3648], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0282, 0.0294, 0.0308, 0.0243, 0.0309, 0.0201, 0.0355], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 22:12:15,884 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41045.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:12:26,918 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.2872, 2.0995, 2.5375, 3.8325, 3.4650, 3.8707, 2.9632, 2.1304], device='cuda:1'), covar=tensor([0.0276, 0.1098, 0.0832, 0.0048, 0.0232, 0.0053, 0.0665, 0.1041], device='cuda:1'), in_proj_covar=tensor([0.0115, 0.0149, 0.0162, 0.0089, 0.0116, 0.0087, 0.0166, 0.0148], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-16 22:12:27,794 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41058.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:12:54,709 INFO [train.py:893] (1/4) Epoch 16, batch 600, loss[loss=0.1746, simple_loss=0.2254, pruned_loss=0.06186, over 13457.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2524, pruned_loss=0.07115, over 2530425.60 frames. ], batch size: 65, lr: 8.28e-03, grad_scale: 32.0 2023-04-16 22:13:09,386 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41102.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:13:12,652 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41106.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:13:19,899 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.056e+02 2.815e+02 3.188e+02 4.197e+02 8.444e+02, threshold=6.376e+02, percent-clipped=3.0 2023-04-16 22:13:41,404 INFO [train.py:893] (1/4) Epoch 16, batch 650, loss[loss=0.1844, simple_loss=0.2408, pruned_loss=0.064, over 13045.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2512, pruned_loss=0.07059, over 2559365.02 frames. ], batch size: 142, lr: 8.28e-03, grad_scale: 32.0 2023-04-16 22:13:51,639 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41148.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:14:05,890 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41163.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:14:28,920 INFO [train.py:893] (1/4) Epoch 16, batch 700, loss[loss=0.1734, simple_loss=0.2309, pruned_loss=0.05797, over 13538.00 frames. ], tot_loss[loss=0.1953, simple_loss=0.2504, pruned_loss=0.0701, over 2580755.47 frames. ], batch size: 83, lr: 8.27e-03, grad_scale: 32.0 2023-04-16 22:14:37,419 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41196.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:14:42,242 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-16 22:14:42,409 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.86 vs. limit=5.0 2023-04-16 22:14:54,221 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.098e+02 2.785e+02 3.219e+02 3.700e+02 6.141e+02, threshold=6.437e+02, percent-clipped=0.0 2023-04-16 22:15:01,288 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41221.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:15:09,486 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-04-16 22:15:16,660 INFO [train.py:893] (1/4) Epoch 16, batch 750, loss[loss=0.1805, simple_loss=0.2305, pruned_loss=0.06528, over 13374.00 frames. ], tot_loss[loss=0.1943, simple_loss=0.2491, pruned_loss=0.06969, over 2589296.97 frames. ], batch size: 62, lr: 8.27e-03, grad_scale: 32.0 2023-04-16 22:15:34,495 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41256.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 22:16:03,267 INFO [train.py:893] (1/4) Epoch 16, batch 800, loss[loss=0.165, simple_loss=0.2261, pruned_loss=0.05189, over 13498.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.2507, pruned_loss=0.07051, over 2604388.92 frames. ], batch size: 81, lr: 8.26e-03, grad_scale: 32.0 2023-04-16 22:16:16,093 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4526, 3.0108, 2.3503, 4.4204, 4.9634, 3.6703, 4.7506, 4.5583], device='cuda:1'), covar=tensor([0.0109, 0.0735, 0.0933, 0.0091, 0.0052, 0.0406, 0.0087, 0.0077], device='cuda:1'), in_proj_covar=tensor([0.0071, 0.0086, 0.0095, 0.0075, 0.0061, 0.0078, 0.0051, 0.0072], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 22:16:22,705 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41307.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:16:29,238 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.237e+02 2.904e+02 3.409e+02 3.941e+02 8.201e+02, threshold=6.817e+02, percent-clipped=2.0 2023-04-16 22:16:51,226 INFO [train.py:893] (1/4) Epoch 16, batch 850, loss[loss=0.1961, simple_loss=0.2473, pruned_loss=0.07247, over 13365.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2517, pruned_loss=0.07136, over 2620088.82 frames. ], batch size: 67, lr: 8.26e-03, grad_scale: 32.0 2023-04-16 22:17:06,370 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41353.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:17:08,120 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41355.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:17:36,890 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=2.81 vs. limit=5.0 2023-04-16 22:17:38,113 INFO [train.py:893] (1/4) Epoch 16, batch 900, loss[loss=0.1937, simple_loss=0.2394, pruned_loss=0.07398, over 13531.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2517, pruned_loss=0.07167, over 2634108.07 frames. ], batch size: 87, lr: 8.25e-03, grad_scale: 32.0 2023-04-16 22:17:38,420 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.2613, 2.0253, 3.9448, 3.7780, 3.7923, 3.0487, 3.6281, 2.9323], device='cuda:1'), covar=tensor([0.1806, 0.1512, 0.0089, 0.0184, 0.0160, 0.0654, 0.0215, 0.0702], device='cuda:1'), in_proj_covar=tensor([0.0208, 0.0181, 0.0111, 0.0120, 0.0122, 0.0170, 0.0133, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 22:17:52,155 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41401.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:18:03,459 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.089e+02 3.038e+02 3.486e+02 4.012e+02 7.844e+02, threshold=6.973e+02, percent-clipped=1.0 2023-04-16 22:18:09,454 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6919, 3.4333, 2.8461, 3.1829, 2.8615, 2.0266, 3.4588, 1.8895], device='cuda:1'), covar=tensor([0.0715, 0.0580, 0.0497, 0.0426, 0.0708, 0.1956, 0.1176, 0.1381], device='cuda:1'), in_proj_covar=tensor([0.0132, 0.0128, 0.0123, 0.0106, 0.0142, 0.0177, 0.0154, 0.0134], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 22:18:09,951 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-16 22:18:24,931 INFO [train.py:893] (1/4) Epoch 16, batch 950, loss[loss=0.1632, simple_loss=0.2214, pruned_loss=0.05245, over 13365.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2505, pruned_loss=0.07147, over 2642740.73 frames. ], batch size: 73, lr: 8.25e-03, grad_scale: 32.0 2023-04-16 22:18:31,539 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.75 vs. limit=5.0 2023-04-16 22:18:45,287 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=41458.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:19:11,655 INFO [train.py:893] (1/4) Epoch 16, batch 1000, loss[loss=0.19, simple_loss=0.2403, pruned_loss=0.06984, over 13383.00 frames. ], tot_loss[loss=0.1945, simple_loss=0.2483, pruned_loss=0.07034, over 2647315.15 frames. ], batch size: 62, lr: 8.24e-03, grad_scale: 32.0 2023-04-16 22:19:14,398 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0267, 4.3984, 4.1811, 4.1552, 4.1864, 4.0364, 4.4611, 4.5030], device='cuda:1'), covar=tensor([0.0241, 0.0217, 0.0210, 0.0403, 0.0253, 0.0258, 0.0264, 0.0229], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0185, 0.0144, 0.0168, 0.0132, 0.0182, 0.0122, 0.0158], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 22:19:37,201 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.105e+02 2.855e+02 3.387e+02 3.992e+02 7.617e+02, threshold=6.774e+02, percent-clipped=2.0 2023-04-16 22:19:42,449 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0876, 4.5678, 4.3409, 4.3345, 4.3205, 4.1011, 4.5984, 4.6426], device='cuda:1'), covar=tensor([0.0227, 0.0220, 0.0195, 0.0339, 0.0285, 0.0321, 0.0275, 0.0226], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0185, 0.0143, 0.0167, 0.0132, 0.0181, 0.0123, 0.0157], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 22:19:44,056 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41521.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:19:58,469 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7479, 4.1965, 3.9898, 3.9816, 3.9876, 3.8407, 4.2073, 4.2973], device='cuda:1'), covar=tensor([0.0420, 0.0331, 0.0300, 0.0442, 0.0400, 0.0390, 0.0395, 0.0296], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0185, 0.0143, 0.0167, 0.0131, 0.0181, 0.0123, 0.0156], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 22:19:59,062 INFO [train.py:893] (1/4) Epoch 16, batch 1050, loss[loss=0.2016, simple_loss=0.2572, pruned_loss=0.07298, over 13249.00 frames. ], tot_loss[loss=0.1917, simple_loss=0.2459, pruned_loss=0.06878, over 2646582.54 frames. ], batch size: 124, lr: 8.24e-03, grad_scale: 32.0 2023-04-16 22:20:12,963 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-04-16 22:20:16,862 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41556.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 22:20:29,122 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41569.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:20:45,605 INFO [train.py:893] (1/4) Epoch 16, batch 1100, loss[loss=0.2081, simple_loss=0.2541, pruned_loss=0.08102, over 11752.00 frames. ], tot_loss[loss=0.1916, simple_loss=0.2464, pruned_loss=0.0684, over 2650397.42 frames. ], batch size: 157, lr: 8.23e-03, grad_scale: 32.0 2023-04-16 22:21:02,301 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41604.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:21:11,235 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.019e+02 2.712e+02 3.141e+02 3.905e+02 7.780e+02, threshold=6.282e+02, percent-clipped=1.0 2023-04-16 22:21:19,142 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0694, 4.4768, 4.1539, 4.1727, 4.1882, 4.6096, 4.3473, 4.3378], device='cuda:1'), covar=tensor([0.0323, 0.0259, 0.0342, 0.0965, 0.0364, 0.0247, 0.0317, 0.0278], device='cuda:1'), in_proj_covar=tensor([0.0143, 0.0140, 0.0158, 0.0252, 0.0158, 0.0174, 0.0156, 0.0106], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 22:21:34,301 INFO [train.py:893] (1/4) Epoch 16, batch 1150, loss[loss=0.1791, simple_loss=0.2423, pruned_loss=0.05796, over 13378.00 frames. ], tot_loss[loss=0.1905, simple_loss=0.2462, pruned_loss=0.06747, over 2651909.14 frames. ], batch size: 109, lr: 8.23e-03, grad_scale: 32.0 2023-04-16 22:21:48,718 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41653.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:21:49,622 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41654.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:22:20,426 INFO [train.py:893] (1/4) Epoch 16, batch 1200, loss[loss=0.1695, simple_loss=0.2362, pruned_loss=0.05139, over 13371.00 frames. ], tot_loss[loss=0.1904, simple_loss=0.2464, pruned_loss=0.0672, over 2647614.22 frames. ], batch size: 118, lr: 8.23e-03, grad_scale: 16.0 2023-04-16 22:22:28,112 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 2023-04-16 22:22:34,030 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41701.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:22:34,083 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41701.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:22:46,472 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.687e+02 2.840e+02 3.425e+02 4.080e+02 1.396e+03, threshold=6.851e+02, percent-clipped=2.0 2023-04-16 22:22:46,832 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41715.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 22:22:47,453 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-16 22:22:59,784 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-16 22:23:06,421 INFO [train.py:893] (1/4) Epoch 16, batch 1250, loss[loss=0.2108, simple_loss=0.2659, pruned_loss=0.0779, over 13578.00 frames. ], tot_loss[loss=0.191, simple_loss=0.2469, pruned_loss=0.0675, over 2651901.29 frames. ], batch size: 89, lr: 8.22e-03, grad_scale: 16.0 2023-04-16 22:23:16,950 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7367, 2.3736, 2.3590, 2.6972, 1.9759, 2.7917, 2.6796, 2.3459], device='cuda:1'), covar=tensor([0.0069, 0.0163, 0.0136, 0.0142, 0.0205, 0.0115, 0.0172, 0.0154], device='cuda:1'), in_proj_covar=tensor([0.0091, 0.0097, 0.0106, 0.0100, 0.0111, 0.0091, 0.0091, 0.0091], device='cuda:1'), out_proj_covar=tensor([9.5664e-05, 1.0490e-04, 1.1716e-04, 1.0922e-04, 1.2309e-04, 9.7171e-05, 9.9401e-05, 9.7191e-05], device='cuda:1') 2023-04-16 22:23:17,622 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41749.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:23:25,491 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9681, 4.5234, 4.2656, 4.2715, 4.2656, 4.1498, 4.5493, 4.6048], device='cuda:1'), covar=tensor([0.0230, 0.0193, 0.0171, 0.0314, 0.0280, 0.0227, 0.0257, 0.0203], device='cuda:1'), in_proj_covar=tensor([0.0123, 0.0182, 0.0142, 0.0165, 0.0130, 0.0179, 0.0122, 0.0155], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 22:23:26,951 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=41758.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:23:52,548 INFO [train.py:893] (1/4) Epoch 16, batch 1300, loss[loss=0.1806, simple_loss=0.2411, pruned_loss=0.06006, over 13543.00 frames. ], tot_loss[loss=0.1932, simple_loss=0.2491, pruned_loss=0.06867, over 2652012.91 frames. ], batch size: 78, lr: 8.22e-03, grad_scale: 16.0 2023-04-16 22:23:59,350 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1092, 4.4256, 4.2044, 4.2049, 4.1984, 4.5962, 4.4092, 4.3045], device='cuda:1'), covar=tensor([0.0322, 0.0285, 0.0314, 0.0966, 0.0334, 0.0264, 0.0286, 0.0288], device='cuda:1'), in_proj_covar=tensor([0.0144, 0.0141, 0.0158, 0.0256, 0.0159, 0.0176, 0.0156, 0.0107], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 22:24:11,501 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=41806.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:24:18,910 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.763e+02 2.850e+02 3.302e+02 3.897e+02 7.866e+02, threshold=6.605e+02, percent-clipped=2.0 2023-04-16 22:24:40,585 INFO [train.py:893] (1/4) Epoch 16, batch 1350, loss[loss=0.1776, simple_loss=0.2372, pruned_loss=0.05898, over 13544.00 frames. ], tot_loss[loss=0.193, simple_loss=0.2488, pruned_loss=0.06856, over 2653281.50 frames. ], batch size: 78, lr: 8.21e-03, grad_scale: 16.0 2023-04-16 22:25:27,120 INFO [train.py:893] (1/4) Epoch 16, batch 1400, loss[loss=0.2022, simple_loss=0.2528, pruned_loss=0.07582, over 13429.00 frames. ], tot_loss[loss=0.1926, simple_loss=0.248, pruned_loss=0.06863, over 2654935.01 frames. ], batch size: 95, lr: 8.21e-03, grad_scale: 16.0 2023-04-16 22:25:45,095 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41906.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:25:53,692 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.042e+02 2.927e+02 3.297e+02 3.876e+02 1.002e+03, threshold=6.593e+02, percent-clipped=4.0 2023-04-16 22:26:13,621 INFO [train.py:893] (1/4) Epoch 16, batch 1450, loss[loss=0.2031, simple_loss=0.2561, pruned_loss=0.07508, over 13370.00 frames. ], tot_loss[loss=0.1934, simple_loss=0.2483, pruned_loss=0.06927, over 2655134.02 frames. ], batch size: 118, lr: 8.20e-03, grad_scale: 16.0 2023-04-16 22:26:24,814 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=41948.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 22:26:42,313 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=41967.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:26:45,773 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.8118, 2.5963, 2.0383, 1.4201, 1.5758, 2.3019, 2.1592, 2.8392], device='cuda:1'), covar=tensor([0.0915, 0.0323, 0.0798, 0.1605, 0.0268, 0.0509, 0.0778, 0.0189], device='cuda:1'), in_proj_covar=tensor([0.0154, 0.0127, 0.0109, 0.0195, 0.0102, 0.0146, 0.0155, 0.0120], device='cuda:1'), out_proj_covar=tensor([1.1509e-04, 9.5449e-05, 8.6285e-05, 1.4792e-04, 7.7044e-05, 1.1014e-04, 1.1861e-04, 8.8777e-05], device='cuda:1') 2023-04-16 22:26:48,245 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.5412, 3.2225, 2.5967, 2.6913, 2.6026, 1.9315, 3.2239, 1.7906], device='cuda:1'), covar=tensor([0.0660, 0.0665, 0.0501, 0.0490, 0.0756, 0.1877, 0.1160, 0.1465], device='cuda:1'), in_proj_covar=tensor([0.0132, 0.0130, 0.0126, 0.0108, 0.0143, 0.0179, 0.0156, 0.0135], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 22:26:51,494 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0355, 3.8316, 3.0373, 3.6244, 3.0412, 2.4072, 3.7840, 1.9838], device='cuda:1'), covar=tensor([0.0637, 0.0516, 0.0492, 0.0259, 0.0702, 0.1604, 0.1063, 0.1378], device='cuda:1'), in_proj_covar=tensor([0.0132, 0.0130, 0.0126, 0.0108, 0.0143, 0.0179, 0.0157, 0.0135], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 22:27:00,187 INFO [train.py:893] (1/4) Epoch 16, batch 1500, loss[loss=0.1808, simple_loss=0.2265, pruned_loss=0.06758, over 13188.00 frames. ], tot_loss[loss=0.1938, simple_loss=0.2492, pruned_loss=0.06923, over 2658107.17 frames. ], batch size: 58, lr: 8.20e-03, grad_scale: 16.0 2023-04-16 22:27:24,210 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.07 vs. limit=5.0 2023-04-16 22:27:26,469 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42009.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 22:27:27,161 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42010.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 22:27:31,082 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.141e+02 2.694e+02 3.135e+02 3.781e+02 7.515e+02, threshold=6.271e+02, percent-clipped=2.0 2023-04-16 22:27:52,311 INFO [train.py:893] (1/4) Epoch 16, batch 1550, loss[loss=0.1825, simple_loss=0.2376, pruned_loss=0.0637, over 13370.00 frames. ], tot_loss[loss=0.1933, simple_loss=0.2491, pruned_loss=0.06877, over 2660138.86 frames. ], batch size: 67, lr: 8.19e-03, grad_scale: 16.0 2023-04-16 22:28:39,477 INFO [train.py:893] (1/4) Epoch 16, batch 1600, loss[loss=0.1781, simple_loss=0.2444, pruned_loss=0.05592, over 13460.00 frames. ], tot_loss[loss=0.1935, simple_loss=0.2495, pruned_loss=0.06871, over 2661076.65 frames. ], batch size: 106, lr: 8.19e-03, grad_scale: 16.0 2023-04-16 22:29:02,543 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0075, 4.2816, 4.0831, 4.0887, 4.1465, 4.4517, 4.2813, 4.0364], device='cuda:1'), covar=tensor([0.0302, 0.0283, 0.0331, 0.0896, 0.0306, 0.0246, 0.0304, 0.0430], device='cuda:1'), in_proj_covar=tensor([0.0145, 0.0142, 0.0159, 0.0256, 0.0160, 0.0178, 0.0157, 0.0106], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 22:29:05,744 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.862e+02 2.671e+02 3.172e+02 3.980e+02 8.718e+02, threshold=6.343e+02, percent-clipped=3.0 2023-04-16 22:29:27,688 INFO [train.py:893] (1/4) Epoch 16, batch 1650, loss[loss=0.211, simple_loss=0.2684, pruned_loss=0.07678, over 13487.00 frames. ], tot_loss[loss=0.1928, simple_loss=0.2492, pruned_loss=0.0682, over 2659310.50 frames. ], batch size: 93, lr: 8.18e-03, grad_scale: 16.0 2023-04-16 22:29:28,219 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.22 vs. limit=5.0 2023-04-16 22:29:39,754 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3704, 1.9410, 2.3529, 3.8508, 3.4557, 3.8348, 2.9075, 2.0518], device='cuda:1'), covar=tensor([0.0292, 0.1409, 0.1115, 0.0059, 0.0272, 0.0077, 0.0761, 0.1279], device='cuda:1'), in_proj_covar=tensor([0.0116, 0.0148, 0.0164, 0.0089, 0.0117, 0.0086, 0.0166, 0.0146], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-16 22:30:14,897 INFO [train.py:893] (1/4) Epoch 16, batch 1700, loss[loss=0.1945, simple_loss=0.2492, pruned_loss=0.06993, over 13561.00 frames. ], tot_loss[loss=0.1927, simple_loss=0.2495, pruned_loss=0.06792, over 2664127.99 frames. ], batch size: 89, lr: 8.18e-03, grad_scale: 16.0 2023-04-16 22:30:41,189 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.067e+02 2.979e+02 3.408e+02 4.299e+02 7.415e+02, threshold=6.815e+02, percent-clipped=3.0 2023-04-16 22:31:02,236 INFO [train.py:893] (1/4) Epoch 16, batch 1750, loss[loss=0.2001, simple_loss=0.2514, pruned_loss=0.07434, over 13376.00 frames. ], tot_loss[loss=0.1915, simple_loss=0.2486, pruned_loss=0.0672, over 2665436.19 frames. ], batch size: 113, lr: 8.17e-03, grad_scale: 16.0 2023-04-16 22:31:20,755 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.3877, 2.0993, 4.2259, 3.9002, 4.1526, 3.2932, 3.9117, 3.0598], device='cuda:1'), covar=tensor([0.1753, 0.1616, 0.0122, 0.0221, 0.0182, 0.0578, 0.0201, 0.0795], device='cuda:1'), in_proj_covar=tensor([0.0205, 0.0180, 0.0111, 0.0118, 0.0122, 0.0168, 0.0132, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 22:31:26,451 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42262.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:31:26,531 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4403, 4.7216, 4.3850, 4.5036, 4.5304, 4.9023, 4.6346, 4.5135], device='cuda:1'), covar=tensor([0.0326, 0.0251, 0.0334, 0.0894, 0.0286, 0.0210, 0.0317, 0.0223], device='cuda:1'), in_proj_covar=tensor([0.0145, 0.0141, 0.0160, 0.0256, 0.0160, 0.0177, 0.0157, 0.0106], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 22:31:46,809 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.8998, 4.3757, 4.3124, 4.4373, 4.2585, 4.2255, 4.8697, 4.4118], device='cuda:1'), covar=tensor([0.0717, 0.1253, 0.2492, 0.2685, 0.0939, 0.1504, 0.1010, 0.1268], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0343, 0.0436, 0.0443, 0.0262, 0.0326, 0.0392, 0.0315], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 22:31:49,970 INFO [train.py:893] (1/4) Epoch 16, batch 1800, loss[loss=0.19, simple_loss=0.2501, pruned_loss=0.06494, over 13439.00 frames. ], tot_loss[loss=0.1893, simple_loss=0.2467, pruned_loss=0.06599, over 2666210.24 frames. ], batch size: 103, lr: 8.17e-03, grad_scale: 16.0 2023-04-16 22:32:04,641 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42304.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 22:32:10,523 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42310.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 22:32:15,943 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.816e+02 2.666e+02 3.144e+02 3.768e+02 7.198e+02, threshold=6.287e+02, percent-clipped=1.0 2023-04-16 22:32:26,578 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8043, 4.2678, 4.0968, 4.0556, 4.0241, 3.9101, 4.2901, 4.3534], device='cuda:1'), covar=tensor([0.0229, 0.0220, 0.0211, 0.0362, 0.0301, 0.0280, 0.0309, 0.0254], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0185, 0.0145, 0.0168, 0.0131, 0.0182, 0.0123, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 22:32:36,104 INFO [train.py:893] (1/4) Epoch 16, batch 1850, loss[loss=0.1931, simple_loss=0.2374, pruned_loss=0.07443, over 13416.00 frames. ], tot_loss[loss=0.1898, simple_loss=0.2469, pruned_loss=0.06642, over 2662004.71 frames. ], batch size: 65, lr: 8.16e-03, grad_scale: 16.0 2023-04-16 22:32:39,367 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-16 22:32:56,309 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42358.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:33:03,986 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8105, 2.5512, 2.5476, 2.8928, 2.3019, 3.0736, 3.0008, 2.5338], device='cuda:1'), covar=tensor([0.0101, 0.0184, 0.0187, 0.0157, 0.0218, 0.0107, 0.0158, 0.0168], device='cuda:1'), in_proj_covar=tensor([0.0093, 0.0098, 0.0109, 0.0103, 0.0113, 0.0092, 0.0094, 0.0092], device='cuda:1'), out_proj_covar=tensor([9.8496e-05, 1.0630e-04, 1.2025e-04, 1.1183e-04, 1.2485e-04, 9.8083e-05, 1.0269e-04, 9.8072e-05], device='cuda:1') 2023-04-16 22:33:16,537 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42379.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:33:23,834 INFO [train.py:893] (1/4) Epoch 16, batch 1900, loss[loss=0.1963, simple_loss=0.2497, pruned_loss=0.07148, over 13520.00 frames. ], tot_loss[loss=0.1899, simple_loss=0.2466, pruned_loss=0.06655, over 2663590.05 frames. ], batch size: 85, lr: 8.16e-03, grad_scale: 16.0 2023-04-16 22:33:50,090 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2303, 4.3989, 2.9242, 4.1987, 4.1995, 2.5913, 3.7323, 2.8828], device='cuda:1'), covar=tensor([0.0297, 0.0263, 0.1096, 0.0348, 0.0218, 0.1290, 0.0539, 0.1358], device='cuda:1'), in_proj_covar=tensor([0.0151, 0.0162, 0.0174, 0.0180, 0.0131, 0.0158, 0.0158, 0.0159], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 22:33:50,544 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.457e+02 2.816e+02 3.160e+02 3.935e+02 6.934e+02, threshold=6.320e+02, percent-clipped=1.0 2023-04-16 22:34:09,987 INFO [train.py:893] (1/4) Epoch 16, batch 1950, loss[loss=0.205, simple_loss=0.2636, pruned_loss=0.07322, over 13465.00 frames. ], tot_loss[loss=0.1906, simple_loss=0.2477, pruned_loss=0.06679, over 2663807.57 frames. ], batch size: 100, lr: 8.15e-03, grad_scale: 16.0 2023-04-16 22:34:14,606 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42440.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:34:17,203 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-04-16 22:34:57,423 INFO [train.py:893] (1/4) Epoch 16, batch 2000, loss[loss=0.2034, simple_loss=0.2586, pruned_loss=0.07413, over 13184.00 frames. ], tot_loss[loss=0.1907, simple_loss=0.2477, pruned_loss=0.06688, over 2665649.35 frames. ], batch size: 132, lr: 8.15e-03, grad_scale: 16.0 2023-04-16 22:35:00,350 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1224, 2.5237, 2.1320, 4.0134, 4.5592, 3.4676, 4.4606, 4.2102], device='cuda:1'), covar=tensor([0.0103, 0.0915, 0.1024, 0.0106, 0.0058, 0.0437, 0.0070, 0.0084], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0085, 0.0093, 0.0074, 0.0061, 0.0077, 0.0051, 0.0071], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:1') 2023-04-16 22:35:05,260 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-16 22:35:17,151 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42506.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:35:24,432 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.996e+02 2.838e+02 3.154e+02 3.694e+02 1.052e+03, threshold=6.307e+02, percent-clipped=1.0 2023-04-16 22:35:46,082 INFO [train.py:893] (1/4) Epoch 16, batch 2050, loss[loss=0.215, simple_loss=0.2725, pruned_loss=0.07876, over 13360.00 frames. ], tot_loss[loss=0.1927, simple_loss=0.2492, pruned_loss=0.06811, over 2664367.64 frames. ], batch size: 118, lr: 8.14e-03, grad_scale: 16.0 2023-04-16 22:35:53,421 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.51 vs. limit=2.0 2023-04-16 22:35:58,907 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0337, 4.6464, 4.4421, 4.3982, 4.4018, 4.2318, 4.6584, 4.7080], device='cuda:1'), covar=tensor([0.0260, 0.0189, 0.0191, 0.0321, 0.0256, 0.0266, 0.0271, 0.0218], device='cuda:1'), in_proj_covar=tensor([0.0125, 0.0186, 0.0145, 0.0168, 0.0133, 0.0183, 0.0123, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 22:36:08,889 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42562.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:36:14,106 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42567.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:36:31,594 INFO [train.py:893] (1/4) Epoch 16, batch 2100, loss[loss=0.2234, simple_loss=0.2825, pruned_loss=0.08214, over 13330.00 frames. ], tot_loss[loss=0.1917, simple_loss=0.2482, pruned_loss=0.06753, over 2665383.32 frames. ], batch size: 118, lr: 8.14e-03, grad_scale: 16.0 2023-04-16 22:36:48,891 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=42604.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 22:36:53,945 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42610.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:36:57,949 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.948e+02 2.754e+02 3.328e+02 3.868e+02 6.469e+02, threshold=6.656e+02, percent-clipped=2.0 2023-04-16 22:37:19,348 INFO [train.py:893] (1/4) Epoch 16, batch 2150, loss[loss=0.2226, simple_loss=0.2737, pruned_loss=0.0858, over 13032.00 frames. ], tot_loss[loss=0.192, simple_loss=0.2489, pruned_loss=0.06752, over 2661626.21 frames. ], batch size: 142, lr: 8.13e-03, grad_scale: 16.0 2023-04-16 22:37:29,666 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42649.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:37:32,910 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=42652.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 22:38:05,985 INFO [train.py:893] (1/4) Epoch 16, batch 2200, loss[loss=0.1934, simple_loss=0.2469, pruned_loss=0.06997, over 13011.00 frames. ], tot_loss[loss=0.1921, simple_loss=0.249, pruned_loss=0.06762, over 2657071.70 frames. ], batch size: 142, lr: 8.13e-03, grad_scale: 16.0 2023-04-16 22:38:12,245 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8457, 2.5329, 2.6186, 2.9443, 2.2148, 3.0746, 3.0004, 2.5512], device='cuda:1'), covar=tensor([0.0072, 0.0169, 0.0128, 0.0134, 0.0210, 0.0099, 0.0143, 0.0133], device='cuda:1'), in_proj_covar=tensor([0.0094, 0.0100, 0.0110, 0.0103, 0.0114, 0.0093, 0.0095, 0.0095], device='cuda:1'), out_proj_covar=tensor([9.8809e-05, 1.0838e-04, 1.2151e-04, 1.1182e-04, 1.2640e-04, 9.9797e-05, 1.0284e-04, 1.0116e-04], device='cuda:1') 2023-04-16 22:38:27,152 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42710.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:38:31,883 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.739e+02 2.651e+02 3.047e+02 3.675e+02 6.847e+02, threshold=6.094e+02, percent-clipped=1.0 2023-04-16 22:38:50,859 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42735.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:38:52,284 INFO [train.py:893] (1/4) Epoch 16, batch 2250, loss[loss=0.1646, simple_loss=0.2199, pruned_loss=0.05468, over 13172.00 frames. ], tot_loss[loss=0.1895, simple_loss=0.2467, pruned_loss=0.06611, over 2655212.19 frames. ], batch size: 58, lr: 8.12e-03, grad_scale: 16.0 2023-04-16 22:39:39,969 INFO [train.py:893] (1/4) Epoch 16, batch 2300, loss[loss=0.1845, simple_loss=0.248, pruned_loss=0.0605, over 13486.00 frames. ], tot_loss[loss=0.188, simple_loss=0.2452, pruned_loss=0.06535, over 2656262.81 frames. ], batch size: 100, lr: 8.12e-03, grad_scale: 16.0 2023-04-16 22:39:57,463 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.4881, 4.9734, 4.9583, 5.0121, 4.8347, 4.8087, 5.4798, 4.9578], device='cuda:1'), covar=tensor([0.0854, 0.1053, 0.2328, 0.2765, 0.0843, 0.1718, 0.0929, 0.1053], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0347, 0.0442, 0.0450, 0.0265, 0.0330, 0.0396, 0.0319], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 22:40:05,866 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.686e+02 2.521e+02 3.073e+02 3.579e+02 5.794e+02, threshold=6.147e+02, percent-clipped=0.0 2023-04-16 22:40:25,374 INFO [train.py:893] (1/4) Epoch 16, batch 2350, loss[loss=0.1984, simple_loss=0.2567, pruned_loss=0.07007, over 13274.00 frames. ], tot_loss[loss=0.187, simple_loss=0.2444, pruned_loss=0.06485, over 2659474.04 frames. ], batch size: 124, lr: 8.12e-03, grad_scale: 16.0 2023-04-16 22:40:42,973 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6848, 4.4508, 4.7795, 4.6476, 5.0146, 4.5247, 5.0245, 5.0113], device='cuda:1'), covar=tensor([0.0406, 0.0653, 0.0663, 0.0530, 0.0560, 0.0836, 0.0506, 0.0524], device='cuda:1'), in_proj_covar=tensor([0.0265, 0.0276, 0.0277, 0.0209, 0.0396, 0.0319, 0.0245, 0.0274], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 22:40:47,119 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42859.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:40:49,483 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-16 22:40:49,645 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=42862.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:40:49,961 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-04-16 22:41:06,600 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=42879.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:41:13,719 INFO [train.py:893] (1/4) Epoch 16, batch 2400, loss[loss=0.2067, simple_loss=0.2623, pruned_loss=0.07549, over 13570.00 frames. ], tot_loss[loss=0.1874, simple_loss=0.2443, pruned_loss=0.06531, over 2659094.23 frames. ], batch size: 89, lr: 8.11e-03, grad_scale: 16.0 2023-04-16 22:41:39,543 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.683e+02 2.661e+02 3.119e+02 3.858e+02 1.033e+03, threshold=6.237e+02, percent-clipped=2.0 2023-04-16 22:41:44,068 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42920.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:41:46,942 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-04-16 22:41:54,363 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4204, 3.4470, 4.1167, 2.9523, 2.6661, 2.7890, 4.3610, 4.4590], device='cuda:1'), covar=tensor([0.1649, 0.1619, 0.0362, 0.1678, 0.1519, 0.1494, 0.0261, 0.0248], device='cuda:1'), in_proj_covar=tensor([0.0236, 0.0254, 0.0185, 0.0216, 0.0211, 0.0177, 0.0193, 0.0163], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 22:41:58,912 INFO [train.py:893] (1/4) Epoch 16, batch 2450, loss[loss=0.1948, simple_loss=0.2553, pruned_loss=0.06711, over 13526.00 frames. ], tot_loss[loss=0.1881, simple_loss=0.2449, pruned_loss=0.06563, over 2659192.51 frames. ], batch size: 98, lr: 8.11e-03, grad_scale: 16.0 2023-04-16 22:42:02,606 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=42940.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:42:14,246 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8521, 3.9357, 3.9489, 3.4086, 3.8609, 4.1156, 4.0531, 3.7843], device='cuda:1'), covar=tensor([0.0279, 0.0331, 0.0296, 0.1640, 0.0325, 0.0365, 0.0344, 0.0741], device='cuda:1'), in_proj_covar=tensor([0.0146, 0.0144, 0.0159, 0.0257, 0.0164, 0.0179, 0.0159, 0.0108], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 22:42:46,443 INFO [train.py:893] (1/4) Epoch 16, batch 2500, loss[loss=0.176, simple_loss=0.236, pruned_loss=0.05802, over 13053.00 frames. ], tot_loss[loss=0.1872, simple_loss=0.244, pruned_loss=0.06525, over 2648134.16 frames. ], batch size: 142, lr: 8.10e-03, grad_scale: 16.0 2023-04-16 22:43:03,559 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.91 vs. limit=5.0 2023-04-16 22:43:04,006 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43005.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:43:13,892 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.663e+02 2.605e+02 2.966e+02 3.580e+02 6.970e+02, threshold=5.932e+02, percent-clipped=2.0 2023-04-16 22:43:17,136 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-04-16 22:43:32,651 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43035.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:43:33,436 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2216, 2.4062, 1.9538, 4.1457, 4.6316, 3.4551, 4.5488, 4.3204], device='cuda:1'), covar=tensor([0.0098, 0.0933, 0.1032, 0.0090, 0.0074, 0.0420, 0.0074, 0.0081], device='cuda:1'), in_proj_covar=tensor([0.0071, 0.0087, 0.0094, 0.0076, 0.0062, 0.0078, 0.0051, 0.0072], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 22:43:34,008 INFO [train.py:893] (1/4) Epoch 16, batch 2550, loss[loss=0.1915, simple_loss=0.245, pruned_loss=0.06907, over 13410.00 frames. ], tot_loss[loss=0.1868, simple_loss=0.2438, pruned_loss=0.06492, over 2654199.91 frames. ], batch size: 88, lr: 8.10e-03, grad_scale: 16.0 2023-04-16 22:43:58,317 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-16 22:44:12,133 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3138, 3.2575, 3.8954, 2.7817, 2.5259, 2.6702, 4.1388, 4.2849], device='cuda:1'), covar=tensor([0.1102, 0.1634, 0.0401, 0.1582, 0.1538, 0.1460, 0.0277, 0.0200], device='cuda:1'), in_proj_covar=tensor([0.0231, 0.0249, 0.0181, 0.0211, 0.0207, 0.0173, 0.0190, 0.0159], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 22:44:16,822 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43083.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:44:20,083 INFO [train.py:893] (1/4) Epoch 16, batch 2600, loss[loss=0.1891, simple_loss=0.2468, pruned_loss=0.06572, over 13519.00 frames. ], tot_loss[loss=0.1877, simple_loss=0.2443, pruned_loss=0.06553, over 2657731.38 frames. ], batch size: 91, lr: 8.09e-03, grad_scale: 16.0 2023-04-16 22:44:46,500 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.090e+02 2.759e+02 3.409e+02 4.306e+02 1.371e+03, threshold=6.818e+02, percent-clipped=5.0 2023-04-16 22:45:03,096 INFO [train.py:893] (1/4) Epoch 16, batch 2650, loss[loss=0.2035, simple_loss=0.2534, pruned_loss=0.07679, over 13353.00 frames. ], tot_loss[loss=0.1893, simple_loss=0.2455, pruned_loss=0.06653, over 2659933.45 frames. ], batch size: 118, lr: 8.09e-03, grad_scale: 16.0 2023-04-16 22:45:22,170 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43162.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:46:00,016 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 22:46:10,670 INFO [train.py:893] (1/4) Epoch 17, batch 0, loss[loss=0.246, simple_loss=0.2837, pruned_loss=0.1042, over 13557.00 frames. ], tot_loss[loss=0.246, simple_loss=0.2837, pruned_loss=0.1042, over 13557.00 frames. ], batch size: 87, lr: 7.84e-03, grad_scale: 16.0 2023-04-16 22:46:10,671 INFO [train.py:918] (1/4) Computing validation loss 2023-04-16 22:46:33,606 INFO [train.py:927] (1/4) Epoch 17, validation: loss=0.1429, simple_loss=0.2017, pruned_loss=0.04209, over 2446609.00 frames. 2023-04-16 22:46:33,607 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12829MB 2023-04-16 22:46:33,912 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9623, 1.9519, 3.7001, 3.6005, 3.5406, 2.6339, 3.3922, 2.7522], device='cuda:1'), covar=tensor([0.2050, 0.1382, 0.0124, 0.0197, 0.0219, 0.0820, 0.0241, 0.0750], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0181, 0.0113, 0.0122, 0.0125, 0.0169, 0.0134, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 22:46:55,939 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43210.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:47:01,609 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.048e+02 2.768e+02 3.169e+02 4.178e+02 7.323e+02, threshold=6.338e+02, percent-clipped=1.0 2023-04-16 22:47:01,819 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43215.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:47:20,227 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43235.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:47:20,827 INFO [train.py:893] (1/4) Epoch 17, batch 50, loss[loss=0.2149, simple_loss=0.2645, pruned_loss=0.08261, over 13463.00 frames. ], tot_loss[loss=0.1847, simple_loss=0.2391, pruned_loss=0.06512, over 600826.18 frames. ], batch size: 106, lr: 7.83e-03, grad_scale: 16.0 2023-04-16 22:47:23,704 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4394, 3.2732, 4.0282, 2.8665, 2.6851, 2.7831, 4.2247, 4.3915], device='cuda:1'), covar=tensor([0.1036, 0.1886, 0.0309, 0.1556, 0.1366, 0.1343, 0.0250, 0.0193], device='cuda:1'), in_proj_covar=tensor([0.0232, 0.0249, 0.0181, 0.0213, 0.0207, 0.0174, 0.0191, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 22:47:25,474 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1988, 4.3713, 3.4734, 3.0933, 3.1833, 2.5995, 4.5093, 2.6485], device='cuda:1'), covar=tensor([0.1419, 0.0293, 0.0946, 0.1686, 0.0686, 0.2994, 0.0184, 0.3284], device='cuda:1'), in_proj_covar=tensor([0.0302, 0.0279, 0.0298, 0.0311, 0.0245, 0.0313, 0.0200, 0.0353], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 22:47:26,479 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-04-16 22:47:45,121 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-16 22:47:45,122 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-16 22:47:45,122 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-16 22:47:45,128 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-16 22:47:45,136 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-16 22:47:45,158 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-16 22:47:45,174 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-16 22:48:06,954 INFO [train.py:893] (1/4) Epoch 17, batch 100, loss[loss=0.2037, simple_loss=0.2605, pruned_loss=0.07346, over 13099.00 frames. ], tot_loss[loss=0.187, simple_loss=0.2414, pruned_loss=0.06634, over 1060167.84 frames. ], batch size: 142, lr: 7.83e-03, grad_scale: 16.0 2023-04-16 22:48:11,567 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.3117, 2.5493, 4.3683, 3.8999, 4.1391, 3.3888, 3.9179, 2.9836], device='cuda:1'), covar=tensor([0.1862, 0.1196, 0.0070, 0.0230, 0.0167, 0.0529, 0.0182, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0183, 0.0114, 0.0123, 0.0127, 0.0171, 0.0135, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 22:48:18,490 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-04-16 22:48:20,969 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-04-16 22:48:24,952 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43305.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:48:34,474 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.195e+02 2.957e+02 3.496e+02 4.133e+02 8.444e+02, threshold=6.991e+02, percent-clipped=1.0 2023-04-16 22:48:53,794 INFO [train.py:893] (1/4) Epoch 17, batch 150, loss[loss=0.2187, simple_loss=0.274, pruned_loss=0.08176, over 13472.00 frames. ], tot_loss[loss=0.1898, simple_loss=0.2434, pruned_loss=0.06813, over 1407541.23 frames. ], batch size: 100, lr: 7.83e-03, grad_scale: 16.0 2023-04-16 22:49:02,442 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.1991, 1.8832, 4.0768, 3.8565, 3.8877, 2.9518, 3.7400, 2.9167], device='cuda:1'), covar=tensor([0.1895, 0.1624, 0.0085, 0.0232, 0.0211, 0.0796, 0.0199, 0.0797], device='cuda:1'), in_proj_covar=tensor([0.0215, 0.0185, 0.0115, 0.0124, 0.0128, 0.0172, 0.0136, 0.0181], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 22:49:09,911 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43353.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:49:41,157 INFO [train.py:893] (1/4) Epoch 17, batch 200, loss[loss=0.1664, simple_loss=0.2069, pruned_loss=0.06297, over 8026.00 frames. ], tot_loss[loss=0.1914, simple_loss=0.245, pruned_loss=0.06888, over 1675359.96 frames. ], batch size: 31, lr: 7.82e-03, grad_scale: 16.0 2023-04-16 22:50:08,279 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.91 vs. limit=5.0 2023-04-16 22:50:09,168 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.896e+02 2.754e+02 3.114e+02 3.666e+02 8.146e+02, threshold=6.228e+02, percent-clipped=1.0 2023-04-16 22:50:29,319 INFO [train.py:893] (1/4) Epoch 17, batch 250, loss[loss=0.1962, simple_loss=0.2471, pruned_loss=0.07261, over 13519.00 frames. ], tot_loss[loss=0.1936, simple_loss=0.2472, pruned_loss=0.07002, over 1889810.20 frames. ], batch size: 85, lr: 7.82e-03, grad_scale: 16.0 2023-04-16 22:50:59,434 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.79 vs. limit=5.0 2023-04-16 22:51:17,453 INFO [train.py:893] (1/4) Epoch 17, batch 300, loss[loss=0.2215, simple_loss=0.2642, pruned_loss=0.08937, over 11674.00 frames. ], tot_loss[loss=0.1929, simple_loss=0.2472, pruned_loss=0.06934, over 2048559.11 frames. ], batch size: 157, lr: 7.81e-03, grad_scale: 16.0 2023-04-16 22:51:23,482 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.9212, 4.3945, 4.3345, 4.3931, 4.1887, 4.2109, 4.8672, 4.4320], device='cuda:1'), covar=tensor([0.0699, 0.1243, 0.2109, 0.2750, 0.1105, 0.1557, 0.0954, 0.1148], device='cuda:1'), in_proj_covar=tensor([0.0301, 0.0344, 0.0437, 0.0448, 0.0263, 0.0329, 0.0400, 0.0319], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 22:51:43,632 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43514.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:51:44,181 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.065e+02 2.747e+02 3.096e+02 3.753e+02 1.093e+03, threshold=6.191e+02, percent-clipped=3.0 2023-04-16 22:51:44,437 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43515.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:51:55,131 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2023-04-16 22:52:03,577 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=43535.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:52:04,118 INFO [train.py:893] (1/4) Epoch 17, batch 350, loss[loss=0.2073, simple_loss=0.2639, pruned_loss=0.07532, over 13514.00 frames. ], tot_loss[loss=0.1934, simple_loss=0.2479, pruned_loss=0.06945, over 2183502.40 frames. ], batch size: 91, lr: 7.81e-03, grad_scale: 16.0 2023-04-16 22:52:29,652 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43563.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:52:40,620 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43575.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:52:41,048 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-04-16 22:52:48,954 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=43583.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:52:51,304 INFO [train.py:893] (1/4) Epoch 17, batch 400, loss[loss=0.2982, simple_loss=0.3355, pruned_loss=0.1304, over 11743.00 frames. ], tot_loss[loss=0.1948, simple_loss=0.2495, pruned_loss=0.07006, over 2285172.19 frames. ], batch size: 158, lr: 7.80e-03, grad_scale: 16.0 2023-04-16 22:53:00,684 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-16 22:53:18,637 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1665, 4.1524, 2.8437, 3.9847, 4.0358, 2.5395, 3.6036, 2.8826], device='cuda:1'), covar=tensor([0.0332, 0.0260, 0.1081, 0.0341, 0.0238, 0.1274, 0.0528, 0.1339], device='cuda:1'), in_proj_covar=tensor([0.0155, 0.0163, 0.0174, 0.0182, 0.0131, 0.0158, 0.0158, 0.0159], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 22:53:19,102 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.981e+02 2.711e+02 3.272e+02 3.876e+02 8.692e+02, threshold=6.544e+02, percent-clipped=2.0 2023-04-16 22:53:38,406 INFO [train.py:893] (1/4) Epoch 17, batch 450, loss[loss=0.1747, simple_loss=0.2248, pruned_loss=0.06232, over 12821.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.2504, pruned_loss=0.0703, over 2373127.00 frames. ], batch size: 52, lr: 7.80e-03, grad_scale: 16.0 2023-04-16 22:53:55,604 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8162, 3.9350, 3.8879, 3.4447, 3.7507, 4.1247, 4.0718, 3.7670], device='cuda:1'), covar=tensor([0.0362, 0.0336, 0.0441, 0.1492, 0.0461, 0.0390, 0.0388, 0.0829], device='cuda:1'), in_proj_covar=tensor([0.0147, 0.0144, 0.0160, 0.0257, 0.0164, 0.0180, 0.0158, 0.0107], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0006, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 22:54:04,438 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-16 22:54:26,047 INFO [train.py:893] (1/4) Epoch 17, batch 500, loss[loss=0.1922, simple_loss=0.2547, pruned_loss=0.06484, over 13346.00 frames. ], tot_loss[loss=0.1952, simple_loss=0.2504, pruned_loss=0.07001, over 2435099.34 frames. ], batch size: 118, lr: 7.79e-03, grad_scale: 32.0 2023-04-16 22:54:52,826 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.848e+02 2.764e+02 3.201e+02 3.711e+02 6.287e+02, threshold=6.402e+02, percent-clipped=0.0 2023-04-16 22:55:13,750 INFO [train.py:893] (1/4) Epoch 17, batch 550, loss[loss=0.1984, simple_loss=0.256, pruned_loss=0.07042, over 13442.00 frames. ], tot_loss[loss=0.1944, simple_loss=0.2499, pruned_loss=0.06949, over 2481821.93 frames. ], batch size: 103, lr: 7.79e-03, grad_scale: 32.0 2023-04-16 22:55:46,074 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-04-16 22:55:59,539 INFO [train.py:893] (1/4) Epoch 17, batch 600, loss[loss=0.1998, simple_loss=0.2542, pruned_loss=0.07275, over 13449.00 frames. ], tot_loss[loss=0.1931, simple_loss=0.2484, pruned_loss=0.0689, over 2518030.52 frames. ], batch size: 100, lr: 7.79e-03, grad_scale: 32.0 2023-04-16 22:56:00,588 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43787.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:56:17,287 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.02 vs. limit=5.0 2023-04-16 22:56:27,189 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9344, 3.9303, 3.1276, 3.6110, 3.1315, 2.1216, 3.9143, 2.2620], device='cuda:1'), covar=tensor([0.0716, 0.0423, 0.0526, 0.0323, 0.0780, 0.2154, 0.0798, 0.1396], device='cuda:1'), in_proj_covar=tensor([0.0135, 0.0131, 0.0127, 0.0111, 0.0145, 0.0181, 0.0160, 0.0136], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 22:56:27,681 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.795e+02 2.647e+02 3.088e+02 3.705e+02 5.223e+02, threshold=6.176e+02, percent-clipped=0.0 2023-04-16 22:56:35,470 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7775, 2.3390, 2.3101, 2.6993, 1.9695, 2.8984, 2.7769, 2.3239], device='cuda:1'), covar=tensor([0.0085, 0.0199, 0.0165, 0.0165, 0.0234, 0.0107, 0.0168, 0.0182], device='cuda:1'), in_proj_covar=tensor([0.0093, 0.0100, 0.0107, 0.0103, 0.0115, 0.0093, 0.0095, 0.0093], device='cuda:1'), out_proj_covar=tensor([9.7255e-05, 1.0850e-04, 1.1804e-04, 1.1120e-04, 1.2705e-04, 9.8770e-05, 1.0336e-04, 9.9249e-05], device='cuda:1') 2023-04-16 22:56:47,895 INFO [train.py:893] (1/4) Epoch 17, batch 650, loss[loss=0.1876, simple_loss=0.2449, pruned_loss=0.0651, over 13523.00 frames. ], tot_loss[loss=0.1927, simple_loss=0.2481, pruned_loss=0.06861, over 2555329.73 frames. ], batch size: 78, lr: 7.78e-03, grad_scale: 32.0 2023-04-16 22:56:58,901 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43848.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:57:19,632 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=43870.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:57:32,199 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.2108, 4.7777, 4.6543, 4.6333, 4.3597, 4.5347, 5.1529, 4.6578], device='cuda:1'), covar=tensor([0.0657, 0.1011, 0.1948, 0.2682, 0.1052, 0.1507, 0.0870, 0.1133], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0343, 0.0437, 0.0444, 0.0267, 0.0327, 0.0399, 0.0320], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 22:57:34,549 INFO [train.py:893] (1/4) Epoch 17, batch 700, loss[loss=0.1793, simple_loss=0.2427, pruned_loss=0.05801, over 13227.00 frames. ], tot_loss[loss=0.1908, simple_loss=0.247, pruned_loss=0.06735, over 2581894.92 frames. ], batch size: 124, lr: 7.78e-03, grad_scale: 32.0 2023-04-16 22:57:56,598 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43909.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:58:01,413 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.703e+02 2.597e+02 2.999e+02 3.684e+02 5.228e+02, threshold=5.998e+02, percent-clipped=0.0 2023-04-16 22:58:21,225 INFO [train.py:893] (1/4) Epoch 17, batch 750, loss[loss=0.2072, simple_loss=0.2545, pruned_loss=0.07991, over 13274.00 frames. ], tot_loss[loss=0.192, simple_loss=0.2476, pruned_loss=0.06817, over 2595645.25 frames. ], batch size: 117, lr: 7.77e-03, grad_scale: 32.0 2023-04-16 22:58:43,850 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=43959.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:58:53,167 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=43970.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:59:07,816 INFO [train.py:893] (1/4) Epoch 17, batch 800, loss[loss=0.1707, simple_loss=0.2372, pruned_loss=0.05211, over 13524.00 frames. ], tot_loss[loss=0.1921, simple_loss=0.2481, pruned_loss=0.06802, over 2609518.31 frames. ], batch size: 78, lr: 7.77e-03, grad_scale: 32.0 2023-04-16 22:59:38,395 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.890e+02 2.754e+02 3.341e+02 3.760e+02 1.385e+03, threshold=6.683e+02, percent-clipped=4.0 2023-04-16 22:59:44,366 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44020.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 22:59:51,852 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.9464, 4.7293, 4.9462, 4.7704, 5.2296, 4.6677, 5.2307, 5.2191], device='cuda:1'), covar=tensor([0.0347, 0.0553, 0.0646, 0.0602, 0.0488, 0.0825, 0.0411, 0.0438], device='cuda:1'), in_proj_covar=tensor([0.0267, 0.0279, 0.0284, 0.0213, 0.0402, 0.0327, 0.0251, 0.0278], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-16 22:59:56,768 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3340, 2.3082, 2.6133, 3.9134, 3.4969, 3.9549, 3.0187, 2.3094], device='cuda:1'), covar=tensor([0.0327, 0.0920, 0.0799, 0.0047, 0.0255, 0.0051, 0.0653, 0.0969], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0151, 0.0164, 0.0092, 0.0116, 0.0089, 0.0169, 0.0149], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-16 22:59:58,185 INFO [train.py:893] (1/4) Epoch 17, batch 850, loss[loss=0.2003, simple_loss=0.258, pruned_loss=0.07129, over 13545.00 frames. ], tot_loss[loss=0.1929, simple_loss=0.2491, pruned_loss=0.06837, over 2620857.03 frames. ], batch size: 87, lr: 7.76e-03, grad_scale: 32.0 2023-04-16 22:59:58,719 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-04-16 23:00:20,369 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-04-16 23:00:40,184 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.88 vs. limit=2.0 2023-04-16 23:00:44,720 INFO [train.py:893] (1/4) Epoch 17, batch 900, loss[loss=0.2104, simple_loss=0.2542, pruned_loss=0.08333, over 13055.00 frames. ], tot_loss[loss=0.1937, simple_loss=0.2495, pruned_loss=0.06892, over 2630284.49 frames. ], batch size: 142, lr: 7.76e-03, grad_scale: 32.0 2023-04-16 23:01:11,103 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.969e+02 2.735e+02 3.180e+02 3.676e+02 7.064e+02, threshold=6.360e+02, percent-clipped=1.0 2023-04-16 23:01:15,502 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-16 23:01:29,520 INFO [train.py:893] (1/4) Epoch 17, batch 950, loss[loss=0.18, simple_loss=0.2368, pruned_loss=0.06165, over 13468.00 frames. ], tot_loss[loss=0.1927, simple_loss=0.2482, pruned_loss=0.0686, over 2639678.99 frames. ], batch size: 100, lr: 7.75e-03, grad_scale: 32.0 2023-04-16 23:01:37,266 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44143.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:02:00,579 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.2586, 2.1583, 4.1411, 3.9521, 4.0759, 3.2702, 3.7888, 3.1851], device='cuda:1'), covar=tensor([0.1958, 0.1549, 0.0092, 0.0188, 0.0155, 0.0602, 0.0251, 0.0752], device='cuda:1'), in_proj_covar=tensor([0.0214, 0.0182, 0.0116, 0.0122, 0.0126, 0.0170, 0.0138, 0.0180], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 23:02:01,397 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44170.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:02:17,147 INFO [train.py:893] (1/4) Epoch 17, batch 1000, loss[loss=0.2176, simple_loss=0.2634, pruned_loss=0.08586, over 13518.00 frames. ], tot_loss[loss=0.1906, simple_loss=0.2457, pruned_loss=0.06774, over 2647906.25 frames. ], batch size: 76, lr: 7.75e-03, grad_scale: 32.0 2023-04-16 23:02:44,692 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.003e+02 2.659e+02 3.160e+02 3.851e+02 7.237e+02, threshold=6.320e+02, percent-clipped=1.0 2023-04-16 23:02:47,440 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44218.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:02:50,305 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-04-16 23:03:02,975 INFO [train.py:893] (1/4) Epoch 17, batch 1050, loss[loss=0.1814, simple_loss=0.2412, pruned_loss=0.06076, over 13452.00 frames. ], tot_loss[loss=0.189, simple_loss=0.2446, pruned_loss=0.06672, over 2655240.26 frames. ], batch size: 103, lr: 7.75e-03, grad_scale: 32.0 2023-04-16 23:03:29,713 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5750, 2.7552, 2.6172, 4.5355, 5.0595, 3.6937, 4.8120, 4.6357], device='cuda:1'), covar=tensor([0.0109, 0.0803, 0.0874, 0.0092, 0.0070, 0.0393, 0.0081, 0.0088], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0087, 0.0094, 0.0076, 0.0062, 0.0078, 0.0051, 0.0072], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 23:03:30,435 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44265.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:03:50,131 INFO [train.py:893] (1/4) Epoch 17, batch 1100, loss[loss=0.1851, simple_loss=0.2437, pruned_loss=0.0632, over 13504.00 frames. ], tot_loss[loss=0.1886, simple_loss=0.2447, pruned_loss=0.06627, over 2653035.42 frames. ], batch size: 93, lr: 7.74e-03, grad_scale: 32.0 2023-04-16 23:04:16,538 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.587e+02 2.690e+02 3.160e+02 3.946e+02 7.228e+02, threshold=6.319e+02, percent-clipped=1.0 2023-04-16 23:04:16,808 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44315.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:04:36,902 INFO [train.py:893] (1/4) Epoch 17, batch 1150, loss[loss=0.1914, simple_loss=0.2597, pruned_loss=0.06151, over 12974.00 frames. ], tot_loss[loss=0.1872, simple_loss=0.244, pruned_loss=0.06517, over 2656631.61 frames. ], batch size: 142, lr: 7.74e-03, grad_scale: 32.0 2023-04-16 23:05:08,036 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1825, 2.0662, 2.4190, 3.6319, 3.2901, 3.6771, 2.9220, 2.2092], device='cuda:1'), covar=tensor([0.0312, 0.1039, 0.0806, 0.0062, 0.0277, 0.0064, 0.0676, 0.1038], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0149, 0.0161, 0.0091, 0.0114, 0.0088, 0.0165, 0.0147], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-16 23:05:22,751 INFO [train.py:893] (1/4) Epoch 17, batch 1200, loss[loss=0.2039, simple_loss=0.2569, pruned_loss=0.07541, over 13434.00 frames. ], tot_loss[loss=0.188, simple_loss=0.245, pruned_loss=0.06545, over 2659364.07 frames. ], batch size: 95, lr: 7.73e-03, grad_scale: 32.0 2023-04-16 23:05:23,827 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9351, 4.2334, 3.9880, 3.9446, 4.0269, 4.3319, 4.1382, 4.0006], device='cuda:1'), covar=tensor([0.0269, 0.0245, 0.0291, 0.0884, 0.0255, 0.0220, 0.0267, 0.0371], device='cuda:1'), in_proj_covar=tensor([0.0149, 0.0145, 0.0162, 0.0258, 0.0165, 0.0180, 0.0160, 0.0107], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0006, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 23:05:39,314 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2789, 3.5339, 3.3831, 3.9816, 2.1460, 2.8823, 3.6024, 2.0848], device='cuda:1'), covar=tensor([0.0140, 0.0572, 0.0744, 0.0572, 0.1569, 0.0945, 0.0619, 0.1842], device='cuda:1'), in_proj_covar=tensor([0.0115, 0.0178, 0.0198, 0.0226, 0.0179, 0.0194, 0.0175, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 23:05:50,373 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.921e+02 2.713e+02 3.116e+02 3.508e+02 5.884e+02, threshold=6.232e+02, percent-clipped=0.0 2023-04-16 23:05:51,352 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-16 23:06:00,766 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1602, 4.6401, 4.3838, 4.3883, 4.3830, 4.2356, 4.6903, 4.7184], device='cuda:1'), covar=tensor([0.0277, 0.0250, 0.0315, 0.0335, 0.0296, 0.0403, 0.0288, 0.0274], device='cuda:1'), in_proj_covar=tensor([0.0127, 0.0187, 0.0148, 0.0171, 0.0135, 0.0186, 0.0124, 0.0165], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 23:06:03,782 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-16 23:06:06,502 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6409, 3.7410, 2.5174, 3.4597, 3.6343, 2.3469, 3.2126, 2.5103], device='cuda:1'), covar=tensor([0.0317, 0.0260, 0.1137, 0.0344, 0.0285, 0.1233, 0.0570, 0.1329], device='cuda:1'), in_proj_covar=tensor([0.0153, 0.0163, 0.0171, 0.0182, 0.0130, 0.0156, 0.0154, 0.0156], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 23:06:09,559 INFO [train.py:893] (1/4) Epoch 17, batch 1250, loss[loss=0.1698, simple_loss=0.2293, pruned_loss=0.05516, over 13356.00 frames. ], tot_loss[loss=0.1899, simple_loss=0.2465, pruned_loss=0.06662, over 2658848.55 frames. ], batch size: 73, lr: 7.73e-03, grad_scale: 32.0 2023-04-16 23:06:10,709 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9889, 1.7474, 3.2276, 3.2408, 3.0712, 2.6249, 3.0513, 2.3793], device='cuda:1'), covar=tensor([0.1581, 0.1473, 0.0166, 0.0202, 0.0270, 0.0703, 0.0255, 0.0804], device='cuda:1'), in_proj_covar=tensor([0.0212, 0.0180, 0.0115, 0.0121, 0.0125, 0.0167, 0.0137, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0001, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 23:06:16,211 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44443.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:06:43,531 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.2600, 2.0523, 3.9876, 3.8166, 3.9466, 3.0480, 3.7027, 3.0004], device='cuda:1'), covar=tensor([0.2040, 0.1715, 0.0160, 0.0266, 0.0159, 0.0773, 0.0304, 0.0897], device='cuda:1'), in_proj_covar=tensor([0.0214, 0.0181, 0.0116, 0.0122, 0.0126, 0.0170, 0.0139, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 23:06:54,451 INFO [train.py:893] (1/4) Epoch 17, batch 1300, loss[loss=0.2282, simple_loss=0.2775, pruned_loss=0.08947, over 13521.00 frames. ], tot_loss[loss=0.1913, simple_loss=0.2479, pruned_loss=0.06736, over 2659410.47 frames. ], batch size: 85, lr: 7.72e-03, grad_scale: 32.0 2023-04-16 23:07:00,362 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44491.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:07:21,526 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.930e+02 2.691e+02 3.241e+02 3.696e+02 7.455e+02, threshold=6.482e+02, percent-clipped=2.0 2023-04-16 23:07:29,131 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.52 vs. limit=5.0 2023-04-16 23:07:34,577 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0283, 4.3742, 4.0252, 4.1799, 4.1015, 4.4830, 4.2682, 4.0888], device='cuda:1'), covar=tensor([0.0298, 0.0232, 0.0336, 0.0753, 0.0319, 0.0229, 0.0279, 0.0384], device='cuda:1'), in_proj_covar=tensor([0.0151, 0.0147, 0.0165, 0.0260, 0.0167, 0.0183, 0.0161, 0.0109], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-16 23:07:34,641 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44527.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:07:41,941 INFO [train.py:893] (1/4) Epoch 17, batch 1350, loss[loss=0.1973, simple_loss=0.2523, pruned_loss=0.07122, over 13494.00 frames. ], tot_loss[loss=0.1923, simple_loss=0.2487, pruned_loss=0.06795, over 2655712.85 frames. ], batch size: 93, lr: 7.72e-03, grad_scale: 32.0 2023-04-16 23:07:55,214 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=44550.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:07:57,878 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1922, 4.3548, 3.4419, 2.9499, 3.2402, 2.6779, 4.5706, 2.6001], device='cuda:1'), covar=tensor([0.1601, 0.0301, 0.1035, 0.1996, 0.0739, 0.3050, 0.0181, 0.3730], device='cuda:1'), in_proj_covar=tensor([0.0305, 0.0281, 0.0301, 0.0314, 0.0248, 0.0315, 0.0202, 0.0354], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 23:08:09,365 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44565.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:08:28,911 INFO [train.py:893] (1/4) Epoch 17, batch 1400, loss[loss=0.1863, simple_loss=0.2403, pruned_loss=0.06621, over 13361.00 frames. ], tot_loss[loss=0.1913, simple_loss=0.2479, pruned_loss=0.06732, over 2657067.88 frames. ], batch size: 109, lr: 7.72e-03, grad_scale: 32.0 2023-04-16 23:08:30,964 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44588.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:08:51,529 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=44611.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:08:53,042 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44613.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:08:55,325 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.769e+02 2.732e+02 3.117e+02 3.575e+02 6.095e+02, threshold=6.234e+02, percent-clipped=0.0 2023-04-16 23:08:55,583 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=44615.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:09:14,430 INFO [train.py:893] (1/4) Epoch 17, batch 1450, loss[loss=0.2041, simple_loss=0.2558, pruned_loss=0.07619, over 13529.00 frames. ], tot_loss[loss=0.1915, simple_loss=0.2479, pruned_loss=0.0676, over 2657693.78 frames. ], batch size: 70, lr: 7.71e-03, grad_scale: 32.0 2023-04-16 23:09:39,612 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=44663.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:09:48,905 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-04-16 23:09:49,549 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2039, 3.5678, 3.3627, 3.8382, 2.1873, 2.9400, 3.6766, 2.1332], device='cuda:1'), covar=tensor([0.0136, 0.0459, 0.0764, 0.0528, 0.1509, 0.0904, 0.0499, 0.1713], device='cuda:1'), in_proj_covar=tensor([0.0114, 0.0177, 0.0198, 0.0227, 0.0179, 0.0194, 0.0174, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 23:10:01,295 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8767, 2.0086, 1.8160, 3.7341, 4.1316, 3.1583, 4.1658, 3.9072], device='cuda:1'), covar=tensor([0.0176, 0.1357, 0.1420, 0.0160, 0.0256, 0.0661, 0.0130, 0.0162], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0086, 0.0093, 0.0076, 0.0062, 0.0077, 0.0051, 0.0072], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 23:10:01,806 INFO [train.py:893] (1/4) Epoch 17, batch 1500, loss[loss=0.2069, simple_loss=0.2589, pruned_loss=0.07739, over 13327.00 frames. ], tot_loss[loss=0.1906, simple_loss=0.2471, pruned_loss=0.06705, over 2653562.12 frames. ], batch size: 118, lr: 7.71e-03, grad_scale: 32.0 2023-04-16 23:10:29,105 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.915e+02 2.738e+02 3.164e+02 3.676e+02 8.107e+02, threshold=6.328e+02, percent-clipped=2.0 2023-04-16 23:10:48,296 INFO [train.py:893] (1/4) Epoch 17, batch 1550, loss[loss=0.1809, simple_loss=0.242, pruned_loss=0.05989, over 13107.00 frames. ], tot_loss[loss=0.1902, simple_loss=0.2468, pruned_loss=0.06678, over 2653015.67 frames. ], batch size: 142, lr: 7.70e-03, grad_scale: 32.0 2023-04-16 23:11:33,087 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.91 vs. limit=5.0 2023-04-16 23:11:34,992 INFO [train.py:893] (1/4) Epoch 17, batch 1600, loss[loss=0.166, simple_loss=0.2319, pruned_loss=0.05011, over 13464.00 frames. ], tot_loss[loss=0.1894, simple_loss=0.2465, pruned_loss=0.06616, over 2655834.24 frames. ], batch size: 79, lr: 7.70e-03, grad_scale: 32.0 2023-04-16 23:12:02,570 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.839e+02 2.671e+02 3.065e+02 4.037e+02 6.217e+02, threshold=6.130e+02, percent-clipped=0.0 2023-04-16 23:12:23,119 INFO [train.py:893] (1/4) Epoch 17, batch 1650, loss[loss=0.1907, simple_loss=0.2513, pruned_loss=0.06511, over 11931.00 frames. ], tot_loss[loss=0.1898, simple_loss=0.2476, pruned_loss=0.06598, over 2655594.84 frames. ], batch size: 157, lr: 7.69e-03, grad_scale: 32.0 2023-04-16 23:12:55,650 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-04-16 23:13:06,090 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44883.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:13:08,325 INFO [train.py:893] (1/4) Epoch 17, batch 1700, loss[loss=0.198, simple_loss=0.2565, pruned_loss=0.06975, over 13450.00 frames. ], tot_loss[loss=0.1898, simple_loss=0.2479, pruned_loss=0.06591, over 2661446.87 frames. ], batch size: 103, lr: 7.69e-03, grad_scale: 32.0 2023-04-16 23:13:27,729 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=44906.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:13:35,564 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.839e+02 2.756e+02 3.136e+02 3.621e+02 5.395e+02, threshold=6.271e+02, percent-clipped=0.0 2023-04-16 23:13:55,252 INFO [train.py:893] (1/4) Epoch 17, batch 1750, loss[loss=0.1938, simple_loss=0.2547, pruned_loss=0.06641, over 13346.00 frames. ], tot_loss[loss=0.1875, simple_loss=0.2455, pruned_loss=0.06471, over 2664132.68 frames. ], batch size: 118, lr: 7.69e-03, grad_scale: 32.0 2023-04-16 23:14:02,148 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8810, 3.5991, 3.7767, 2.3961, 4.0679, 3.9174, 3.8624, 3.9947], device='cuda:1'), covar=tensor([0.0209, 0.0177, 0.0150, 0.1064, 0.0136, 0.0202, 0.0126, 0.0109], device='cuda:1'), in_proj_covar=tensor([0.0085, 0.0050, 0.0077, 0.0100, 0.0094, 0.0098, 0.0076, 0.0070], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 23:14:14,898 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.5758, 2.2239, 2.3165, 2.5877, 1.9180, 2.6902, 2.6640, 2.2185], device='cuda:1'), covar=tensor([0.0098, 0.0236, 0.0163, 0.0146, 0.0234, 0.0128, 0.0174, 0.0172], device='cuda:1'), in_proj_covar=tensor([0.0095, 0.0102, 0.0110, 0.0106, 0.0118, 0.0095, 0.0098, 0.0096], device='cuda:1'), out_proj_covar=tensor([9.8856e-05, 1.0990e-04, 1.2145e-04, 1.1495e-04, 1.3030e-04, 1.0153e-04, 1.0643e-04, 1.0167e-04], device='cuda:1') 2023-04-16 23:14:40,132 INFO [train.py:893] (1/4) Epoch 17, batch 1800, loss[loss=0.2121, simple_loss=0.2648, pruned_loss=0.07968, over 13518.00 frames. ], tot_loss[loss=0.1867, simple_loss=0.2447, pruned_loss=0.06429, over 2662736.21 frames. ], batch size: 98, lr: 7.68e-03, grad_scale: 16.0 2023-04-16 23:14:51,366 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6092, 3.1861, 2.7407, 4.6203, 5.1248, 3.9638, 5.0213, 4.7616], device='cuda:1'), covar=tensor([0.0114, 0.0635, 0.0808, 0.0082, 0.0063, 0.0365, 0.0058, 0.0077], device='cuda:1'), in_proj_covar=tensor([0.0070, 0.0086, 0.0094, 0.0076, 0.0063, 0.0078, 0.0052, 0.0073], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 23:15:08,027 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.790e+02 2.583e+02 3.065e+02 3.779e+02 6.192e+02, threshold=6.130e+02, percent-clipped=0.0 2023-04-16 23:15:28,118 INFO [train.py:893] (1/4) Epoch 17, batch 1850, loss[loss=0.171, simple_loss=0.2331, pruned_loss=0.0544, over 13530.00 frames. ], tot_loss[loss=0.1856, simple_loss=0.2438, pruned_loss=0.06367, over 2664108.61 frames. ], batch size: 72, lr: 7.68e-03, grad_scale: 16.0 2023-04-16 23:15:31,427 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-16 23:15:47,430 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45056.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:15:51,610 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0907, 3.4280, 3.3600, 3.7358, 2.1892, 2.8958, 3.5462, 2.0098], device='cuda:1'), covar=tensor([0.0109, 0.0527, 0.0657, 0.0438, 0.1425, 0.0883, 0.0582, 0.1909], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0180, 0.0200, 0.0231, 0.0180, 0.0195, 0.0176, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 23:16:09,809 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45081.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:16:14,530 INFO [train.py:893] (1/4) Epoch 17, batch 1900, loss[loss=0.1785, simple_loss=0.2394, pruned_loss=0.05878, over 13246.00 frames. ], tot_loss[loss=0.1857, simple_loss=0.2436, pruned_loss=0.0639, over 2665519.44 frames. ], batch size: 124, lr: 7.67e-03, grad_scale: 16.0 2023-04-16 23:16:31,332 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9628, 4.3911, 4.2504, 4.1464, 4.2187, 4.0467, 4.4290, 4.4654], device='cuda:1'), covar=tensor([0.0217, 0.0207, 0.0163, 0.0316, 0.0224, 0.0243, 0.0253, 0.0255], device='cuda:1'), in_proj_covar=tensor([0.0127, 0.0189, 0.0148, 0.0171, 0.0134, 0.0186, 0.0124, 0.0166], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 23:16:42,600 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.971e+02 2.721e+02 3.190e+02 3.626e+02 8.319e+02, threshold=6.379e+02, percent-clipped=2.0 2023-04-16 23:16:43,803 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45117.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 23:16:59,503 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 2023-04-16 23:16:59,858 INFO [train.py:893] (1/4) Epoch 17, batch 1950, loss[loss=0.203, simple_loss=0.2629, pruned_loss=0.07158, over 13428.00 frames. ], tot_loss[loss=0.1854, simple_loss=0.2433, pruned_loss=0.06374, over 2661471.98 frames. ], batch size: 113, lr: 7.67e-03, grad_scale: 16.0 2023-04-16 23:17:05,102 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45142.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:17:17,921 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-04-16 23:17:17,932 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-04-16 23:17:44,077 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45183.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:17:46,411 INFO [train.py:893] (1/4) Epoch 17, batch 2000, loss[loss=0.2214, simple_loss=0.2729, pruned_loss=0.08491, over 13434.00 frames. ], tot_loss[loss=0.1866, simple_loss=0.2445, pruned_loss=0.06435, over 2666974.33 frames. ], batch size: 106, lr: 7.67e-03, grad_scale: 16.0 2023-04-16 23:17:53,072 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-16 23:18:06,200 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45206.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:18:14,875 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.975e+02 2.748e+02 3.264e+02 3.939e+02 8.760e+02, threshold=6.528e+02, percent-clipped=3.0 2023-04-16 23:18:28,246 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45231.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:18:32,266 INFO [train.py:893] (1/4) Epoch 17, batch 2050, loss[loss=0.1645, simple_loss=0.2151, pruned_loss=0.05693, over 13382.00 frames. ], tot_loss[loss=0.1891, simple_loss=0.2467, pruned_loss=0.06576, over 2668479.63 frames. ], batch size: 62, lr: 7.66e-03, grad_scale: 16.0 2023-04-16 23:18:50,612 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45254.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:19:20,260 INFO [train.py:893] (1/4) Epoch 17, batch 2100, loss[loss=0.1607, simple_loss=0.2098, pruned_loss=0.05576, over 12783.00 frames. ], tot_loss[loss=0.1887, simple_loss=0.2462, pruned_loss=0.06558, over 2667893.24 frames. ], batch size: 52, lr: 7.66e-03, grad_scale: 16.0 2023-04-16 23:19:48,309 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.819e+02 2.694e+02 3.157e+02 3.891e+02 9.018e+02, threshold=6.314e+02, percent-clipped=1.0 2023-04-16 23:20:06,991 INFO [train.py:893] (1/4) Epoch 17, batch 2150, loss[loss=0.1918, simple_loss=0.2549, pruned_loss=0.06431, over 13542.00 frames. ], tot_loss[loss=0.1884, simple_loss=0.2463, pruned_loss=0.06527, over 2666388.66 frames. ], batch size: 91, lr: 7.65e-03, grad_scale: 16.0 2023-04-16 23:20:18,101 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45348.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 23:20:18,929 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9025, 3.8786, 3.1263, 3.6389, 3.1255, 2.2895, 3.8534, 2.1156], device='cuda:1'), covar=tensor([0.0775, 0.0413, 0.0490, 0.0279, 0.0757, 0.1932, 0.0888, 0.1418], device='cuda:1'), in_proj_covar=tensor([0.0137, 0.0135, 0.0131, 0.0112, 0.0148, 0.0184, 0.0164, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 23:20:36,712 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8873, 4.0093, 3.0823, 2.7089, 2.7620, 2.3943, 4.0624, 2.2313], device='cuda:1'), covar=tensor([0.1541, 0.0327, 0.1086, 0.1999, 0.0875, 0.3156, 0.0232, 0.4141], device='cuda:1'), in_proj_covar=tensor([0.0306, 0.0279, 0.0302, 0.0316, 0.0249, 0.0316, 0.0202, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 23:20:52,825 INFO [train.py:893] (1/4) Epoch 17, batch 2200, loss[loss=0.1651, simple_loss=0.2179, pruned_loss=0.05612, over 13433.00 frames. ], tot_loss[loss=0.1874, simple_loss=0.2454, pruned_loss=0.06476, over 2666019.60 frames. ], batch size: 65, lr: 7.65e-03, grad_scale: 16.0 2023-04-16 23:21:15,875 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45409.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 23:21:16,856 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-04-16 23:21:18,255 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45412.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 23:21:21,301 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.722e+02 2.681e+02 3.203e+02 3.685e+02 6.103e+02, threshold=6.406e+02, percent-clipped=0.0 2023-04-16 23:21:40,211 INFO [train.py:893] (1/4) Epoch 17, batch 2250, loss[loss=0.1653, simple_loss=0.2265, pruned_loss=0.05209, over 13067.00 frames. ], tot_loss[loss=0.1863, simple_loss=0.2439, pruned_loss=0.06436, over 2664391.10 frames. ], batch size: 142, lr: 7.64e-03, grad_scale: 16.0 2023-04-16 23:21:41,257 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45437.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:21:59,849 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-04-16 23:22:11,961 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7468, 2.5431, 4.3835, 4.1086, 4.3181, 3.5118, 4.1100, 3.2185], device='cuda:1'), covar=tensor([0.1765, 0.1504, 0.0079, 0.0202, 0.0106, 0.0531, 0.0170, 0.0752], device='cuda:1'), in_proj_covar=tensor([0.0215, 0.0182, 0.0115, 0.0121, 0.0127, 0.0172, 0.0137, 0.0180], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 23:22:24,771 INFO [train.py:893] (1/4) Epoch 17, batch 2300, loss[loss=0.1775, simple_loss=0.2398, pruned_loss=0.05755, over 13525.00 frames. ], tot_loss[loss=0.1857, simple_loss=0.2435, pruned_loss=0.06398, over 2664646.54 frames. ], batch size: 85, lr: 7.64e-03, grad_scale: 16.0 2023-04-16 23:22:51,733 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.737e+02 2.559e+02 2.932e+02 3.587e+02 6.349e+02, threshold=5.863e+02, percent-clipped=0.0 2023-04-16 23:23:11,512 INFO [train.py:893] (1/4) Epoch 17, batch 2350, loss[loss=0.1934, simple_loss=0.246, pruned_loss=0.0704, over 13070.00 frames. ], tot_loss[loss=0.1855, simple_loss=0.2433, pruned_loss=0.06387, over 2665816.48 frames. ], batch size: 142, lr: 7.64e-03, grad_scale: 16.0 2023-04-16 23:23:34,118 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-16 23:23:58,211 INFO [train.py:893] (1/4) Epoch 17, batch 2400, loss[loss=0.1918, simple_loss=0.2473, pruned_loss=0.06814, over 13522.00 frames. ], tot_loss[loss=0.1857, simple_loss=0.2431, pruned_loss=0.0641, over 2666420.19 frames. ], batch size: 83, lr: 7.63e-03, grad_scale: 16.0 2023-04-16 23:24:10,738 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0325, 2.7623, 2.2245, 1.5867, 1.8110, 2.4309, 2.4685, 3.0189], device='cuda:1'), covar=tensor([0.0802, 0.0284, 0.0761, 0.1624, 0.0317, 0.0441, 0.0670, 0.0220], device='cuda:1'), in_proj_covar=tensor([0.0153, 0.0127, 0.0110, 0.0199, 0.0103, 0.0149, 0.0163, 0.0122], device='cuda:1'), out_proj_covar=tensor([1.1400e-04, 9.5133e-05, 8.6664e-05, 1.4933e-04, 7.6934e-05, 1.1257e-04, 1.2340e-04, 9.0745e-05], device='cuda:1') 2023-04-16 23:24:26,227 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.7840, 5.0011, 4.8160, 4.8500, 4.8441, 5.1548, 4.9846, 4.7764], device='cuda:1'), covar=tensor([0.0209, 0.0243, 0.0236, 0.0649, 0.0248, 0.0174, 0.0214, 0.0157], device='cuda:1'), in_proj_covar=tensor([0.0150, 0.0145, 0.0164, 0.0257, 0.0167, 0.0182, 0.0160, 0.0107], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-16 23:24:26,797 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.045e+02 2.728e+02 3.241e+02 3.826e+02 5.379e+02, threshold=6.482e+02, percent-clipped=0.0 2023-04-16 23:24:28,163 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 2023-04-16 23:24:45,229 INFO [train.py:893] (1/4) Epoch 17, batch 2450, loss[loss=0.1886, simple_loss=0.2468, pruned_loss=0.06517, over 13269.00 frames. ], tot_loss[loss=0.1859, simple_loss=0.2435, pruned_loss=0.06409, over 2666749.43 frames. ], batch size: 124, lr: 7.63e-03, grad_scale: 16.0 2023-04-16 23:25:12,983 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9553, 3.7996, 3.1954, 3.6070, 3.1626, 2.0709, 3.8840, 2.1864], device='cuda:1'), covar=tensor([0.0733, 0.0504, 0.0400, 0.0338, 0.0652, 0.2077, 0.0818, 0.1354], device='cuda:1'), in_proj_covar=tensor([0.0137, 0.0135, 0.0130, 0.0112, 0.0147, 0.0184, 0.0162, 0.0138], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 23:25:32,480 INFO [train.py:893] (1/4) Epoch 17, batch 2500, loss[loss=0.1755, simple_loss=0.2395, pruned_loss=0.0558, over 13520.00 frames. ], tot_loss[loss=0.1852, simple_loss=0.243, pruned_loss=0.06371, over 2668721.07 frames. ], batch size: 91, lr: 7.62e-03, grad_scale: 16.0 2023-04-16 23:25:46,832 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45702.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:25:48,436 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=45704.0, num_to_drop=1, layers_to_drop={2} 2023-04-16 23:25:56,869 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45712.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 23:26:00,709 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.951e+02 2.760e+02 3.151e+02 3.629e+02 9.778e+02, threshold=6.301e+02, percent-clipped=0.0 2023-04-16 23:26:17,927 INFO [train.py:893] (1/4) Epoch 17, batch 2550, loss[loss=0.1947, simple_loss=0.2548, pruned_loss=0.06732, over 13522.00 frames. ], tot_loss[loss=0.1851, simple_loss=0.2429, pruned_loss=0.06365, over 2668288.48 frames. ], batch size: 91, lr: 7.62e-03, grad_scale: 16.0 2023-04-16 23:26:18,978 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=45737.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:26:40,097 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-16 23:26:41,033 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45760.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:26:43,657 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45763.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:26:54,100 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45774.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:27:04,764 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=45785.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:27:05,400 INFO [train.py:893] (1/4) Epoch 17, batch 2600, loss[loss=0.2103, simple_loss=0.2609, pruned_loss=0.07986, over 13213.00 frames. ], tot_loss[loss=0.1858, simple_loss=0.2433, pruned_loss=0.06416, over 2671977.62 frames. ], batch size: 132, lr: 7.62e-03, grad_scale: 16.0 2023-04-16 23:27:10,841 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0045, 4.0555, 2.8351, 3.7357, 3.9758, 2.5319, 3.5714, 2.8671], device='cuda:1'), covar=tensor([0.0279, 0.0222, 0.1085, 0.0324, 0.0224, 0.1204, 0.0501, 0.1178], device='cuda:1'), in_proj_covar=tensor([0.0156, 0.0165, 0.0173, 0.0188, 0.0132, 0.0158, 0.0156, 0.0159], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 23:27:30,329 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6043, 4.4925, 4.7086, 4.6518, 4.9786, 4.4748, 4.9964, 4.9513], device='cuda:1'), covar=tensor([0.0416, 0.0569, 0.0677, 0.0538, 0.0561, 0.0846, 0.0458, 0.0512], device='cuda:1'), in_proj_covar=tensor([0.0268, 0.0281, 0.0287, 0.0215, 0.0405, 0.0323, 0.0258, 0.0281], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 23:27:31,661 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.842e+02 2.635e+02 3.152e+02 4.000e+02 8.090e+02, threshold=6.305e+02, percent-clipped=5.0 2023-04-16 23:27:47,023 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45835.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:27:47,566 INFO [train.py:893] (1/4) Epoch 17, batch 2650, loss[loss=0.2116, simple_loss=0.2674, pruned_loss=0.07791, over 13074.00 frames. ], tot_loss[loss=0.1874, simple_loss=0.2445, pruned_loss=0.06509, over 2673113.77 frames. ], batch size: 142, lr: 7.61e-03, grad_scale: 16.0 2023-04-16 23:28:00,043 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-04-16 23:28:05,508 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=45859.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:28:45,052 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-16 23:28:55,841 INFO [train.py:893] (1/4) Epoch 18, batch 0, loss[loss=0.1924, simple_loss=0.2482, pruned_loss=0.06832, over 13534.00 frames. ], tot_loss[loss=0.1924, simple_loss=0.2482, pruned_loss=0.06832, over 13534.00 frames. ], batch size: 91, lr: 7.39e-03, grad_scale: 16.0 2023-04-16 23:28:55,842 INFO [train.py:918] (1/4) Computing validation loss 2023-04-16 23:29:18,849 INFO [train.py:927] (1/4) Epoch 18, validation: loss=0.1416, simple_loss=0.2005, pruned_loss=0.04137, over 2446609.00 frames. 2023-04-16 23:29:18,850 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12829MB 2023-04-16 23:29:35,651 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6426, 2.7070, 2.9838, 4.1507, 3.7608, 4.2116, 3.4706, 2.5419], device='cuda:1'), covar=tensor([0.0249, 0.0892, 0.0761, 0.0050, 0.0199, 0.0043, 0.0579, 0.0944], device='cuda:1'), in_proj_covar=tensor([0.0115, 0.0147, 0.0162, 0.0090, 0.0113, 0.0088, 0.0164, 0.0147], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-16 23:29:46,276 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.926e+02 2.882e+02 3.236e+02 3.835e+02 6.257e+02, threshold=6.473e+02, percent-clipped=0.0 2023-04-16 23:29:51,624 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=45920.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:30:02,455 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9282, 4.0884, 3.1297, 2.7686, 2.8922, 2.4949, 4.2607, 2.3804], device='cuda:1'), covar=tensor([0.1599, 0.0395, 0.1121, 0.2049, 0.0818, 0.3214, 0.0236, 0.3820], device='cuda:1'), in_proj_covar=tensor([0.0306, 0.0280, 0.0302, 0.0317, 0.0248, 0.0318, 0.0202, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 23:30:05,443 INFO [train.py:893] (1/4) Epoch 18, batch 50, loss[loss=0.1829, simple_loss=0.2451, pruned_loss=0.06033, over 13480.00 frames. ], tot_loss[loss=0.1811, simple_loss=0.2368, pruned_loss=0.06268, over 600579.41 frames. ], batch size: 81, lr: 7.39e-03, grad_scale: 16.0 2023-04-16 23:30:17,420 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7114, 2.3875, 2.3585, 2.7822, 2.1066, 2.8414, 2.7412, 2.3570], device='cuda:1'), covar=tensor([0.0084, 0.0176, 0.0148, 0.0145, 0.0203, 0.0102, 0.0184, 0.0160], device='cuda:1'), in_proj_covar=tensor([0.0096, 0.0104, 0.0111, 0.0107, 0.0121, 0.0097, 0.0099, 0.0096], device='cuda:1'), out_proj_covar=tensor([9.9644e-05, 1.1240e-04, 1.2116e-04, 1.1530e-04, 1.3272e-04, 1.0350e-04, 1.0759e-04, 1.0184e-04], device='cuda:1') 2023-04-16 23:30:30,553 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-16 23:30:30,553 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-16 23:30:30,554 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-16 23:30:30,560 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-16 23:30:31,299 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-16 23:30:31,322 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-16 23:30:31,332 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-16 23:30:51,214 INFO [train.py:893] (1/4) Epoch 18, batch 100, loss[loss=0.2094, simple_loss=0.2615, pruned_loss=0.07868, over 13254.00 frames. ], tot_loss[loss=0.1861, simple_loss=0.2413, pruned_loss=0.0655, over 1058447.51 frames. ], batch size: 132, lr: 7.38e-03, grad_scale: 16.0 2023-04-16 23:31:12,794 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46004.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 23:31:23,504 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.126e+02 2.834e+02 3.248e+02 3.832e+02 7.333e+02, threshold=6.497e+02, percent-clipped=1.0 2023-04-16 23:31:40,945 INFO [train.py:893] (1/4) Epoch 18, batch 150, loss[loss=0.1713, simple_loss=0.2297, pruned_loss=0.0565, over 13115.00 frames. ], tot_loss[loss=0.1906, simple_loss=0.245, pruned_loss=0.06814, over 1398467.20 frames. ], batch size: 142, lr: 7.38e-03, grad_scale: 16.0 2023-04-16 23:31:58,072 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46052.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 23:32:03,721 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46058.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:32:27,723 INFO [train.py:893] (1/4) Epoch 18, batch 200, loss[loss=0.1787, simple_loss=0.2426, pruned_loss=0.0574, over 13379.00 frames. ], tot_loss[loss=0.1938, simple_loss=0.2474, pruned_loss=0.07009, over 1667065.71 frames. ], batch size: 113, lr: 7.37e-03, grad_scale: 16.0 2023-04-16 23:32:56,064 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.742e+02 2.872e+02 3.344e+02 4.213e+02 8.067e+02, threshold=6.687e+02, percent-clipped=4.0 2023-04-16 23:33:08,554 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46130.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:33:12,532 INFO [train.py:893] (1/4) Epoch 18, batch 250, loss[loss=0.1791, simple_loss=0.2346, pruned_loss=0.06182, over 13352.00 frames. ], tot_loss[loss=0.1932, simple_loss=0.2475, pruned_loss=0.06946, over 1884845.82 frames. ], batch size: 67, lr: 7.37e-03, grad_scale: 16.0 2023-04-16 23:33:36,709 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2735, 3.6076, 3.3972, 3.9459, 2.0895, 2.9364, 3.7645, 2.0261], device='cuda:1'), covar=tensor([0.0145, 0.0571, 0.0825, 0.0516, 0.1770, 0.1103, 0.0518, 0.2067], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0182, 0.0202, 0.0235, 0.0183, 0.0199, 0.0177, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 23:33:40,071 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.5538, 2.2928, 2.2786, 2.6356, 1.9263, 2.6707, 2.5229, 2.1902], device='cuda:1'), covar=tensor([0.0077, 0.0196, 0.0154, 0.0130, 0.0221, 0.0129, 0.0203, 0.0159], device='cuda:1'), in_proj_covar=tensor([0.0095, 0.0103, 0.0110, 0.0106, 0.0120, 0.0097, 0.0098, 0.0096], device='cuda:1'), out_proj_covar=tensor([9.8442e-05, 1.1121e-04, 1.2066e-04, 1.1386e-04, 1.3178e-04, 1.0319e-04, 1.0638e-04, 1.0243e-04], device='cuda:1') 2023-04-16 23:33:40,106 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1968, 3.5339, 3.3818, 3.7861, 2.1015, 2.9909, 3.6318, 2.0844], device='cuda:1'), covar=tensor([0.0150, 0.0492, 0.0686, 0.0444, 0.1579, 0.0844, 0.0552, 0.1636], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0181, 0.0202, 0.0234, 0.0183, 0.0198, 0.0177, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 23:34:00,295 INFO [train.py:893] (1/4) Epoch 18, batch 300, loss[loss=0.2104, simple_loss=0.271, pruned_loss=0.07492, over 13388.00 frames. ], tot_loss[loss=0.1935, simple_loss=0.2485, pruned_loss=0.06928, over 2053616.58 frames. ], batch size: 109, lr: 7.37e-03, grad_scale: 16.0 2023-04-16 23:34:26,038 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.3697, 4.8759, 4.7888, 4.8061, 4.6240, 4.6601, 5.3344, 4.8785], device='cuda:1'), covar=tensor([0.0719, 0.1340, 0.2165, 0.2601, 0.0880, 0.1608, 0.0889, 0.1286], device='cuda:1'), in_proj_covar=tensor([0.0304, 0.0355, 0.0445, 0.0451, 0.0268, 0.0329, 0.0407, 0.0324], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-16 23:34:29,236 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46215.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:34:29,871 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.905e+02 2.634e+02 3.212e+02 4.032e+02 9.593e+02, threshold=6.425e+02, percent-clipped=2.0 2023-04-16 23:34:46,265 INFO [train.py:893] (1/4) Epoch 18, batch 350, loss[loss=0.1802, simple_loss=0.2411, pruned_loss=0.05962, over 13535.00 frames. ], tot_loss[loss=0.1933, simple_loss=0.2486, pruned_loss=0.06902, over 2187910.93 frames. ], batch size: 85, lr: 7.36e-03, grad_scale: 16.0 2023-04-16 23:35:05,133 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6402, 2.4107, 2.4496, 2.7075, 2.0945, 2.8022, 2.7895, 2.3214], device='cuda:1'), covar=tensor([0.0084, 0.0175, 0.0185, 0.0153, 0.0211, 0.0126, 0.0142, 0.0168], device='cuda:1'), in_proj_covar=tensor([0.0095, 0.0104, 0.0110, 0.0106, 0.0120, 0.0097, 0.0098, 0.0097], device='cuda:1'), out_proj_covar=tensor([9.8889e-05, 1.1202e-04, 1.2123e-04, 1.1412e-04, 1.3238e-04, 1.0400e-04, 1.0634e-04, 1.0339e-04], device='cuda:1') 2023-04-16 23:35:26,197 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46276.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:35:33,391 INFO [train.py:893] (1/4) Epoch 18, batch 400, loss[loss=0.1862, simple_loss=0.2413, pruned_loss=0.06557, over 13526.00 frames. ], tot_loss[loss=0.1924, simple_loss=0.2483, pruned_loss=0.06828, over 2296395.01 frames. ], batch size: 85, lr: 7.36e-03, grad_scale: 16.0 2023-04-16 23:36:02,177 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.971e+02 2.592e+02 3.217e+02 3.805e+02 6.157e+02, threshold=6.435e+02, percent-clipped=0.0 2023-04-16 23:36:07,460 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0440, 4.3271, 4.0293, 4.1374, 4.1340, 4.4943, 4.2751, 4.0754], device='cuda:1'), covar=tensor([0.0275, 0.0286, 0.0329, 0.0819, 0.0280, 0.0205, 0.0273, 0.0377], device='cuda:1'), in_proj_covar=tensor([0.0149, 0.0144, 0.0163, 0.0254, 0.0166, 0.0182, 0.0159, 0.0106], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-16 23:36:18,728 INFO [train.py:893] (1/4) Epoch 18, batch 450, loss[loss=0.1813, simple_loss=0.2423, pruned_loss=0.06014, over 13349.00 frames. ], tot_loss[loss=0.1939, simple_loss=0.25, pruned_loss=0.06896, over 2376616.29 frames. ], batch size: 118, lr: 7.35e-03, grad_scale: 16.0 2023-04-16 23:36:21,606 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46337.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:36:39,134 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46358.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:36:43,037 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-16 23:37:01,914 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-04-16 23:37:04,568 INFO [train.py:893] (1/4) Epoch 18, batch 500, loss[loss=0.1852, simple_loss=0.2417, pruned_loss=0.06436, over 13515.00 frames. ], tot_loss[loss=0.193, simple_loss=0.2495, pruned_loss=0.06827, over 2439391.42 frames. ], batch size: 70, lr: 7.35e-03, grad_scale: 16.0 2023-04-16 23:37:24,792 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46406.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:37:33,008 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.893e+02 2.887e+02 3.261e+02 4.060e+02 6.137e+02, threshold=6.521e+02, percent-clipped=0.0 2023-04-16 23:37:47,276 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46430.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:37:52,086 INFO [train.py:893] (1/4) Epoch 18, batch 550, loss[loss=0.1914, simple_loss=0.2463, pruned_loss=0.06825, over 13248.00 frames. ], tot_loss[loss=0.1918, simple_loss=0.2483, pruned_loss=0.06767, over 2485420.85 frames. ], batch size: 132, lr: 7.35e-03, grad_scale: 16.0 2023-04-16 23:38:31,275 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46478.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:38:36,987 INFO [train.py:893] (1/4) Epoch 18, batch 600, loss[loss=0.1725, simple_loss=0.2278, pruned_loss=0.05863, over 13531.00 frames. ], tot_loss[loss=0.1904, simple_loss=0.2465, pruned_loss=0.06714, over 2524504.28 frames. ], batch size: 70, lr: 7.34e-03, grad_scale: 16.0 2023-04-16 23:39:05,436 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46515.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:39:05,945 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.019e+02 2.668e+02 3.118e+02 3.710e+02 6.528e+02, threshold=6.235e+02, percent-clipped=1.0 2023-04-16 23:39:24,371 INFO [train.py:893] (1/4) Epoch 18, batch 650, loss[loss=0.1867, simple_loss=0.2475, pruned_loss=0.06293, over 13209.00 frames. ], tot_loss[loss=0.1895, simple_loss=0.2457, pruned_loss=0.06666, over 2555621.05 frames. ], batch size: 132, lr: 7.34e-03, grad_scale: 16.0 2023-04-16 23:39:41,297 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4047, 4.1986, 4.3908, 4.3773, 4.6469, 4.1832, 4.6671, 4.6024], device='cuda:1'), covar=tensor([0.0400, 0.0527, 0.0679, 0.0509, 0.0532, 0.0866, 0.0433, 0.0481], device='cuda:1'), in_proj_covar=tensor([0.0263, 0.0273, 0.0280, 0.0211, 0.0395, 0.0317, 0.0255, 0.0275], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 23:39:43,166 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.2129, 4.4850, 3.5117, 2.9891, 3.2529, 2.6533, 4.6286, 2.5724], device='cuda:1'), covar=tensor([0.1517, 0.0282, 0.1018, 0.1988, 0.0751, 0.3089, 0.0189, 0.3657], device='cuda:1'), in_proj_covar=tensor([0.0308, 0.0281, 0.0304, 0.0319, 0.0251, 0.0318, 0.0205, 0.0358], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 23:39:50,480 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46563.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:40:00,834 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-04-16 23:40:05,812 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0146, 4.4817, 4.2435, 4.2636, 4.2542, 4.0829, 4.5579, 4.5574], device='cuda:1'), covar=tensor([0.0230, 0.0221, 0.0194, 0.0324, 0.0256, 0.0309, 0.0248, 0.0237], device='cuda:1'), in_proj_covar=tensor([0.0130, 0.0190, 0.0150, 0.0172, 0.0137, 0.0187, 0.0126, 0.0166], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 23:40:09,748 INFO [train.py:893] (1/4) Epoch 18, batch 700, loss[loss=0.1817, simple_loss=0.2425, pruned_loss=0.06043, over 13421.00 frames. ], tot_loss[loss=0.1882, simple_loss=0.2448, pruned_loss=0.06578, over 2582746.98 frames. ], batch size: 95, lr: 7.33e-03, grad_scale: 16.0 2023-04-16 23:40:19,947 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46595.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:40:26,076 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46602.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:40:38,199 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.838e+02 2.538e+02 2.982e+02 3.442e+02 5.381e+02, threshold=5.963e+02, percent-clipped=0.0 2023-04-16 23:40:53,058 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46632.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:40:53,443 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-04-16 23:40:55,404 INFO [train.py:893] (1/4) Epoch 18, batch 750, loss[loss=0.1743, simple_loss=0.2167, pruned_loss=0.06599, over 12211.00 frames. ], tot_loss[loss=0.1881, simple_loss=0.2443, pruned_loss=0.06595, over 2592865.46 frames. ], batch size: 49, lr: 7.33e-03, grad_scale: 16.0 2023-04-16 23:41:16,458 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8701, 3.7741, 3.0306, 3.3688, 3.0668, 2.1527, 3.7315, 2.1566], device='cuda:1'), covar=tensor([0.0704, 0.0398, 0.0465, 0.0355, 0.0683, 0.1878, 0.0830, 0.1327], device='cuda:1'), in_proj_covar=tensor([0.0140, 0.0136, 0.0133, 0.0113, 0.0149, 0.0185, 0.0165, 0.0141], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-16 23:41:16,481 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46656.0, num_to_drop=1, layers_to_drop={3} 2023-04-16 23:41:22,217 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46663.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:41:43,678 INFO [train.py:893] (1/4) Epoch 18, batch 800, loss[loss=0.1989, simple_loss=0.2489, pruned_loss=0.07449, over 13443.00 frames. ], tot_loss[loss=0.1883, simple_loss=0.2448, pruned_loss=0.06587, over 2609376.09 frames. ], batch size: 65, lr: 7.33e-03, grad_scale: 16.0 2023-04-16 23:41:53,949 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3250, 2.1455, 2.6743, 3.7743, 3.4353, 3.8513, 2.9383, 2.1558], device='cuda:1'), covar=tensor([0.0295, 0.0979, 0.0742, 0.0062, 0.0252, 0.0054, 0.0729, 0.1053], device='cuda:1'), in_proj_covar=tensor([0.0116, 0.0146, 0.0160, 0.0090, 0.0113, 0.0089, 0.0163, 0.0146], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-16 23:42:11,128 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.676e+02 3.155e+02 3.704e+02 5.990e+02, threshold=6.311e+02, percent-clipped=1.0 2023-04-16 23:42:27,928 INFO [train.py:893] (1/4) Epoch 18, batch 850, loss[loss=0.1745, simple_loss=0.2309, pruned_loss=0.05903, over 13532.00 frames. ], tot_loss[loss=0.1892, simple_loss=0.2461, pruned_loss=0.06614, over 2623567.71 frames. ], batch size: 76, lr: 7.32e-03, grad_scale: 16.0 2023-04-16 23:43:09,574 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.3685, 2.0810, 2.0929, 2.3552, 1.8405, 2.3692, 2.2423, 1.9486], device='cuda:1'), covar=tensor([0.0087, 0.0184, 0.0160, 0.0128, 0.0195, 0.0139, 0.0196, 0.0178], device='cuda:1'), in_proj_covar=tensor([0.0098, 0.0107, 0.0114, 0.0111, 0.0125, 0.0100, 0.0101, 0.0100], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-16 23:43:15,263 INFO [train.py:893] (1/4) Epoch 18, batch 900, loss[loss=0.1778, simple_loss=0.239, pruned_loss=0.05826, over 13560.00 frames. ], tot_loss[loss=0.1898, simple_loss=0.246, pruned_loss=0.06677, over 2630340.87 frames. ], batch size: 89, lr: 7.32e-03, grad_scale: 16.0 2023-04-16 23:43:39,056 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=46810.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:43:44,622 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.126e+02 2.808e+02 3.282e+02 4.134e+02 5.929e+02, threshold=6.564e+02, percent-clipped=0.0 2023-04-16 23:43:45,563 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-16 23:43:46,064 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-04-16 23:44:00,569 INFO [train.py:893] (1/4) Epoch 18, batch 950, loss[loss=0.2024, simple_loss=0.2509, pruned_loss=0.07694, over 13385.00 frames. ], tot_loss[loss=0.1897, simple_loss=0.2456, pruned_loss=0.06686, over 2639557.40 frames. ], batch size: 62, lr: 7.32e-03, grad_scale: 16.0 2023-04-16 23:44:20,952 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0649, 2.1599, 3.6252, 3.6059, 3.5433, 2.8260, 3.3562, 2.6686], device='cuda:1'), covar=tensor([0.1896, 0.1333, 0.0171, 0.0206, 0.0245, 0.0717, 0.0242, 0.0868], device='cuda:1'), in_proj_covar=tensor([0.0213, 0.0181, 0.0115, 0.0123, 0.0129, 0.0171, 0.0137, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-16 23:44:35,112 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=46871.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:44:47,222 INFO [train.py:893] (1/4) Epoch 18, batch 1000, loss[loss=0.1768, simple_loss=0.2278, pruned_loss=0.06291, over 13147.00 frames. ], tot_loss[loss=0.1884, simple_loss=0.2441, pruned_loss=0.06637, over 2641053.86 frames. ], batch size: 58, lr: 7.31e-03, grad_scale: 16.0 2023-04-16 23:45:15,979 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.800e+02 2.599e+02 2.993e+02 3.540e+02 6.302e+02, threshold=5.986e+02, percent-clipped=0.0 2023-04-16 23:45:30,998 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=46932.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:45:33,842 INFO [train.py:893] (1/4) Epoch 18, batch 1050, loss[loss=0.1891, simple_loss=0.2497, pruned_loss=0.06421, over 13451.00 frames. ], tot_loss[loss=0.1871, simple_loss=0.2434, pruned_loss=0.06536, over 2649769.33 frames. ], batch size: 106, lr: 7.31e-03, grad_scale: 16.0 2023-04-16 23:45:39,428 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-04-16 23:45:47,993 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46951.0, num_to_drop=1, layers_to_drop={0} 2023-04-16 23:45:53,831 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=46958.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:46:15,054 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=46980.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:46:19,126 INFO [train.py:893] (1/4) Epoch 18, batch 1100, loss[loss=0.1755, simple_loss=0.2449, pruned_loss=0.05308, over 13526.00 frames. ], tot_loss[loss=0.1875, simple_loss=0.2442, pruned_loss=0.06544, over 2654754.45 frames. ], batch size: 85, lr: 7.30e-03, grad_scale: 32.0 2023-04-16 23:46:39,525 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0917, 2.1993, 2.5332, 3.5361, 3.2034, 3.6271, 2.8973, 2.3488], device='cuda:1'), covar=tensor([0.0330, 0.0895, 0.0729, 0.0066, 0.0293, 0.0054, 0.0591, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0149, 0.0164, 0.0091, 0.0115, 0.0089, 0.0166, 0.0147], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-16 23:46:48,439 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.808e+02 2.562e+02 3.010e+02 3.540e+02 7.955e+02, threshold=6.021e+02, percent-clipped=2.0 2023-04-16 23:47:06,594 INFO [train.py:893] (1/4) Epoch 18, batch 1150, loss[loss=0.1809, simple_loss=0.2392, pruned_loss=0.06132, over 13465.00 frames. ], tot_loss[loss=0.1866, simple_loss=0.2437, pruned_loss=0.06475, over 2654871.07 frames. ], batch size: 106, lr: 7.30e-03, grad_scale: 32.0 2023-04-16 23:47:52,591 INFO [train.py:893] (1/4) Epoch 18, batch 1200, loss[loss=0.2002, simple_loss=0.2558, pruned_loss=0.07234, over 13360.00 frames. ], tot_loss[loss=0.1854, simple_loss=0.2432, pruned_loss=0.06382, over 2659357.03 frames. ], batch size: 109, lr: 7.30e-03, grad_scale: 32.0 2023-04-16 23:48:21,773 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.858e+02 2.660e+02 3.050e+02 3.598e+02 7.489e+02, threshold=6.100e+02, percent-clipped=1.0 2023-04-16 23:48:21,864 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-16 23:48:34,972 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-16 23:48:40,793 INFO [train.py:893] (1/4) Epoch 18, batch 1250, loss[loss=0.1835, simple_loss=0.2403, pruned_loss=0.06337, over 13495.00 frames. ], tot_loss[loss=0.1855, simple_loss=0.2431, pruned_loss=0.06396, over 2652768.18 frames. ], batch size: 93, lr: 7.29e-03, grad_scale: 32.0 2023-04-16 23:48:47,819 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1108, 4.2939, 3.3210, 2.8770, 3.0528, 2.5813, 4.4391, 2.5058], device='cuda:1'), covar=tensor([0.1555, 0.0279, 0.1145, 0.2144, 0.0826, 0.3058, 0.0223, 0.3786], device='cuda:1'), in_proj_covar=tensor([0.0309, 0.0284, 0.0307, 0.0321, 0.0252, 0.0320, 0.0206, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-16 23:49:09,910 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47166.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:49:27,696 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-04-16 23:49:28,061 INFO [train.py:893] (1/4) Epoch 18, batch 1300, loss[loss=0.1944, simple_loss=0.2515, pruned_loss=0.06867, over 11462.00 frames. ], tot_loss[loss=0.187, simple_loss=0.2447, pruned_loss=0.06465, over 2654297.24 frames. ], batch size: 157, lr: 7.29e-03, grad_scale: 32.0 2023-04-16 23:49:52,214 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.80 vs. limit=5.0 2023-04-16 23:49:56,998 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.759e+02 2.746e+02 3.213e+02 3.709e+02 7.669e+02, threshold=6.426e+02, percent-clipped=2.0 2023-04-16 23:50:13,560 INFO [train.py:893] (1/4) Epoch 18, batch 1350, loss[loss=0.1609, simple_loss=0.203, pruned_loss=0.05941, over 12137.00 frames. ], tot_loss[loss=0.1875, simple_loss=0.2451, pruned_loss=0.06493, over 2659320.27 frames. ], batch size: 49, lr: 7.28e-03, grad_scale: 32.0 2023-04-16 23:50:19,775 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47241.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:50:29,515 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47251.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:50:35,344 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47258.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:50:44,575 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9444, 3.9406, 2.8120, 3.6323, 3.8483, 2.6047, 3.4914, 2.6476], device='cuda:1'), covar=tensor([0.0236, 0.0241, 0.1063, 0.0376, 0.0280, 0.1222, 0.0549, 0.1369], device='cuda:1'), in_proj_covar=tensor([0.0157, 0.0169, 0.0176, 0.0195, 0.0133, 0.0159, 0.0160, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 23:50:57,009 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-04-16 23:51:00,617 INFO [train.py:893] (1/4) Epoch 18, batch 1400, loss[loss=0.1898, simple_loss=0.2504, pruned_loss=0.06455, over 13566.00 frames. ], tot_loss[loss=0.1873, simple_loss=0.2449, pruned_loss=0.06488, over 2661359.45 frames. ], batch size: 89, lr: 7.28e-03, grad_scale: 32.0 2023-04-16 23:51:13,496 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47299.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:51:16,099 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47302.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:51:20,293 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47306.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:51:29,922 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.066e+02 2.681e+02 3.277e+02 3.724e+02 6.638e+02, threshold=6.553e+02, percent-clipped=2.0 2023-04-16 23:51:46,914 INFO [train.py:893] (1/4) Epoch 18, batch 1450, loss[loss=0.1897, simple_loss=0.26, pruned_loss=0.05968, over 13451.00 frames. ], tot_loss[loss=0.1879, simple_loss=0.2453, pruned_loss=0.06524, over 2657793.56 frames. ], batch size: 100, lr: 7.28e-03, grad_scale: 32.0 2023-04-16 23:52:19,619 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-04-16 23:52:29,358 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47379.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:52:34,026 INFO [train.py:893] (1/4) Epoch 18, batch 1500, loss[loss=0.2011, simple_loss=0.255, pruned_loss=0.07366, over 13526.00 frames. ], tot_loss[loss=0.187, simple_loss=0.2447, pruned_loss=0.06465, over 2652037.55 frames. ], batch size: 87, lr: 7.27e-03, grad_scale: 32.0 2023-04-16 23:53:03,657 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.719e+02 2.645e+02 2.973e+02 3.718e+02 8.279e+02, threshold=5.945e+02, percent-clipped=1.0 2023-04-16 23:53:21,946 INFO [train.py:893] (1/4) Epoch 18, batch 1550, loss[loss=0.2036, simple_loss=0.2553, pruned_loss=0.07598, over 13471.00 frames. ], tot_loss[loss=0.1862, simple_loss=0.244, pruned_loss=0.06423, over 2651280.51 frames. ], batch size: 100, lr: 7.27e-03, grad_scale: 32.0 2023-04-16 23:53:23,322 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-04-16 23:53:27,163 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47440.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:53:51,424 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47466.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:54:08,482 INFO [train.py:893] (1/4) Epoch 18, batch 1600, loss[loss=0.2365, simple_loss=0.2868, pruned_loss=0.09311, over 13394.00 frames. ], tot_loss[loss=0.186, simple_loss=0.244, pruned_loss=0.06399, over 2656499.10 frames. ], batch size: 113, lr: 7.27e-03, grad_scale: 32.0 2023-04-16 23:54:35,023 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47514.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:54:36,507 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.837e+02 2.658e+02 3.121e+02 4.001e+02 6.335e+02, threshold=6.242e+02, percent-clipped=2.0 2023-04-16 23:54:54,969 INFO [train.py:893] (1/4) Epoch 18, batch 1650, loss[loss=0.1691, simple_loss=0.222, pruned_loss=0.0581, over 13530.00 frames. ], tot_loss[loss=0.1858, simple_loss=0.2445, pruned_loss=0.06352, over 2657744.49 frames. ], batch size: 70, lr: 7.26e-03, grad_scale: 32.0 2023-04-16 23:55:41,254 INFO [train.py:893] (1/4) Epoch 18, batch 1700, loss[loss=0.1894, simple_loss=0.2484, pruned_loss=0.06519, over 13259.00 frames. ], tot_loss[loss=0.1864, simple_loss=0.2452, pruned_loss=0.06382, over 2656304.79 frames. ], batch size: 124, lr: 7.26e-03, grad_scale: 32.0 2023-04-16 23:55:52,306 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47597.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:56:01,355 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.73 vs. limit=5.0 2023-04-16 23:56:10,031 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.925e+02 2.763e+02 3.195e+02 3.652e+02 5.835e+02, threshold=6.391e+02, percent-clipped=0.0 2023-04-16 23:56:27,288 INFO [train.py:893] (1/4) Epoch 18, batch 1750, loss[loss=0.1656, simple_loss=0.2243, pruned_loss=0.05347, over 13328.00 frames. ], tot_loss[loss=0.1848, simple_loss=0.2435, pruned_loss=0.06306, over 2656918.25 frames. ], batch size: 69, lr: 7.25e-03, grad_scale: 32.0 2023-04-16 23:56:29,518 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.57 vs. limit=5.0 2023-04-16 23:56:53,250 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-04-16 23:57:02,324 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9332, 3.9252, 2.7112, 3.6717, 3.8181, 2.4255, 3.4885, 2.6393], device='cuda:1'), covar=tensor([0.0282, 0.0202, 0.1124, 0.0408, 0.0239, 0.1335, 0.0555, 0.1335], device='cuda:1'), in_proj_covar=tensor([0.0158, 0.0169, 0.0177, 0.0199, 0.0134, 0.0160, 0.0159, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-16 23:57:14,868 INFO [train.py:893] (1/4) Epoch 18, batch 1800, loss[loss=0.1797, simple_loss=0.2376, pruned_loss=0.06095, over 13553.00 frames. ], tot_loss[loss=0.1842, simple_loss=0.2433, pruned_loss=0.06255, over 2658338.39 frames. ], batch size: 72, lr: 7.25e-03, grad_scale: 32.0 2023-04-16 23:57:43,293 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.712e+02 2.641e+02 3.042e+02 3.514e+02 5.198e+02, threshold=6.083e+02, percent-clipped=0.0 2023-04-16 23:57:59,986 INFO [train.py:893] (1/4) Epoch 18, batch 1850, loss[loss=0.1891, simple_loss=0.2511, pruned_loss=0.06359, over 13285.00 frames. ], tot_loss[loss=0.1836, simple_loss=0.2424, pruned_loss=0.06245, over 2655605.70 frames. ], batch size: 118, lr: 7.25e-03, grad_scale: 32.0 2023-04-16 23:58:01,015 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=47735.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:58:04,698 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47738.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:58:05,304 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-16 23:58:48,057 INFO [train.py:893] (1/4) Epoch 18, batch 1900, loss[loss=0.153, simple_loss=0.2217, pruned_loss=0.04212, over 13465.00 frames. ], tot_loss[loss=0.1835, simple_loss=0.2419, pruned_loss=0.0626, over 2658521.77 frames. ], batch size: 79, lr: 7.24e-03, grad_scale: 32.0 2023-04-16 23:58:50,884 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47788.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:58:52,781 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-04-16 23:59:00,858 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47799.0, num_to_drop=0, layers_to_drop=set() 2023-04-16 23:59:10,469 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=47809.0, num_to_drop=1, layers_to_drop={1} 2023-04-16 23:59:16,787 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.811e+02 2.747e+02 3.190e+02 3.655e+02 6.040e+02, threshold=6.381e+02, percent-clipped=0.0 2023-04-16 23:59:34,986 INFO [train.py:893] (1/4) Epoch 18, batch 1950, loss[loss=0.2196, simple_loss=0.2731, pruned_loss=0.08304, over 13530.00 frames. ], tot_loss[loss=0.1837, simple_loss=0.2421, pruned_loss=0.0627, over 2660086.53 frames. ], batch size: 83, lr: 7.24e-03, grad_scale: 32.0 2023-04-16 23:59:48,077 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47849.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:00:07,752 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=47870.0, num_to_drop=1, layers_to_drop={2} 2023-04-17 00:00:20,509 INFO [train.py:893] (1/4) Epoch 18, batch 2000, loss[loss=0.2265, simple_loss=0.2716, pruned_loss=0.09069, over 11885.00 frames. ], tot_loss[loss=0.1862, simple_loss=0.2442, pruned_loss=0.06413, over 2657870.85 frames. ], batch size: 157, lr: 7.24e-03, grad_scale: 32.0 2023-04-17 00:00:26,473 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-17 00:00:33,184 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=47897.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:00:49,669 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.108e+02 2.833e+02 3.370e+02 4.001e+02 7.543e+02, threshold=6.741e+02, percent-clipped=2.0 2023-04-17 00:00:53,016 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.87 vs. limit=5.0 2023-04-17 00:01:08,117 INFO [train.py:893] (1/4) Epoch 18, batch 2050, loss[loss=0.1568, simple_loss=0.2181, pruned_loss=0.04772, over 13478.00 frames. ], tot_loss[loss=0.1873, simple_loss=0.2452, pruned_loss=0.06466, over 2657599.68 frames. ], batch size: 65, lr: 7.23e-03, grad_scale: 32.0 2023-04-17 00:01:16,692 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=47945.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:01:37,040 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.84 vs. limit=5.0 2023-04-17 00:01:54,811 INFO [train.py:893] (1/4) Epoch 18, batch 2100, loss[loss=0.189, simple_loss=0.2492, pruned_loss=0.06439, over 13386.00 frames. ], tot_loss[loss=0.1864, simple_loss=0.2447, pruned_loss=0.06408, over 2657051.37 frames. ], batch size: 113, lr: 7.23e-03, grad_scale: 32.0 2023-04-17 00:02:27,052 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.728e+02 2.550e+02 3.091e+02 3.728e+02 7.028e+02, threshold=6.181e+02, percent-clipped=2.0 2023-04-17 00:02:45,195 INFO [train.py:893] (1/4) Epoch 18, batch 2150, loss[loss=0.1696, simple_loss=0.2349, pruned_loss=0.05214, over 13514.00 frames. ], tot_loss[loss=0.186, simple_loss=0.2447, pruned_loss=0.06364, over 2659466.38 frames. ], batch size: 85, lr: 7.22e-03, grad_scale: 32.0 2023-04-17 00:02:45,470 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48035.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:03:20,046 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48072.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:03:30,786 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48083.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:03:32,347 INFO [train.py:893] (1/4) Epoch 18, batch 2200, loss[loss=0.1557, simple_loss=0.2146, pruned_loss=0.04835, over 13505.00 frames. ], tot_loss[loss=0.1846, simple_loss=0.2436, pruned_loss=0.06282, over 2659968.35 frames. ], batch size: 70, lr: 7.22e-03, grad_scale: 32.0 2023-04-17 00:03:40,737 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48094.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:03:58,514 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3834, 2.6827, 2.3647, 4.2302, 4.7369, 3.5518, 4.6196, 4.3915], device='cuda:1'), covar=tensor([0.0085, 0.0840, 0.0903, 0.0093, 0.0067, 0.0388, 0.0062, 0.0077], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0089, 0.0097, 0.0079, 0.0064, 0.0080, 0.0054, 0.0074], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 00:04:00,791 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.580e+02 2.476e+02 3.037e+02 3.602e+02 5.277e+02, threshold=6.074e+02, percent-clipped=0.0 2023-04-17 00:04:15,804 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48133.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:04:17,169 INFO [train.py:893] (1/4) Epoch 18, batch 2250, loss[loss=0.147, simple_loss=0.2071, pruned_loss=0.04344, over 13523.00 frames. ], tot_loss[loss=0.1845, simple_loss=0.2432, pruned_loss=0.06288, over 2659073.02 frames. ], batch size: 70, lr: 7.22e-03, grad_scale: 32.0 2023-04-17 00:04:26,581 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48144.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:04:45,109 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48165.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 00:05:04,045 INFO [train.py:893] (1/4) Epoch 18, batch 2300, loss[loss=0.1933, simple_loss=0.2498, pruned_loss=0.06841, over 13546.00 frames. ], tot_loss[loss=0.1828, simple_loss=0.2418, pruned_loss=0.06194, over 2662844.15 frames. ], batch size: 83, lr: 7.21e-03, grad_scale: 32.0 2023-04-17 00:05:30,046 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48212.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 00:05:33,475 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-04-17 00:05:33,716 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.863e+02 2.589e+02 2.983e+02 3.618e+02 6.211e+02, threshold=5.966e+02, percent-clipped=2.0 2023-04-17 00:05:49,815 INFO [train.py:893] (1/4) Epoch 18, batch 2350, loss[loss=0.1552, simple_loss=0.2167, pruned_loss=0.04691, over 13352.00 frames. ], tot_loss[loss=0.1822, simple_loss=0.241, pruned_loss=0.06168, over 2662608.67 frames. ], batch size: 67, lr: 7.21e-03, grad_scale: 32.0 2023-04-17 00:06:05,947 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3904, 4.1869, 4.3170, 2.9993, 4.7345, 4.4325, 4.4096, 4.6228], device='cuda:1'), covar=tensor([0.0250, 0.0139, 0.0168, 0.1006, 0.0150, 0.0259, 0.0145, 0.0144], device='cuda:1'), in_proj_covar=tensor([0.0087, 0.0052, 0.0078, 0.0101, 0.0095, 0.0100, 0.0076, 0.0071], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 00:06:12,408 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-17 00:06:26,107 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48273.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 00:06:36,448 INFO [train.py:893] (1/4) Epoch 18, batch 2400, loss[loss=0.179, simple_loss=0.2337, pruned_loss=0.06217, over 13519.00 frames. ], tot_loss[loss=0.1825, simple_loss=0.241, pruned_loss=0.06197, over 2662417.98 frames. ], batch size: 91, lr: 7.21e-03, grad_scale: 32.0 2023-04-17 00:06:37,680 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1110, 4.3762, 3.3514, 3.0112, 3.1898, 2.5750, 4.5466, 2.5722], device='cuda:1'), covar=tensor([0.1606, 0.0319, 0.1152, 0.1953, 0.0744, 0.3406, 0.0194, 0.3890], device='cuda:1'), in_proj_covar=tensor([0.0306, 0.0282, 0.0304, 0.0319, 0.0251, 0.0319, 0.0204, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 00:06:51,759 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2717, 4.0580, 4.2600, 2.7058, 4.5195, 4.3281, 4.2859, 4.5051], device='cuda:1'), covar=tensor([0.0235, 0.0144, 0.0118, 0.1085, 0.0131, 0.0196, 0.0123, 0.0103], device='cuda:1'), in_proj_covar=tensor([0.0087, 0.0052, 0.0078, 0.0100, 0.0095, 0.0100, 0.0076, 0.0071], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 00:07:02,440 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0410, 4.1512, 2.8262, 3.8948, 3.9904, 2.4947, 3.6794, 2.7645], device='cuda:1'), covar=tensor([0.0272, 0.0262, 0.1080, 0.0425, 0.0296, 0.1286, 0.0529, 0.1192], device='cuda:1'), in_proj_covar=tensor([0.0157, 0.0169, 0.0177, 0.0199, 0.0135, 0.0161, 0.0159, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 00:07:05,451 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.915e+02 2.726e+02 3.153e+02 3.503e+02 4.945e+02, threshold=6.306e+02, percent-clipped=0.0 2023-04-17 00:07:12,376 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48324.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:07:22,289 INFO [train.py:893] (1/4) Epoch 18, batch 2450, loss[loss=0.2019, simple_loss=0.2621, pruned_loss=0.07081, over 13431.00 frames. ], tot_loss[loss=0.1827, simple_loss=0.2415, pruned_loss=0.06202, over 2666418.34 frames. ], batch size: 95, lr: 7.20e-03, grad_scale: 32.0 2023-04-17 00:08:09,077 INFO [train.py:893] (1/4) Epoch 18, batch 2500, loss[loss=0.1675, simple_loss=0.2261, pruned_loss=0.05447, over 13533.00 frames. ], tot_loss[loss=0.182, simple_loss=0.2407, pruned_loss=0.0616, over 2663405.95 frames. ], batch size: 72, lr: 7.20e-03, grad_scale: 32.0 2023-04-17 00:08:09,405 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48385.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:08:16,161 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0447, 4.0576, 2.8418, 3.8069, 3.9664, 2.5736, 3.6421, 2.6848], device='cuda:1'), covar=tensor([0.0263, 0.0234, 0.1103, 0.0359, 0.0267, 0.1273, 0.0515, 0.1324], device='cuda:1'), in_proj_covar=tensor([0.0158, 0.0170, 0.0178, 0.0200, 0.0136, 0.0162, 0.0160, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 00:08:16,878 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48394.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:08:21,137 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48397.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:08:38,951 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.890e+02 2.803e+02 3.213e+02 3.680e+02 8.016e+02, threshold=6.426e+02, percent-clipped=1.0 2023-04-17 00:08:45,239 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48423.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:08:50,109 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48428.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:08:52,040 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-04-17 00:08:52,697 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6343, 3.4825, 4.3926, 3.0403, 2.8971, 2.9643, 4.5576, 4.6199], device='cuda:1'), covar=tensor([0.1399, 0.1801, 0.0354, 0.1854, 0.1620, 0.1533, 0.0252, 0.0382], device='cuda:1'), in_proj_covar=tensor([0.0241, 0.0262, 0.0189, 0.0219, 0.0216, 0.0179, 0.0200, 0.0169], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-17 00:08:57,176 INFO [train.py:893] (1/4) Epoch 18, batch 2550, loss[loss=0.2115, simple_loss=0.2667, pruned_loss=0.07814, over 13234.00 frames. ], tot_loss[loss=0.1813, simple_loss=0.24, pruned_loss=0.06134, over 2660938.72 frames. ], batch size: 117, lr: 7.19e-03, grad_scale: 32.0 2023-04-17 00:09:03,262 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48442.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:09:05,039 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48444.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:09:06,272 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-04-17 00:09:17,659 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-17 00:09:17,887 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48458.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:09:26,027 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48465.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 00:09:26,933 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.8622, 2.9511, 2.5486, 1.8342, 2.0113, 2.5084, 2.4533, 3.1118], device='cuda:1'), covar=tensor([0.0987, 0.0305, 0.0923, 0.1541, 0.0483, 0.0563, 0.0828, 0.0292], device='cuda:1'), in_proj_covar=tensor([0.0158, 0.0130, 0.0114, 0.0201, 0.0104, 0.0154, 0.0160, 0.0125], device='cuda:1'), out_proj_covar=tensor([1.1691e-04, 9.6902e-05, 8.9940e-05, 1.5036e-04, 7.7149e-05, 1.1629e-04, 1.2153e-04, 9.2609e-05], device='cuda:1') 2023-04-17 00:09:30,152 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2996, 4.7850, 4.6004, 4.6124, 4.5407, 4.4007, 4.8861, 4.8867], device='cuda:1'), covar=tensor([0.0201, 0.0209, 0.0175, 0.0267, 0.0313, 0.0246, 0.0259, 0.0204], device='cuda:1'), in_proj_covar=tensor([0.0135, 0.0196, 0.0154, 0.0175, 0.0140, 0.0190, 0.0129, 0.0171], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 00:09:41,885 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48484.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:09:42,389 INFO [train.py:893] (1/4) Epoch 18, batch 2600, loss[loss=0.1824, simple_loss=0.2422, pruned_loss=0.06126, over 13449.00 frames. ], tot_loss[loss=0.1815, simple_loss=0.2401, pruned_loss=0.06149, over 2655442.44 frames. ], batch size: 106, lr: 7.19e-03, grad_scale: 32.0 2023-04-17 00:09:50,696 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48492.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:10:08,213 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48513.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 00:10:10,308 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.856e+02 2.747e+02 3.259e+02 4.082e+02 6.442e+02, threshold=6.518e+02, percent-clipped=1.0 2023-04-17 00:10:24,715 INFO [train.py:893] (1/4) Epoch 18, batch 2650, loss[loss=0.2167, simple_loss=0.2762, pruned_loss=0.07858, over 13236.00 frames. ], tot_loss[loss=0.1825, simple_loss=0.2409, pruned_loss=0.06199, over 2657871.35 frames. ], batch size: 124, lr: 7.19e-03, grad_scale: 32.0 2023-04-17 00:10:49,941 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48568.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 00:11:21,497 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-17 00:11:32,187 INFO [train.py:893] (1/4) Epoch 19, batch 0, loss[loss=0.1742, simple_loss=0.23, pruned_loss=0.05922, over 13256.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.23, pruned_loss=0.05922, over 13256.00 frames. ], batch size: 124, lr: 6.99e-03, grad_scale: 32.0 2023-04-17 00:11:32,187 INFO [train.py:918] (1/4) Computing validation loss 2023-04-17 00:11:53,918 INFO [train.py:927] (1/4) Epoch 19, validation: loss=0.1405, simple_loss=0.1995, pruned_loss=0.04077, over 2446609.00 frames. 2023-04-17 00:11:53,919 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12831MB 2023-04-17 00:11:58,361 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4224, 2.4393, 2.3677, 4.2129, 4.6821, 3.4801, 4.7152, 4.5075], device='cuda:1'), covar=tensor([0.0123, 0.1287, 0.1371, 0.0168, 0.0232, 0.0670, 0.0140, 0.0119], device='cuda:1'), in_proj_covar=tensor([0.0072, 0.0088, 0.0097, 0.0078, 0.0064, 0.0079, 0.0054, 0.0075], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 00:12:23,917 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.860e+02 2.769e+02 3.155e+02 3.700e+02 8.510e+02, threshold=6.310e+02, percent-clipped=1.0 2023-04-17 00:12:41,452 INFO [train.py:893] (1/4) Epoch 19, batch 50, loss[loss=0.1646, simple_loss=0.2265, pruned_loss=0.05142, over 13536.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2334, pruned_loss=0.05959, over 603932.26 frames. ], batch size: 72, lr: 6.99e-03, grad_scale: 32.0 2023-04-17 00:13:06,423 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-17 00:13:06,424 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-17 00:13:06,424 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-17 00:13:06,431 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-17 00:13:07,980 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-17 00:13:07,994 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-17 00:13:08,012 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-17 00:13:23,762 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48680.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:13:26,848 INFO [train.py:893] (1/4) Epoch 19, batch 100, loss[loss=0.2163, simple_loss=0.2631, pruned_loss=0.08478, over 13425.00 frames. ], tot_loss[loss=0.1821, simple_loss=0.2377, pruned_loss=0.06321, over 1055097.86 frames. ], batch size: 88, lr: 6.98e-03, grad_scale: 32.0 2023-04-17 00:13:56,997 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.794e+02 2.831e+02 3.394e+02 4.079e+02 8.667e+02, threshold=6.787e+02, percent-clipped=3.0 2023-04-17 00:14:08,203 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7215, 3.7079, 2.8173, 3.4003, 3.6435, 2.2673, 3.2568, 2.5840], device='cuda:1'), covar=tensor([0.0341, 0.0262, 0.0990, 0.0408, 0.0265, 0.1295, 0.0611, 0.1306], device='cuda:1'), in_proj_covar=tensor([0.0159, 0.0170, 0.0177, 0.0199, 0.0135, 0.0160, 0.0159, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 00:14:08,902 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48728.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:14:14,723 INFO [train.py:893] (1/4) Epoch 19, batch 150, loss[loss=0.1707, simple_loss=0.238, pruned_loss=0.05167, over 13498.00 frames. ], tot_loss[loss=0.1844, simple_loss=0.2401, pruned_loss=0.06433, over 1401766.79 frames. ], batch size: 81, lr: 6.98e-03, grad_scale: 32.0 2023-04-17 00:14:31,377 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48753.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:14:46,621 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7952, 3.2553, 3.1882, 3.4030, 2.1989, 2.9354, 3.4182, 1.9850], device='cuda:1'), covar=tensor([0.0204, 0.0487, 0.0606, 0.0364, 0.1282, 0.0768, 0.0522, 0.1711], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0180, 0.0198, 0.0230, 0.0179, 0.0194, 0.0174, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 00:14:53,791 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48776.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:14:56,384 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=48779.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:14:59,779 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7337, 2.3583, 2.3983, 2.7758, 2.0519, 2.7627, 2.7027, 2.3227], device='cuda:1'), covar=tensor([0.0068, 0.0181, 0.0121, 0.0135, 0.0184, 0.0131, 0.0151, 0.0142], device='cuda:1'), in_proj_covar=tensor([0.0097, 0.0107, 0.0114, 0.0111, 0.0121, 0.0101, 0.0101, 0.0098], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 00:15:00,327 INFO [train.py:893] (1/4) Epoch 19, batch 200, loss[loss=0.1949, simple_loss=0.2553, pruned_loss=0.06731, over 13444.00 frames. ], tot_loss[loss=0.1876, simple_loss=0.243, pruned_loss=0.06605, over 1676004.91 frames. ], batch size: 103, lr: 6.98e-03, grad_scale: 32.0 2023-04-17 00:15:16,065 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-04-17 00:15:17,146 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-04-17 00:15:30,382 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.935e+02 2.775e+02 3.304e+02 3.931e+02 5.890e+02, threshold=6.608e+02, percent-clipped=0.0 2023-04-17 00:15:36,448 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48821.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:15:41,912 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2023-04-17 00:15:47,950 INFO [train.py:893] (1/4) Epoch 19, batch 250, loss[loss=0.2035, simple_loss=0.2617, pruned_loss=0.07265, over 13439.00 frames. ], tot_loss[loss=0.1888, simple_loss=0.2443, pruned_loss=0.06663, over 1891453.81 frames. ], batch size: 106, lr: 6.97e-03, grad_scale: 32.0 2023-04-17 00:15:52,185 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=48838.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:16:20,153 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48868.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 00:16:22,804 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3175, 2.8304, 2.8426, 3.3710, 2.7315, 3.3689, 3.1847, 2.9790], device='cuda:1'), covar=tensor([0.0062, 0.0151, 0.0124, 0.0110, 0.0161, 0.0091, 0.0170, 0.0102], device='cuda:1'), in_proj_covar=tensor([0.0097, 0.0107, 0.0114, 0.0111, 0.0122, 0.0101, 0.0101, 0.0099], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 00:16:31,887 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48882.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:16:34,312 INFO [train.py:893] (1/4) Epoch 19, batch 300, loss[loss=0.1748, simple_loss=0.2358, pruned_loss=0.0569, over 13475.00 frames. ], tot_loss[loss=0.1878, simple_loss=0.2437, pruned_loss=0.06594, over 2054027.25 frames. ], batch size: 100, lr: 6.97e-03, grad_scale: 32.0 2023-04-17 00:16:49,262 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=48899.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:17:01,173 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7578, 3.6426, 4.4863, 3.1984, 3.0079, 3.0787, 4.7001, 4.7600], device='cuda:1'), covar=tensor([0.1056, 0.1537, 0.0287, 0.1463, 0.1465, 0.1359, 0.0283, 0.0259], device='cuda:1'), in_proj_covar=tensor([0.0242, 0.0259, 0.0189, 0.0218, 0.0215, 0.0178, 0.0200, 0.0170], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-17 00:17:03,855 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.60 vs. limit=5.0 2023-04-17 00:17:04,063 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.909e+02 2.666e+02 3.189e+02 3.769e+02 7.720e+02, threshold=6.378e+02, percent-clipped=3.0 2023-04-17 00:17:04,982 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=48916.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 00:17:20,221 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.47 vs. limit=5.0 2023-04-17 00:17:21,370 INFO [train.py:893] (1/4) Epoch 19, batch 350, loss[loss=0.2005, simple_loss=0.2559, pruned_loss=0.07256, over 13456.00 frames. ], tot_loss[loss=0.1891, simple_loss=0.2447, pruned_loss=0.06676, over 2190640.93 frames. ], batch size: 100, lr: 6.96e-03, grad_scale: 32.0 2023-04-17 00:18:04,412 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=48980.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:18:08,473 INFO [train.py:893] (1/4) Epoch 19, batch 400, loss[loss=0.1879, simple_loss=0.2484, pruned_loss=0.06366, over 13433.00 frames. ], tot_loss[loss=0.189, simple_loss=0.2452, pruned_loss=0.06644, over 2293901.06 frames. ], batch size: 95, lr: 6.96e-03, grad_scale: 64.0 2023-04-17 00:18:34,235 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3369, 3.2299, 3.8460, 2.7625, 2.4703, 2.6774, 4.1177, 4.1977], device='cuda:1'), covar=tensor([0.1054, 0.1524, 0.0332, 0.1492, 0.1485, 0.1470, 0.0322, 0.0240], device='cuda:1'), in_proj_covar=tensor([0.0239, 0.0257, 0.0187, 0.0215, 0.0212, 0.0177, 0.0198, 0.0168], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-17 00:18:38,841 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.896e+02 2.600e+02 3.085e+02 3.830e+02 7.566e+02, threshold=6.170e+02, percent-clipped=2.0 2023-04-17 00:18:49,718 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49028.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:18:54,458 INFO [train.py:893] (1/4) Epoch 19, batch 450, loss[loss=0.1813, simple_loss=0.2388, pruned_loss=0.06193, over 13533.00 frames. ], tot_loss[loss=0.1894, simple_loss=0.246, pruned_loss=0.06643, over 2374221.76 frames. ], batch size: 83, lr: 6.96e-03, grad_scale: 64.0 2023-04-17 00:19:12,836 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49053.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:19:21,048 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-17 00:19:38,086 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49079.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:19:41,978 INFO [train.py:893] (1/4) Epoch 19, batch 500, loss[loss=0.1943, simple_loss=0.2545, pruned_loss=0.06703, over 13402.00 frames. ], tot_loss[loss=0.1895, simple_loss=0.2467, pruned_loss=0.06614, over 2437606.71 frames. ], batch size: 113, lr: 6.95e-03, grad_scale: 64.0 2023-04-17 00:19:56,339 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3663, 3.1822, 3.7862, 2.8773, 2.4113, 2.7204, 4.1303, 4.1763], device='cuda:1'), covar=tensor([0.1061, 0.1538, 0.0307, 0.1404, 0.1535, 0.1352, 0.0227, 0.0241], device='cuda:1'), in_proj_covar=tensor([0.0239, 0.0256, 0.0186, 0.0215, 0.0212, 0.0176, 0.0198, 0.0169], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-17 00:19:56,940 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49101.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:20:12,367 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.858e+02 2.668e+02 2.993e+02 3.591e+02 6.396e+02, threshold=5.987e+02, percent-clipped=1.0 2023-04-17 00:20:21,839 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49127.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:20:28,417 INFO [train.py:893] (1/4) Epoch 19, batch 550, loss[loss=0.1818, simple_loss=0.2392, pruned_loss=0.06226, over 13205.00 frames. ], tot_loss[loss=0.1889, simple_loss=0.2463, pruned_loss=0.06578, over 2488524.81 frames. ], batch size: 132, lr: 6.95e-03, grad_scale: 32.0 2023-04-17 00:21:09,539 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49177.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:21:16,000 INFO [train.py:893] (1/4) Epoch 19, batch 600, loss[loss=0.1915, simple_loss=0.2478, pruned_loss=0.0676, over 11898.00 frames. ], tot_loss[loss=0.187, simple_loss=0.2443, pruned_loss=0.06482, over 2523910.11 frames. ], batch size: 157, lr: 6.95e-03, grad_scale: 32.0 2023-04-17 00:21:24,541 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49194.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:21:47,110 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.916e+02 2.549e+02 2.912e+02 3.543e+02 6.651e+02, threshold=5.825e+02, percent-clipped=1.0 2023-04-17 00:21:50,767 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9245, 3.7515, 3.1489, 3.4818, 3.0194, 2.1393, 3.7704, 2.0496], device='cuda:1'), covar=tensor([0.0627, 0.0398, 0.0404, 0.0321, 0.0698, 0.1782, 0.0785, 0.1354], device='cuda:1'), in_proj_covar=tensor([0.0140, 0.0136, 0.0133, 0.0115, 0.0148, 0.0187, 0.0167, 0.0139], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 00:22:02,910 INFO [train.py:893] (1/4) Epoch 19, batch 650, loss[loss=0.1693, simple_loss=0.2361, pruned_loss=0.05129, over 13429.00 frames. ], tot_loss[loss=0.1867, simple_loss=0.2442, pruned_loss=0.06458, over 2553001.75 frames. ], batch size: 106, lr: 6.94e-03, grad_scale: 32.0 2023-04-17 00:22:04,876 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49236.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:22:49,389 INFO [train.py:893] (1/4) Epoch 19, batch 700, loss[loss=0.1686, simple_loss=0.2262, pruned_loss=0.05544, over 13346.00 frames. ], tot_loss[loss=0.1848, simple_loss=0.243, pruned_loss=0.06327, over 2576375.24 frames. ], batch size: 73, lr: 6.94e-03, grad_scale: 32.0 2023-04-17 00:22:49,747 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6053, 2.3370, 2.2643, 2.6795, 2.0409, 2.7020, 2.6520, 2.2398], device='cuda:1'), covar=tensor([0.0126, 0.0211, 0.0159, 0.0150, 0.0212, 0.0158, 0.0199, 0.0189], device='cuda:1'), in_proj_covar=tensor([0.0099, 0.0109, 0.0116, 0.0113, 0.0125, 0.0103, 0.0102, 0.0101], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 00:23:02,182 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49297.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:23:20,167 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.748e+02 2.893e+02 3.409e+02 4.077e+02 6.619e+02, threshold=6.818e+02, percent-clipped=5.0 2023-04-17 00:23:36,837 INFO [train.py:893] (1/4) Epoch 19, batch 750, loss[loss=0.1881, simple_loss=0.2425, pruned_loss=0.06691, over 13467.00 frames. ], tot_loss[loss=0.1848, simple_loss=0.2427, pruned_loss=0.06341, over 2597186.18 frames. ], batch size: 79, lr: 6.94e-03, grad_scale: 32.0 2023-04-17 00:24:23,048 INFO [train.py:893] (1/4) Epoch 19, batch 800, loss[loss=0.1908, simple_loss=0.2521, pruned_loss=0.06476, over 13490.00 frames. ], tot_loss[loss=0.1866, simple_loss=0.2443, pruned_loss=0.06443, over 2606243.58 frames. ], batch size: 93, lr: 6.93e-03, grad_scale: 32.0 2023-04-17 00:24:24,122 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.8618, 4.2477, 3.9923, 4.6273, 2.8959, 3.4967, 4.3821, 2.5683], device='cuda:1'), covar=tensor([0.0213, 0.0372, 0.0630, 0.0523, 0.1200, 0.0810, 0.0321, 0.1593], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0178, 0.0197, 0.0228, 0.0177, 0.0192, 0.0172, 0.0170], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 00:24:55,576 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.943e+02 2.825e+02 3.338e+02 3.802e+02 7.720e+02, threshold=6.675e+02, percent-clipped=1.0 2023-04-17 00:24:59,991 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49422.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:25:09,070 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49432.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 00:25:10,421 INFO [train.py:893] (1/4) Epoch 19, batch 850, loss[loss=0.1595, simple_loss=0.2237, pruned_loss=0.04763, over 13348.00 frames. ], tot_loss[loss=0.1873, simple_loss=0.245, pruned_loss=0.06477, over 2618678.43 frames. ], batch size: 73, lr: 6.93e-03, grad_scale: 32.0 2023-04-17 00:25:52,384 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49477.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:25:57,524 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49483.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:25:57,992 INFO [train.py:893] (1/4) Epoch 19, batch 900, loss[loss=0.1916, simple_loss=0.2463, pruned_loss=0.06844, over 13478.00 frames. ], tot_loss[loss=0.1869, simple_loss=0.2442, pruned_loss=0.06483, over 2626845.62 frames. ], batch size: 93, lr: 6.93e-03, grad_scale: 32.0 2023-04-17 00:26:06,717 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49493.0, num_to_drop=1, layers_to_drop={2} 2023-04-17 00:26:07,389 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49494.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:26:28,267 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.933e+02 2.628e+02 3.112e+02 3.839e+02 6.691e+02, threshold=6.223e+02, percent-clipped=1.0 2023-04-17 00:26:28,336 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-17 00:26:36,309 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49525.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:26:44,233 INFO [train.py:893] (1/4) Epoch 19, batch 950, loss[loss=0.2064, simple_loss=0.2523, pruned_loss=0.08027, over 13347.00 frames. ], tot_loss[loss=0.186, simple_loss=0.2429, pruned_loss=0.06459, over 2636807.45 frames. ], batch size: 73, lr: 6.92e-03, grad_scale: 32.0 2023-04-17 00:26:52,493 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49542.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:27:21,997 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3244, 4.3869, 3.0895, 4.2169, 4.3032, 2.6988, 3.9911, 2.9775], device='cuda:1'), covar=tensor([0.0244, 0.0266, 0.0983, 0.0400, 0.0220, 0.1183, 0.0385, 0.1211], device='cuda:1'), in_proj_covar=tensor([0.0158, 0.0170, 0.0175, 0.0200, 0.0135, 0.0161, 0.0159, 0.0159], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 00:27:30,887 INFO [train.py:893] (1/4) Epoch 19, batch 1000, loss[loss=0.1583, simple_loss=0.2201, pruned_loss=0.04823, over 13510.00 frames. ], tot_loss[loss=0.1844, simple_loss=0.2408, pruned_loss=0.06402, over 2631187.26 frames. ], batch size: 91, lr: 6.92e-03, grad_scale: 32.0 2023-04-17 00:27:39,092 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49592.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:27:45,014 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3666, 4.7465, 4.5127, 4.5425, 4.5129, 4.3750, 4.7980, 4.8313], device='cuda:1'), covar=tensor([0.0182, 0.0200, 0.0179, 0.0299, 0.0235, 0.0253, 0.0228, 0.0195], device='cuda:1'), in_proj_covar=tensor([0.0135, 0.0195, 0.0156, 0.0176, 0.0141, 0.0191, 0.0127, 0.0172], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 00:28:01,912 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.985e+02 2.718e+02 3.260e+02 3.856e+02 5.927e+02, threshold=6.520e+02, percent-clipped=0.0 2023-04-17 00:28:04,273 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-04-17 00:28:17,891 INFO [train.py:893] (1/4) Epoch 19, batch 1050, loss[loss=0.1572, simple_loss=0.2199, pruned_loss=0.0473, over 13460.00 frames. ], tot_loss[loss=0.1832, simple_loss=0.2401, pruned_loss=0.06314, over 2638884.99 frames. ], batch size: 79, lr: 6.92e-03, grad_scale: 32.0 2023-04-17 00:29:04,574 INFO [train.py:893] (1/4) Epoch 19, batch 1100, loss[loss=0.178, simple_loss=0.2386, pruned_loss=0.0587, over 13469.00 frames. ], tot_loss[loss=0.1838, simple_loss=0.2409, pruned_loss=0.0633, over 2638850.82 frames. ], batch size: 79, lr: 6.91e-03, grad_scale: 32.0 2023-04-17 00:29:35,041 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.663e+02 2.716e+02 3.149e+02 3.915e+02 7.191e+02, threshold=6.299e+02, percent-clipped=1.0 2023-04-17 00:29:36,599 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.92 vs. limit=2.0 2023-04-17 00:29:52,589 INFO [train.py:893] (1/4) Epoch 19, batch 1150, loss[loss=0.1872, simple_loss=0.2427, pruned_loss=0.06581, over 13292.00 frames. ], tot_loss[loss=0.183, simple_loss=0.241, pruned_loss=0.06252, over 2643151.18 frames. ], batch size: 124, lr: 6.91e-03, grad_scale: 32.0 2023-04-17 00:29:54,561 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7618, 2.6108, 2.2854, 1.6262, 1.5287, 1.9364, 2.2609, 2.7680], device='cuda:1'), covar=tensor([0.0880, 0.0266, 0.0616, 0.1451, 0.0225, 0.0404, 0.0652, 0.0158], device='cuda:1'), in_proj_covar=tensor([0.0159, 0.0132, 0.0115, 0.0203, 0.0106, 0.0152, 0.0164, 0.0125], device='cuda:1'), out_proj_covar=tensor([1.1779e-04, 9.8647e-05, 9.0001e-05, 1.5205e-04, 7.8144e-05, 1.1464e-04, 1.2458e-04, 9.2948e-05], device='cuda:1') 2023-04-17 00:30:22,771 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=49766.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:30:32,617 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49778.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:30:39,138 INFO [train.py:893] (1/4) Epoch 19, batch 1200, loss[loss=0.2033, simple_loss=0.2641, pruned_loss=0.07123, over 13216.00 frames. ], tot_loss[loss=0.1826, simple_loss=0.2408, pruned_loss=0.06217, over 2647501.53 frames. ], batch size: 132, lr: 6.91e-03, grad_scale: 32.0 2023-04-17 00:30:42,756 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=49788.0, num_to_drop=1, layers_to_drop={2} 2023-04-17 00:30:50,652 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-04-17 00:31:06,133 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-17 00:31:10,210 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.639e+02 2.701e+02 3.177e+02 3.768e+02 6.623e+02, threshold=6.354e+02, percent-clipped=1.0 2023-04-17 00:31:19,512 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-17 00:31:20,632 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=49827.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:31:26,184 INFO [train.py:893] (1/4) Epoch 19, batch 1250, loss[loss=0.169, simple_loss=0.2157, pruned_loss=0.06117, over 12818.00 frames. ], tot_loss[loss=0.1831, simple_loss=0.2414, pruned_loss=0.06241, over 2652571.11 frames. ], batch size: 52, lr: 6.90e-03, grad_scale: 32.0 2023-04-17 00:31:47,376 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8472, 4.0198, 3.0846, 2.7764, 2.8498, 2.4588, 4.1249, 2.3559], device='cuda:1'), covar=tensor([0.1684, 0.0330, 0.1155, 0.1958, 0.0882, 0.3179, 0.0234, 0.3869], device='cuda:1'), in_proj_covar=tensor([0.0307, 0.0283, 0.0306, 0.0320, 0.0251, 0.0320, 0.0205, 0.0359], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 00:32:13,449 INFO [train.py:893] (1/4) Epoch 19, batch 1300, loss[loss=0.1604, simple_loss=0.2278, pruned_loss=0.04647, over 13445.00 frames. ], tot_loss[loss=0.1845, simple_loss=0.2428, pruned_loss=0.06309, over 2657958.86 frames. ], batch size: 79, lr: 6.90e-03, grad_scale: 32.0 2023-04-17 00:32:21,118 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=49892.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:32:44,298 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.904e+02 2.679e+02 3.163e+02 3.782e+02 7.559e+02, threshold=6.326e+02, percent-clipped=2.0 2023-04-17 00:32:59,140 INFO [train.py:893] (1/4) Epoch 19, batch 1350, loss[loss=0.1974, simple_loss=0.2637, pruned_loss=0.06555, over 13448.00 frames. ], tot_loss[loss=0.1851, simple_loss=0.2436, pruned_loss=0.06332, over 2659954.18 frames. ], batch size: 106, lr: 6.90e-03, grad_scale: 32.0 2023-04-17 00:33:05,816 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=49940.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:33:17,527 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1846, 2.7456, 2.1305, 4.1282, 4.6340, 3.5252, 4.5454, 4.3323], device='cuda:1'), covar=tensor([0.0099, 0.0775, 0.0992, 0.0096, 0.0052, 0.0402, 0.0067, 0.0072], device='cuda:1'), in_proj_covar=tensor([0.0074, 0.0088, 0.0096, 0.0079, 0.0064, 0.0079, 0.0054, 0.0075], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 00:33:36,416 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2023-04-17 00:33:38,563 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4521, 3.4733, 3.9757, 2.7719, 2.6082, 2.7400, 4.3457, 4.4527], device='cuda:1'), covar=tensor([0.1157, 0.1526, 0.0400, 0.1754, 0.1687, 0.1593, 0.0295, 0.0219], device='cuda:1'), in_proj_covar=tensor([0.0244, 0.0261, 0.0191, 0.0220, 0.0216, 0.0181, 0.0204, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-17 00:33:41,691 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9641, 3.8603, 3.0926, 3.7194, 3.1348, 2.2343, 3.8643, 2.1925], device='cuda:1'), covar=tensor([0.0679, 0.0627, 0.0506, 0.0294, 0.0681, 0.2006, 0.0942, 0.1441], device='cuda:1'), in_proj_covar=tensor([0.0140, 0.0137, 0.0133, 0.0113, 0.0147, 0.0187, 0.0168, 0.0141], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 00:33:47,101 INFO [train.py:893] (1/4) Epoch 19, batch 1400, loss[loss=0.174, simple_loss=0.2245, pruned_loss=0.06173, over 13189.00 frames. ], tot_loss[loss=0.1849, simple_loss=0.2431, pruned_loss=0.06338, over 2659493.82 frames. ], batch size: 58, lr: 6.89e-03, grad_scale: 32.0 2023-04-17 00:34:21,725 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.018e+02 2.575e+02 3.265e+02 4.093e+02 1.236e+03, threshold=6.530e+02, percent-clipped=2.0 2023-04-17 00:34:36,620 INFO [train.py:893] (1/4) Epoch 19, batch 1450, loss[loss=0.1905, simple_loss=0.246, pruned_loss=0.06753, over 13364.00 frames. ], tot_loss[loss=0.1851, simple_loss=0.2432, pruned_loss=0.06352, over 2660665.94 frames. ], batch size: 62, lr: 6.89e-03, grad_scale: 32.0 2023-04-17 00:34:52,551 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5228, 4.2991, 4.5276, 4.4812, 4.7830, 4.3122, 4.7717, 4.7508], device='cuda:1'), covar=tensor([0.0416, 0.0599, 0.0657, 0.0529, 0.0590, 0.0856, 0.0516, 0.0479], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0286, 0.0289, 0.0215, 0.0413, 0.0330, 0.0266, 0.0286], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 00:35:19,170 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50078.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:35:23,959 INFO [train.py:893] (1/4) Epoch 19, batch 1500, loss[loss=0.2064, simple_loss=0.2603, pruned_loss=0.07622, over 13532.00 frames. ], tot_loss[loss=0.1853, simple_loss=0.2433, pruned_loss=0.0636, over 2661279.77 frames. ], batch size: 76, lr: 6.88e-03, grad_scale: 32.0 2023-04-17 00:35:25,039 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9068, 3.7186, 3.8283, 2.1362, 4.0425, 3.9594, 3.9294, 4.0484], device='cuda:1'), covar=tensor([0.0212, 0.0132, 0.0128, 0.1252, 0.0137, 0.0189, 0.0121, 0.0102], device='cuda:1'), in_proj_covar=tensor([0.0087, 0.0051, 0.0077, 0.0098, 0.0094, 0.0101, 0.0075, 0.0070], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 00:35:27,486 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50088.0, num_to_drop=1, layers_to_drop={2} 2023-04-17 00:35:27,638 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-04-17 00:35:54,508 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.061e+02 2.681e+02 3.044e+02 3.631e+02 7.393e+02, threshold=6.088e+02, percent-clipped=1.0 2023-04-17 00:35:58,873 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50122.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:36:02,806 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50126.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:36:04,668 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50128.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:36:11,246 INFO [train.py:893] (1/4) Epoch 19, batch 1550, loss[loss=0.1833, simple_loss=0.2435, pruned_loss=0.06161, over 13235.00 frames. ], tot_loss[loss=0.1849, simple_loss=0.2433, pruned_loss=0.0633, over 2662108.04 frames. ], batch size: 124, lr: 6.88e-03, grad_scale: 32.0 2023-04-17 00:36:13,793 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50136.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 00:36:55,678 INFO [train.py:893] (1/4) Epoch 19, batch 1600, loss[loss=0.2074, simple_loss=0.2688, pruned_loss=0.07299, over 13470.00 frames. ], tot_loss[loss=0.1855, simple_loss=0.2437, pruned_loss=0.06366, over 2658273.14 frames. ], batch size: 100, lr: 6.88e-03, grad_scale: 32.0 2023-04-17 00:37:00,836 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50189.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:37:05,304 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-04-17 00:37:09,819 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50197.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 00:37:26,428 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.888e+02 2.851e+02 3.229e+02 3.845e+02 6.840e+02, threshold=6.459e+02, percent-clipped=1.0 2023-04-17 00:37:43,051 INFO [train.py:893] (1/4) Epoch 19, batch 1650, loss[loss=0.1837, simple_loss=0.2451, pruned_loss=0.06111, over 13398.00 frames. ], tot_loss[loss=0.1857, simple_loss=0.2446, pruned_loss=0.06341, over 2659029.00 frames. ], batch size: 88, lr: 6.87e-03, grad_scale: 32.0 2023-04-17 00:38:04,826 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50258.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 00:38:28,022 INFO [train.py:893] (1/4) Epoch 19, batch 1700, loss[loss=0.1745, simple_loss=0.2317, pruned_loss=0.05865, over 13517.00 frames. ], tot_loss[loss=0.1852, simple_loss=0.2446, pruned_loss=0.06287, over 2662961.22 frames. ], batch size: 70, lr: 6.87e-03, grad_scale: 32.0 2023-04-17 00:38:57,982 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.960e+02 2.607e+02 3.098e+02 3.565e+02 8.055e+02, threshold=6.196e+02, percent-clipped=3.0 2023-04-17 00:39:14,610 INFO [train.py:893] (1/4) Epoch 19, batch 1750, loss[loss=0.2032, simple_loss=0.2522, pruned_loss=0.07711, over 13559.00 frames. ], tot_loss[loss=0.184, simple_loss=0.2432, pruned_loss=0.06243, over 2663258.72 frames. ], batch size: 89, lr: 6.87e-03, grad_scale: 32.0 2023-04-17 00:40:01,627 INFO [train.py:893] (1/4) Epoch 19, batch 1800, loss[loss=0.1593, simple_loss=0.2172, pruned_loss=0.05071, over 13216.00 frames. ], tot_loss[loss=0.1829, simple_loss=0.2419, pruned_loss=0.06191, over 2664193.69 frames. ], batch size: 58, lr: 6.86e-03, grad_scale: 32.0 2023-04-17 00:40:13,765 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1663, 2.7544, 2.6675, 3.1062, 2.5430, 3.1711, 3.1666, 2.6984], device='cuda:1'), covar=tensor([0.0069, 0.0153, 0.0147, 0.0138, 0.0175, 0.0119, 0.0159, 0.0157], device='cuda:1'), in_proj_covar=tensor([0.0099, 0.0107, 0.0116, 0.0113, 0.0124, 0.0102, 0.0102, 0.0100], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 00:40:33,679 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.914e+02 2.700e+02 3.166e+02 3.667e+02 8.040e+02, threshold=6.333e+02, percent-clipped=2.0 2023-04-17 00:40:38,155 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50422.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:40:39,898 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0303, 2.6325, 2.5559, 2.9967, 2.3629, 3.1400, 3.0375, 2.5945], device='cuda:1'), covar=tensor([0.0078, 0.0160, 0.0162, 0.0151, 0.0209, 0.0111, 0.0162, 0.0184], device='cuda:1'), in_proj_covar=tensor([0.0099, 0.0107, 0.0117, 0.0114, 0.0125, 0.0103, 0.0102, 0.0101], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 00:40:48,087 INFO [train.py:893] (1/4) Epoch 19, batch 1850, loss[loss=0.1691, simple_loss=0.2318, pruned_loss=0.05323, over 13474.00 frames. ], tot_loss[loss=0.1826, simple_loss=0.2418, pruned_loss=0.06167, over 2664822.33 frames. ], batch size: 79, lr: 6.86e-03, grad_scale: 32.0 2023-04-17 00:40:52,313 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-17 00:40:52,498 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.2990, 5.1805, 5.3285, 5.0408, 5.6130, 5.1164, 5.5777, 5.5681], device='cuda:1'), covar=tensor([0.0388, 0.0453, 0.0600, 0.0610, 0.0553, 0.0814, 0.0600, 0.0553], device='cuda:1'), in_proj_covar=tensor([0.0273, 0.0285, 0.0291, 0.0217, 0.0414, 0.0331, 0.0267, 0.0287], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 00:41:22,270 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50470.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:41:36,348 INFO [train.py:893] (1/4) Epoch 19, batch 1900, loss[loss=0.2103, simple_loss=0.2689, pruned_loss=0.07584, over 13367.00 frames. ], tot_loss[loss=0.1822, simple_loss=0.2412, pruned_loss=0.06164, over 2664588.63 frames. ], batch size: 113, lr: 6.86e-03, grad_scale: 32.0 2023-04-17 00:41:36,525 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50484.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:42:05,950 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.964e+02 2.672e+02 3.086e+02 3.593e+02 6.893e+02, threshold=6.171e+02, percent-clipped=1.0 2023-04-17 00:42:20,678 INFO [train.py:893] (1/4) Epoch 19, batch 1950, loss[loss=0.1791, simple_loss=0.2341, pruned_loss=0.06199, over 13558.00 frames. ], tot_loss[loss=0.1821, simple_loss=0.2407, pruned_loss=0.06179, over 2658798.20 frames. ], batch size: 76, lr: 6.85e-03, grad_scale: 32.0 2023-04-17 00:42:32,867 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50546.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:42:33,795 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5133, 3.9384, 3.6401, 4.2218, 2.3723, 3.3180, 3.9487, 2.2929], device='cuda:1'), covar=tensor([0.0155, 0.0429, 0.0758, 0.0601, 0.1453, 0.0809, 0.0592, 0.1742], device='cuda:1'), in_proj_covar=tensor([0.0123, 0.0181, 0.0201, 0.0234, 0.0182, 0.0196, 0.0175, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 00:42:38,660 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50553.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 00:42:58,932 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2023-04-17 00:43:03,594 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1809, 2.7177, 2.7115, 3.1616, 2.5599, 3.2731, 3.2430, 2.6820], device='cuda:1'), covar=tensor([0.0073, 0.0157, 0.0129, 0.0165, 0.0190, 0.0102, 0.0125, 0.0129], device='cuda:1'), in_proj_covar=tensor([0.0101, 0.0109, 0.0119, 0.0115, 0.0126, 0.0104, 0.0103, 0.0103], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 00:43:07,595 INFO [train.py:893] (1/4) Epoch 19, batch 2000, loss[loss=0.1951, simple_loss=0.2556, pruned_loss=0.06728, over 13456.00 frames. ], tot_loss[loss=0.1845, simple_loss=0.2433, pruned_loss=0.06288, over 2660638.93 frames. ], batch size: 103, lr: 6.85e-03, grad_scale: 32.0 2023-04-17 00:43:12,646 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-17 00:43:30,397 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50607.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:43:38,610 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.951e+02 2.994e+02 3.373e+02 4.024e+02 7.060e+02, threshold=6.745e+02, percent-clipped=2.0 2023-04-17 00:43:51,853 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.89 vs. limit=2.0 2023-04-17 00:43:56,142 INFO [train.py:893] (1/4) Epoch 19, batch 2050, loss[loss=0.2262, simple_loss=0.2782, pruned_loss=0.08712, over 13253.00 frames. ], tot_loss[loss=0.1844, simple_loss=0.2436, pruned_loss=0.06266, over 2663812.10 frames. ], batch size: 124, lr: 6.85e-03, grad_scale: 32.0 2023-04-17 00:44:41,408 INFO [train.py:893] (1/4) Epoch 19, batch 2100, loss[loss=0.2063, simple_loss=0.2599, pruned_loss=0.07637, over 13490.00 frames. ], tot_loss[loss=0.1837, simple_loss=0.2427, pruned_loss=0.06234, over 2662889.50 frames. ], batch size: 93, lr: 6.84e-03, grad_scale: 32.0 2023-04-17 00:44:56,928 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-04-17 00:44:59,077 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0597, 2.6479, 2.6744, 3.0751, 2.5005, 3.1541, 3.0574, 2.5969], device='cuda:1'), covar=tensor([0.0108, 0.0184, 0.0134, 0.0152, 0.0204, 0.0112, 0.0167, 0.0173], device='cuda:1'), in_proj_covar=tensor([0.0101, 0.0109, 0.0119, 0.0115, 0.0127, 0.0104, 0.0103, 0.0103], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 00:45:12,559 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.790e+02 2.504e+02 2.896e+02 3.497e+02 6.589e+02, threshold=5.792e+02, percent-clipped=0.0 2023-04-17 00:45:21,526 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.54 vs. limit=2.0 2023-04-17 00:45:26,498 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9784, 2.0966, 2.3393, 3.3001, 2.9638, 3.2949, 2.5882, 2.1367], device='cuda:1'), covar=tensor([0.0275, 0.0795, 0.0647, 0.0079, 0.0295, 0.0078, 0.0578, 0.0788], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0152, 0.0166, 0.0094, 0.0117, 0.0094, 0.0168, 0.0149], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 00:45:29,436 INFO [train.py:893] (1/4) Epoch 19, batch 2150, loss[loss=0.2255, simple_loss=0.2802, pruned_loss=0.08539, over 13222.00 frames. ], tot_loss[loss=0.1834, simple_loss=0.2428, pruned_loss=0.06206, over 2660005.32 frames. ], batch size: 132, lr: 6.84e-03, grad_scale: 32.0 2023-04-17 00:46:13,443 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50782.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:46:14,789 INFO [train.py:893] (1/4) Epoch 19, batch 2200, loss[loss=0.1772, simple_loss=0.2367, pruned_loss=0.05881, over 13041.00 frames. ], tot_loss[loss=0.1836, simple_loss=0.2429, pruned_loss=0.06212, over 2658765.50 frames. ], batch size: 142, lr: 6.84e-03, grad_scale: 16.0 2023-04-17 00:46:15,013 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50784.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:46:21,789 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.90 vs. limit=2.0 2023-04-17 00:46:37,778 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50808.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 00:46:46,611 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.993e+02 2.746e+02 3.205e+02 4.052e+02 6.409e+02, threshold=6.411e+02, percent-clipped=1.0 2023-04-17 00:47:00,758 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50832.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:47:02,337 INFO [train.py:893] (1/4) Epoch 19, batch 2250, loss[loss=0.1567, simple_loss=0.2138, pruned_loss=0.04979, over 13428.00 frames. ], tot_loss[loss=0.1818, simple_loss=0.2409, pruned_loss=0.06131, over 2662823.90 frames. ], batch size: 65, lr: 6.83e-03, grad_scale: 16.0 2023-04-17 00:47:10,780 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50843.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:47:19,637 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=50853.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 00:47:33,936 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=50869.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 00:47:47,567 INFO [train.py:893] (1/4) Epoch 19, batch 2300, loss[loss=0.168, simple_loss=0.2249, pruned_loss=0.05554, over 13424.00 frames. ], tot_loss[loss=0.1814, simple_loss=0.2405, pruned_loss=0.0612, over 2662213.54 frames. ], batch size: 65, lr: 6.83e-03, grad_scale: 16.0 2023-04-17 00:48:03,644 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=50901.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 00:48:04,486 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=50902.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:48:20,400 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.979e+02 2.847e+02 3.302e+02 3.831e+02 7.686e+02, threshold=6.603e+02, percent-clipped=1.0 2023-04-17 00:48:33,726 INFO [train.py:893] (1/4) Epoch 19, batch 2350, loss[loss=0.1951, simple_loss=0.2504, pruned_loss=0.06986, over 13271.00 frames. ], tot_loss[loss=0.1801, simple_loss=0.2393, pruned_loss=0.06044, over 2660901.38 frames. ], batch size: 124, lr: 6.83e-03, grad_scale: 16.0 2023-04-17 00:48:43,156 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9458, 3.7510, 3.8287, 2.3303, 4.1223, 3.9618, 3.9265, 4.0236], device='cuda:1'), covar=tensor([0.0220, 0.0132, 0.0146, 0.1150, 0.0139, 0.0232, 0.0129, 0.0116], device='cuda:1'), in_proj_covar=tensor([0.0089, 0.0053, 0.0079, 0.0100, 0.0095, 0.0103, 0.0077, 0.0072], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0002], device='cuda:1') 2023-04-17 00:48:45,581 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=50946.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:48:57,031 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-17 00:48:59,156 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-04-17 00:49:14,733 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.8636, 4.7426, 4.9280, 4.7990, 5.1923, 4.7041, 5.2140, 5.1345], device='cuda:1'), covar=tensor([0.0388, 0.0541, 0.0710, 0.0635, 0.0513, 0.0900, 0.0405, 0.0510], device='cuda:1'), in_proj_covar=tensor([0.0274, 0.0287, 0.0291, 0.0217, 0.0414, 0.0330, 0.0267, 0.0287], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 00:49:21,052 INFO [train.py:893] (1/4) Epoch 19, batch 2400, loss[loss=0.1519, simple_loss=0.2174, pruned_loss=0.04316, over 13469.00 frames. ], tot_loss[loss=0.18, simple_loss=0.2389, pruned_loss=0.06056, over 2663480.24 frames. ], batch size: 79, lr: 6.82e-03, grad_scale: 16.0 2023-04-17 00:49:41,829 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51007.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:49:53,372 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.675e+02 2.579e+02 3.050e+02 3.500e+02 7.370e+02, threshold=6.100e+02, percent-clipped=1.0 2023-04-17 00:50:07,320 INFO [train.py:893] (1/4) Epoch 19, batch 2450, loss[loss=0.1742, simple_loss=0.2312, pruned_loss=0.05859, over 13367.00 frames. ], tot_loss[loss=0.1803, simple_loss=0.2392, pruned_loss=0.0607, over 2664806.54 frames. ], batch size: 62, lr: 6.82e-03, grad_scale: 16.0 2023-04-17 00:50:31,585 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0213, 2.7776, 2.4207, 1.9957, 1.9737, 2.5076, 2.5706, 3.0512], device='cuda:1'), covar=tensor([0.0971, 0.0366, 0.0676, 0.1500, 0.0424, 0.0634, 0.0784, 0.0237], device='cuda:1'), in_proj_covar=tensor([0.0158, 0.0134, 0.0116, 0.0203, 0.0107, 0.0152, 0.0164, 0.0124], device='cuda:1'), out_proj_covar=tensor([1.1760e-04, 1.0035e-04, 9.0915e-05, 1.5179e-04, 7.8665e-05, 1.1469e-04, 1.2440e-04, 9.2005e-05], device='cuda:1') 2023-04-17 00:50:33,228 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51062.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:50:54,325 INFO [train.py:893] (1/4) Epoch 19, batch 2500, loss[loss=0.1703, simple_loss=0.2327, pruned_loss=0.05391, over 13540.00 frames. ], tot_loss[loss=0.18, simple_loss=0.2392, pruned_loss=0.06039, over 2667294.16 frames. ], batch size: 72, lr: 6.82e-03, grad_scale: 16.0 2023-04-17 00:51:25,492 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.099e+02 2.551e+02 2.924e+02 3.342e+02 5.751e+02, threshold=5.847e+02, percent-clipped=0.0 2023-04-17 00:51:30,038 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51123.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:51:39,536 INFO [train.py:893] (1/4) Epoch 19, batch 2550, loss[loss=0.173, simple_loss=0.2258, pruned_loss=0.06009, over 13472.00 frames. ], tot_loss[loss=0.1813, simple_loss=0.2402, pruned_loss=0.06118, over 2662561.69 frames. ], batch size: 65, lr: 6.81e-03, grad_scale: 16.0 2023-04-17 00:51:43,812 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51138.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:52:03,731 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-17 00:52:08,001 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51164.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 00:52:17,286 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51172.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 00:52:27,012 INFO [train.py:893] (1/4) Epoch 19, batch 2600, loss[loss=0.1686, simple_loss=0.2253, pruned_loss=0.05591, over 13428.00 frames. ], tot_loss[loss=0.182, simple_loss=0.2406, pruned_loss=0.06167, over 2659386.86 frames. ], batch size: 65, lr: 6.81e-03, grad_scale: 16.0 2023-04-17 00:52:28,396 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.05 vs. limit=5.0 2023-04-17 00:52:33,187 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-04-17 00:52:45,114 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51202.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:52:52,693 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.5471, 2.1028, 2.2080, 2.5885, 1.9493, 2.6286, 2.4869, 2.0321], device='cuda:1'), covar=tensor([0.0078, 0.0237, 0.0161, 0.0143, 0.0235, 0.0115, 0.0177, 0.0193], device='cuda:1'), in_proj_covar=tensor([0.0102, 0.0110, 0.0120, 0.0116, 0.0128, 0.0105, 0.0105, 0.0103], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 00:52:57,644 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.905e+02 2.632e+02 3.230e+02 4.137e+02 6.424e+02, threshold=6.459e+02, percent-clipped=3.0 2023-04-17 00:53:01,208 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.52 vs. limit=2.0 2023-04-17 00:53:09,484 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51233.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 00:53:09,975 INFO [train.py:893] (1/4) Epoch 19, batch 2650, loss[loss=0.1825, simple_loss=0.2444, pruned_loss=0.06032, over 13360.00 frames. ], tot_loss[loss=0.1823, simple_loss=0.2406, pruned_loss=0.06199, over 2660421.71 frames. ], batch size: 67, lr: 6.81e-03, grad_scale: 16.0 2023-04-17 00:53:21,989 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51250.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:53:33,495 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.5335, 2.2942, 2.2783, 2.6377, 2.1539, 2.6377, 2.6980, 2.1320], device='cuda:1'), covar=tensor([0.0088, 0.0192, 0.0151, 0.0167, 0.0203, 0.0153, 0.0158, 0.0173], device='cuda:1'), in_proj_covar=tensor([0.0101, 0.0109, 0.0120, 0.0115, 0.0127, 0.0105, 0.0104, 0.0102], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 00:54:07,531 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-17 00:54:18,257 INFO [train.py:893] (1/4) Epoch 20, batch 0, loss[loss=0.1917, simple_loss=0.2559, pruned_loss=0.06372, over 13048.00 frames. ], tot_loss[loss=0.1917, simple_loss=0.2559, pruned_loss=0.06372, over 13048.00 frames. ], batch size: 142, lr: 6.63e-03, grad_scale: 16.0 2023-04-17 00:54:18,258 INFO [train.py:918] (1/4) Computing validation loss 2023-04-17 00:54:26,768 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.3947, 4.9148, 4.9128, 4.9727, 4.8573, 4.9406, 5.3882, 4.9983], device='cuda:1'), covar=tensor([0.0665, 0.1198, 0.1986, 0.2074, 0.0817, 0.1378, 0.0840, 0.1044], device='cuda:1'), in_proj_covar=tensor([0.0317, 0.0360, 0.0449, 0.0460, 0.0280, 0.0337, 0.0417, 0.0329], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-17 00:54:29,526 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6112, 2.8989, 3.1510, 4.2778, 3.8137, 4.3092, 3.5327, 2.8259], device='cuda:1'), covar=tensor([0.0310, 0.0782, 0.0627, 0.0063, 0.0221, 0.0051, 0.0534, 0.0776], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0151, 0.0167, 0.0095, 0.0117, 0.0094, 0.0167, 0.0149], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 00:54:33,518 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5785, 3.1351, 3.1573, 3.4654, 3.2856, 3.5990, 3.2165, 3.2536], device='cuda:1'), covar=tensor([0.0052, 0.0142, 0.0121, 0.0172, 0.0116, 0.0074, 0.0158, 0.0085], device='cuda:1'), in_proj_covar=tensor([0.0101, 0.0109, 0.0119, 0.0115, 0.0126, 0.0104, 0.0103, 0.0102], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 00:54:41,315 INFO [train.py:927] (1/4) Epoch 20, validation: loss=0.1383, simple_loss=0.1976, pruned_loss=0.03948, over 2446609.00 frames. 2023-04-17 00:54:41,316 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12831MB 2023-04-17 00:54:47,459 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1907, 4.7094, 4.4860, 4.4894, 4.5222, 4.3863, 4.7348, 4.7911], device='cuda:1'), covar=tensor([0.0261, 0.0199, 0.0220, 0.0353, 0.0250, 0.0264, 0.0275, 0.0201], device='cuda:1'), in_proj_covar=tensor([0.0137, 0.0196, 0.0159, 0.0178, 0.0146, 0.0195, 0.0130, 0.0173], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 00:54:50,728 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4421, 4.2261, 4.4203, 4.4177, 4.6615, 4.2810, 4.6394, 4.6094], device='cuda:1'), covar=tensor([0.0406, 0.0540, 0.0606, 0.0522, 0.0541, 0.0765, 0.0421, 0.0475], device='cuda:1'), in_proj_covar=tensor([0.0277, 0.0287, 0.0293, 0.0220, 0.0420, 0.0334, 0.0271, 0.0289], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 00:54:58,938 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51302.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:55:13,676 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.680e+02 2.582e+02 3.103e+02 3.598e+02 6.188e+02, threshold=6.206e+02, percent-clipped=0.0 2023-04-17 00:55:17,768 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.53 vs. limit=5.0 2023-04-17 00:55:27,791 INFO [train.py:893] (1/4) Epoch 20, batch 50, loss[loss=0.1534, simple_loss=0.2115, pruned_loss=0.04762, over 13485.00 frames. ], tot_loss[loss=0.1786, simple_loss=0.2347, pruned_loss=0.06126, over 601913.59 frames. ], batch size: 70, lr: 6.63e-03, grad_scale: 16.0 2023-04-17 00:55:52,741 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-17 00:55:52,741 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-17 00:55:52,741 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-17 00:55:52,749 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-17 00:55:52,768 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-17 00:55:52,784 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-17 00:55:52,800 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-17 00:56:12,663 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.4665, 3.7738, 2.5637, 3.2674, 2.9896, 1.9227, 3.7865, 1.9145], device='cuda:1'), covar=tensor([0.0899, 0.0356, 0.0743, 0.0506, 0.0699, 0.2540, 0.0858, 0.1653], device='cuda:1'), in_proj_covar=tensor([0.0140, 0.0136, 0.0133, 0.0114, 0.0147, 0.0188, 0.0168, 0.0140], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 00:56:13,441 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51382.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:56:14,089 INFO [train.py:893] (1/4) Epoch 20, batch 100, loss[loss=0.1949, simple_loss=0.2464, pruned_loss=0.07165, over 13081.00 frames. ], tot_loss[loss=0.184, simple_loss=0.2394, pruned_loss=0.06426, over 1063192.63 frames. ], batch size: 142, lr: 6.62e-03, grad_scale: 16.0 2023-04-17 00:56:14,438 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.1637, 2.1857, 4.0415, 3.8083, 3.9490, 3.0432, 3.6449, 3.0123], device='cuda:1'), covar=tensor([0.2086, 0.1441, 0.0117, 0.0211, 0.0218, 0.0727, 0.0250, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0217, 0.0185, 0.0119, 0.0125, 0.0131, 0.0174, 0.0140, 0.0181], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-17 00:56:23,441 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51392.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:56:47,333 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.741e+02 2.853e+02 3.337e+02 4.244e+02 6.093e+02, threshold=6.673e+02, percent-clipped=0.0 2023-04-17 00:56:47,552 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51418.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:57:00,424 INFO [train.py:893] (1/4) Epoch 20, batch 150, loss[loss=0.1737, simple_loss=0.2315, pruned_loss=0.05793, over 13536.00 frames. ], tot_loss[loss=0.1865, simple_loss=0.2413, pruned_loss=0.06584, over 1417676.52 frames. ], batch size: 76, lr: 6.62e-03, grad_scale: 16.0 2023-04-17 00:57:05,567 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51438.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:57:10,539 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51443.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:57:20,422 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51453.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:57:29,310 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51464.0, num_to_drop=1, layers_to_drop={2} 2023-04-17 00:57:47,194 INFO [train.py:893] (1/4) Epoch 20, batch 200, loss[loss=0.2064, simple_loss=0.2608, pruned_loss=0.07594, over 13395.00 frames. ], tot_loss[loss=0.1877, simple_loss=0.2429, pruned_loss=0.06626, over 1679820.26 frames. ], batch size: 113, lr: 6.62e-03, grad_scale: 16.0 2023-04-17 00:57:49,741 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51486.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:57:58,385 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51495.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:58:14,008 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51512.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 00:58:20,424 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.908e+02 2.666e+02 3.383e+02 3.892e+02 7.780e+02, threshold=6.766e+02, percent-clipped=1.0 2023-04-17 00:58:28,939 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51528.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 00:58:32,952 INFO [train.py:893] (1/4) Epoch 20, batch 250, loss[loss=0.1767, simple_loss=0.2401, pruned_loss=0.05662, over 13539.00 frames. ], tot_loss[loss=0.1881, simple_loss=0.2433, pruned_loss=0.0664, over 1880806.88 frames. ], batch size: 83, lr: 6.61e-03, grad_scale: 16.0 2023-04-17 00:58:55,550 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51556.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:59:20,632 INFO [train.py:893] (1/4) Epoch 20, batch 300, loss[loss=0.1972, simple_loss=0.2579, pruned_loss=0.06828, over 13559.00 frames. ], tot_loss[loss=0.187, simple_loss=0.2433, pruned_loss=0.06533, over 2051391.87 frames. ], batch size: 89, lr: 6.61e-03, grad_scale: 16.0 2023-04-17 00:59:37,788 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51602.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 00:59:53,135 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.976e+02 2.888e+02 3.248e+02 3.960e+02 8.740e+02, threshold=6.497e+02, percent-clipped=3.0 2023-04-17 01:00:06,405 INFO [train.py:893] (1/4) Epoch 20, batch 350, loss[loss=0.1985, simple_loss=0.2627, pruned_loss=0.06719, over 13492.00 frames. ], tot_loss[loss=0.1867, simple_loss=0.2431, pruned_loss=0.06517, over 2173408.19 frames. ], batch size: 93, lr: 6.61e-03, grad_scale: 16.0 2023-04-17 01:00:23,159 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51650.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:00:54,200 INFO [train.py:893] (1/4) Epoch 20, batch 400, loss[loss=0.1746, simple_loss=0.2369, pruned_loss=0.05612, over 13485.00 frames. ], tot_loss[loss=0.1857, simple_loss=0.2426, pruned_loss=0.06441, over 2279019.10 frames. ], batch size: 81, lr: 6.60e-03, grad_scale: 16.0 2023-04-17 01:01:26,015 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.927e+02 2.630e+02 3.030e+02 3.596e+02 5.249e+02, threshold=6.061e+02, percent-clipped=0.0 2023-04-17 01:01:26,270 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51718.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:01:27,465 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-04-17 01:01:31,967 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0712, 2.6914, 2.6737, 3.0576, 2.6243, 3.0663, 3.0889, 2.6145], device='cuda:1'), covar=tensor([0.0069, 0.0152, 0.0130, 0.0129, 0.0161, 0.0111, 0.0148, 0.0126], device='cuda:1'), in_proj_covar=tensor([0.0101, 0.0109, 0.0118, 0.0116, 0.0126, 0.0105, 0.0104, 0.0103], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 01:01:39,787 INFO [train.py:893] (1/4) Epoch 20, batch 450, loss[loss=0.165, simple_loss=0.2192, pruned_loss=0.05535, over 13170.00 frames. ], tot_loss[loss=0.1864, simple_loss=0.2437, pruned_loss=0.06456, over 2361462.70 frames. ], batch size: 58, lr: 6.60e-03, grad_scale: 16.0 2023-04-17 01:01:43,964 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51738.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:01:52,733 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=51747.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:01:53,457 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51748.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:02:05,664 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-17 01:02:09,904 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51766.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:02:25,778 INFO [train.py:893] (1/4) Epoch 20, batch 500, loss[loss=0.1712, simple_loss=0.2317, pruned_loss=0.05531, over 13476.00 frames. ], tot_loss[loss=0.187, simple_loss=0.2445, pruned_loss=0.06471, over 2428592.93 frames. ], batch size: 79, lr: 6.60e-03, grad_scale: 16.0 2023-04-17 01:02:49,867 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=51808.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:02:58,571 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.164e+02 2.958e+02 3.503e+02 4.173e+02 8.375e+02, threshold=7.006e+02, percent-clipped=2.0 2023-04-17 01:03:08,033 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=51828.0, num_to_drop=1, layers_to_drop={2} 2023-04-17 01:03:12,423 INFO [train.py:893] (1/4) Epoch 20, batch 550, loss[loss=0.2167, simple_loss=0.2604, pruned_loss=0.08652, over 11806.00 frames. ], tot_loss[loss=0.1863, simple_loss=0.2441, pruned_loss=0.06425, over 2475868.23 frames. ], batch size: 157, lr: 6.59e-03, grad_scale: 16.0 2023-04-17 01:03:29,450 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=51851.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:03:52,751 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=51876.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 01:03:59,060 INFO [train.py:893] (1/4) Epoch 20, batch 600, loss[loss=0.1667, simple_loss=0.2288, pruned_loss=0.05229, over 13496.00 frames. ], tot_loss[loss=0.1842, simple_loss=0.2421, pruned_loss=0.06316, over 2517158.59 frames. ], batch size: 81, lr: 6.59e-03, grad_scale: 16.0 2023-04-17 01:04:09,380 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0870, 2.1342, 4.0719, 3.9053, 3.9690, 3.1481, 3.6311, 2.9047], device='cuda:1'), covar=tensor([0.2098, 0.1409, 0.0108, 0.0197, 0.0175, 0.0598, 0.0251, 0.0819], device='cuda:1'), in_proj_covar=tensor([0.0211, 0.0181, 0.0117, 0.0122, 0.0127, 0.0169, 0.0138, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-17 01:04:10,209 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5747, 3.9333, 3.4931, 4.5086, 2.2195, 2.7415, 3.9887, 2.2571], device='cuda:1'), covar=tensor([0.0142, 0.0496, 0.0886, 0.0453, 0.1954, 0.1389, 0.0641, 0.2186], device='cuda:1'), in_proj_covar=tensor([0.0123, 0.0181, 0.0200, 0.0237, 0.0182, 0.0195, 0.0175, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 01:04:19,184 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1696, 3.9962, 4.0789, 2.5995, 4.5151, 4.2417, 4.2408, 4.4448], device='cuda:1'), covar=tensor([0.0226, 0.0139, 0.0141, 0.0986, 0.0129, 0.0219, 0.0135, 0.0101], device='cuda:1'), in_proj_covar=tensor([0.0090, 0.0054, 0.0080, 0.0101, 0.0097, 0.0106, 0.0078, 0.0073], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 01:04:22,432 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4137, 2.2378, 2.6553, 3.8997, 3.4752, 3.9610, 3.0833, 2.2573], device='cuda:1'), covar=tensor([0.0278, 0.0953, 0.0742, 0.0051, 0.0242, 0.0047, 0.0624, 0.0941], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0150, 0.0165, 0.0094, 0.0116, 0.0093, 0.0165, 0.0146], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 01:04:31,320 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.718e+02 2.517e+02 2.993e+02 3.459e+02 6.026e+02, threshold=5.986e+02, percent-clipped=0.0 2023-04-17 01:04:37,683 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.59 vs. limit=2.0 2023-04-17 01:04:46,681 INFO [train.py:893] (1/4) Epoch 20, batch 650, loss[loss=0.1868, simple_loss=0.2236, pruned_loss=0.07497, over 8708.00 frames. ], tot_loss[loss=0.1831, simple_loss=0.2411, pruned_loss=0.0625, over 2545621.48 frames. ], batch size: 35, lr: 6.59e-03, grad_scale: 16.0 2023-04-17 01:04:55,472 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-04-17 01:05:16,722 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.7431, 4.5796, 4.8229, 4.7395, 5.0898, 4.6350, 5.0775, 5.0438], device='cuda:1'), covar=tensor([0.0504, 0.0581, 0.0599, 0.0536, 0.0525, 0.0969, 0.0439, 0.0498], device='cuda:1'), in_proj_covar=tensor([0.0278, 0.0286, 0.0293, 0.0217, 0.0418, 0.0335, 0.0268, 0.0291], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 01:05:32,337 INFO [train.py:893] (1/4) Epoch 20, batch 700, loss[loss=0.1813, simple_loss=0.224, pruned_loss=0.06929, over 13347.00 frames. ], tot_loss[loss=0.1816, simple_loss=0.24, pruned_loss=0.06157, over 2572293.87 frames. ], batch size: 62, lr: 6.59e-03, grad_scale: 16.0 2023-04-17 01:06:10,148 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.613e+02 2.564e+02 3.049e+02 3.631e+02 6.863e+02, threshold=6.097e+02, percent-clipped=2.0 2023-04-17 01:06:24,285 INFO [train.py:893] (1/4) Epoch 20, batch 750, loss[loss=0.1706, simple_loss=0.2363, pruned_loss=0.05251, over 13451.00 frames. ], tot_loss[loss=0.1819, simple_loss=0.24, pruned_loss=0.06188, over 2590630.94 frames. ], batch size: 79, lr: 6.58e-03, grad_scale: 16.0 2023-04-17 01:06:28,653 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52038.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:06:31,077 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52040.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:06:38,305 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52048.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:06:50,151 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-04-17 01:06:59,870 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-04-17 01:07:10,863 INFO [train.py:893] (1/4) Epoch 20, batch 800, loss[loss=0.2026, simple_loss=0.2587, pruned_loss=0.07327, over 13512.00 frames. ], tot_loss[loss=0.1834, simple_loss=0.2415, pruned_loss=0.06266, over 2606587.08 frames. ], batch size: 91, lr: 6.58e-03, grad_scale: 16.0 2023-04-17 01:07:13,477 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52086.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:07:22,597 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52096.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:07:27,021 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52101.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:07:29,234 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52103.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:07:42,863 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.162e+02 2.907e+02 3.374e+02 4.122e+02 8.456e+02, threshold=6.748e+02, percent-clipped=6.0 2023-04-17 01:07:57,076 INFO [train.py:893] (1/4) Epoch 20, batch 850, loss[loss=0.171, simple_loss=0.2359, pruned_loss=0.05303, over 13469.00 frames. ], tot_loss[loss=0.1838, simple_loss=0.2423, pruned_loss=0.06264, over 2615520.76 frames. ], batch size: 79, lr: 6.58e-03, grad_scale: 16.0 2023-04-17 01:08:09,223 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.19 vs. limit=5.0 2023-04-17 01:08:14,677 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52151.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:08:28,025 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.83 vs. limit=5.0 2023-04-17 01:08:32,530 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0520, 2.3680, 2.0395, 3.8724, 4.4555, 3.3078, 4.3892, 4.1739], device='cuda:1'), covar=tensor([0.0123, 0.1037, 0.1067, 0.0117, 0.0070, 0.0497, 0.0095, 0.0099], device='cuda:1'), in_proj_covar=tensor([0.0075, 0.0090, 0.0097, 0.0081, 0.0066, 0.0081, 0.0055, 0.0077], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 01:08:43,837 INFO [train.py:893] (1/4) Epoch 20, batch 900, loss[loss=0.1764, simple_loss=0.2371, pruned_loss=0.05782, over 13537.00 frames. ], tot_loss[loss=0.1833, simple_loss=0.2418, pruned_loss=0.06237, over 2626720.21 frames. ], batch size: 83, lr: 6.57e-03, grad_scale: 16.0 2023-04-17 01:08:52,396 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52193.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:08:57,980 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52199.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:09:15,517 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-17 01:09:15,757 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2115, 4.5438, 4.2539, 4.3216, 4.3493, 4.7045, 4.4810, 4.2900], device='cuda:1'), covar=tensor([0.0255, 0.0251, 0.0318, 0.0822, 0.0226, 0.0189, 0.0266, 0.0311], device='cuda:1'), in_proj_covar=tensor([0.0157, 0.0152, 0.0172, 0.0261, 0.0172, 0.0188, 0.0169, 0.0110], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 01:09:16,317 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.090e+02 2.805e+02 3.291e+02 3.728e+02 9.830e+02, threshold=6.583e+02, percent-clipped=1.0 2023-04-17 01:09:29,519 INFO [train.py:893] (1/4) Epoch 20, batch 950, loss[loss=0.1984, simple_loss=0.2599, pruned_loss=0.0685, over 13422.00 frames. ], tot_loss[loss=0.1838, simple_loss=0.2418, pruned_loss=0.06289, over 2640089.82 frames. ], batch size: 95, lr: 6.57e-03, grad_scale: 16.0 2023-04-17 01:09:49,629 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52254.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:10:16,225 INFO [train.py:893] (1/4) Epoch 20, batch 1000, loss[loss=0.1731, simple_loss=0.2357, pruned_loss=0.05525, over 13367.00 frames. ], tot_loss[loss=0.1823, simple_loss=0.2399, pruned_loss=0.06231, over 2645367.89 frames. ], batch size: 109, lr: 6.57e-03, grad_scale: 16.0 2023-04-17 01:10:18,396 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-04-17 01:10:43,680 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52312.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:10:48,276 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.733e+02 2.626e+02 3.000e+02 3.496e+02 6.136e+02, threshold=6.000e+02, percent-clipped=0.0 2023-04-17 01:10:59,106 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3534, 4.1225, 4.4028, 4.3739, 4.6023, 4.2254, 4.6279, 4.5686], device='cuda:1'), covar=tensor([0.0453, 0.0638, 0.0601, 0.0479, 0.0609, 0.0832, 0.0453, 0.0543], device='cuda:1'), in_proj_covar=tensor([0.0279, 0.0288, 0.0293, 0.0218, 0.0422, 0.0335, 0.0268, 0.0293], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 01:11:02,261 INFO [train.py:893] (1/4) Epoch 20, batch 1050, loss[loss=0.1866, simple_loss=0.2438, pruned_loss=0.06471, over 13512.00 frames. ], tot_loss[loss=0.1802, simple_loss=0.2381, pruned_loss=0.06116, over 2651189.49 frames. ], batch size: 93, lr: 6.56e-03, grad_scale: 16.0 2023-04-17 01:11:25,900 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5340, 3.8153, 3.6966, 4.2741, 2.3754, 3.3077, 3.9392, 2.2984], device='cuda:1'), covar=tensor([0.0107, 0.0511, 0.0765, 0.0591, 0.1615, 0.0898, 0.0519, 0.1843], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0183, 0.0206, 0.0242, 0.0184, 0.0200, 0.0177, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 01:11:40,605 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52373.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:11:48,024 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.1893, 4.6948, 4.6278, 4.6968, 4.4644, 4.5897, 5.1688, 4.7313], device='cuda:1'), covar=tensor([0.0627, 0.1104, 0.1958, 0.2436, 0.0937, 0.1465, 0.0864, 0.1128], device='cuda:1'), in_proj_covar=tensor([0.0321, 0.0371, 0.0462, 0.0471, 0.0288, 0.0348, 0.0430, 0.0340], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 01:11:48,676 INFO [train.py:893] (1/4) Epoch 20, batch 1100, loss[loss=0.1698, simple_loss=0.2344, pruned_loss=0.05264, over 13459.00 frames. ], tot_loss[loss=0.1798, simple_loss=0.2386, pruned_loss=0.06049, over 2656649.50 frames. ], batch size: 100, lr: 6.56e-03, grad_scale: 16.0 2023-04-17 01:12:00,966 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52396.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:12:08,614 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52403.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:12:21,580 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.978e+02 2.765e+02 3.180e+02 3.679e+02 6.525e+02, threshold=6.360e+02, percent-clipped=1.0 2023-04-17 01:12:36,196 INFO [train.py:893] (1/4) Epoch 20, batch 1150, loss[loss=0.1903, simple_loss=0.2492, pruned_loss=0.06572, over 13384.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.2384, pruned_loss=0.06007, over 2658768.37 frames. ], batch size: 84, lr: 6.56e-03, grad_scale: 16.0 2023-04-17 01:12:53,215 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52451.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:13:22,952 INFO [train.py:893] (1/4) Epoch 20, batch 1200, loss[loss=0.1755, simple_loss=0.2351, pruned_loss=0.05795, over 13526.00 frames. ], tot_loss[loss=0.1798, simple_loss=0.2393, pruned_loss=0.06014, over 2663288.64 frames. ], batch size: 76, lr: 6.55e-03, grad_scale: 16.0 2023-04-17 01:13:49,577 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-17 01:13:49,804 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.1393, 2.1530, 4.1622, 3.8647, 4.0349, 3.1450, 3.7668, 3.0351], device='cuda:1'), covar=tensor([0.2145, 0.1521, 0.0101, 0.0253, 0.0185, 0.0587, 0.0249, 0.0812], device='cuda:1'), in_proj_covar=tensor([0.0217, 0.0186, 0.0120, 0.0126, 0.0129, 0.0174, 0.0141, 0.0182], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-17 01:13:56,084 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.988e+02 2.539e+02 3.132e+02 3.497e+02 5.997e+02, threshold=6.265e+02, percent-clipped=0.0 2023-04-17 01:14:02,644 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-17 01:14:10,240 INFO [train.py:893] (1/4) Epoch 20, batch 1250, loss[loss=0.1679, simple_loss=0.2226, pruned_loss=0.05659, over 13360.00 frames. ], tot_loss[loss=0.1804, simple_loss=0.2394, pruned_loss=0.06068, over 2665898.53 frames. ], batch size: 67, lr: 6.55e-03, grad_scale: 16.0 2023-04-17 01:14:25,463 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52549.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:14:31,113 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52555.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 01:14:56,398 INFO [train.py:893] (1/4) Epoch 20, batch 1300, loss[loss=0.168, simple_loss=0.2295, pruned_loss=0.05323, over 13367.00 frames. ], tot_loss[loss=0.1809, simple_loss=0.2405, pruned_loss=0.06069, over 2665816.79 frames. ], batch size: 73, lr: 6.55e-03, grad_scale: 16.0 2023-04-17 01:15:28,403 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52616.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 01:15:29,627 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.989e+02 2.703e+02 3.302e+02 3.792e+02 7.848e+02, threshold=6.603e+02, percent-clipped=1.0 2023-04-17 01:15:43,052 INFO [train.py:893] (1/4) Epoch 20, batch 1350, loss[loss=0.2061, simple_loss=0.263, pruned_loss=0.07463, over 13477.00 frames. ], tot_loss[loss=0.1816, simple_loss=0.2412, pruned_loss=0.06105, over 2667119.38 frames. ], batch size: 100, lr: 6.54e-03, grad_scale: 16.0 2023-04-17 01:16:15,628 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52668.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:16:30,166 INFO [train.py:893] (1/4) Epoch 20, batch 1400, loss[loss=0.1813, simple_loss=0.2422, pruned_loss=0.06017, over 13520.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.2401, pruned_loss=0.06047, over 2670089.90 frames. ], batch size: 81, lr: 6.54e-03, grad_scale: 16.0 2023-04-17 01:16:38,215 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-04-17 01:16:41,815 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52696.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:16:48,387 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4755, 3.2622, 3.9430, 2.8028, 2.6219, 2.7549, 4.2614, 4.3227], device='cuda:1'), covar=tensor([0.1248, 0.1869, 0.0379, 0.1783, 0.1601, 0.1474, 0.0235, 0.0204], device='cuda:1'), in_proj_covar=tensor([0.0243, 0.0261, 0.0190, 0.0220, 0.0216, 0.0179, 0.0203, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-17 01:17:02,871 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.002e+02 2.642e+02 3.013e+02 3.621e+02 7.035e+02, threshold=6.026e+02, percent-clipped=2.0 2023-04-17 01:17:16,198 INFO [train.py:893] (1/4) Epoch 20, batch 1450, loss[loss=0.1658, simple_loss=0.2098, pruned_loss=0.06085, over 12799.00 frames. ], tot_loss[loss=0.1809, simple_loss=0.2402, pruned_loss=0.0608, over 2666741.86 frames. ], batch size: 52, lr: 6.54e-03, grad_scale: 16.0 2023-04-17 01:17:26,930 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52744.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:17:34,546 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=52752.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:17:37,370 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0133, 2.9579, 2.7169, 1.9989, 2.1169, 2.6442, 2.7041, 3.2209], device='cuda:1'), covar=tensor([0.1041, 0.0304, 0.0587, 0.1514, 0.0484, 0.0488, 0.0774, 0.0204], device='cuda:1'), in_proj_covar=tensor([0.0163, 0.0137, 0.0117, 0.0206, 0.0109, 0.0157, 0.0169, 0.0127], device='cuda:1'), out_proj_covar=tensor([1.2039e-04, 1.0264e-04, 9.1416e-05, 1.5465e-04, 8.0275e-05, 1.1875e-04, 1.2755e-04, 9.4415e-05], device='cuda:1') 2023-04-17 01:18:00,994 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9360, 4.2318, 3.9666, 4.0127, 4.0987, 4.3839, 4.2247, 4.0285], device='cuda:1'), covar=tensor([0.0332, 0.0314, 0.0363, 0.0893, 0.0287, 0.0255, 0.0323, 0.0381], device='cuda:1'), in_proj_covar=tensor([0.0158, 0.0152, 0.0171, 0.0262, 0.0171, 0.0188, 0.0168, 0.0109], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 01:18:03,214 INFO [train.py:893] (1/4) Epoch 20, batch 1500, loss[loss=0.1602, simple_loss=0.2262, pruned_loss=0.04712, over 13504.00 frames. ], tot_loss[loss=0.1812, simple_loss=0.2408, pruned_loss=0.06078, over 2668932.48 frames. ], batch size: 81, lr: 6.54e-03, grad_scale: 32.0 2023-04-17 01:18:11,559 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2023-04-17 01:18:31,479 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=52813.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:18:36,241 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.621e+02 2.570e+02 2.979e+02 3.357e+02 4.933e+02, threshold=5.957e+02, percent-clipped=0.0 2023-04-17 01:18:50,221 INFO [train.py:893] (1/4) Epoch 20, batch 1550, loss[loss=0.1931, simple_loss=0.259, pruned_loss=0.06367, over 13080.00 frames. ], tot_loss[loss=0.1811, simple_loss=0.2409, pruned_loss=0.06067, over 2667638.85 frames. ], batch size: 142, lr: 6.53e-03, grad_scale: 32.0 2023-04-17 01:19:05,526 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52849.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:19:08,241 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.68 vs. limit=5.0 2023-04-17 01:19:37,147 INFO [train.py:893] (1/4) Epoch 20, batch 1600, loss[loss=0.1703, simple_loss=0.227, pruned_loss=0.05686, over 11919.00 frames. ], tot_loss[loss=0.1808, simple_loss=0.2407, pruned_loss=0.06051, over 2663833.24 frames. ], batch size: 157, lr: 6.53e-03, grad_scale: 32.0 2023-04-17 01:19:50,407 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=52897.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:20:03,638 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=52911.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 01:20:09,268 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.827e+02 2.611e+02 3.148e+02 3.639e+02 5.620e+02, threshold=6.296e+02, percent-clipped=0.0 2023-04-17 01:20:16,868 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.94 vs. limit=5.0 2023-04-17 01:20:23,718 INFO [train.py:893] (1/4) Epoch 20, batch 1650, loss[loss=0.1847, simple_loss=0.2387, pruned_loss=0.06541, over 13426.00 frames. ], tot_loss[loss=0.1813, simple_loss=0.2415, pruned_loss=0.06054, over 2659165.03 frames. ], batch size: 65, lr: 6.53e-03, grad_scale: 32.0 2023-04-17 01:20:49,887 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.95 vs. limit=5.0 2023-04-17 01:20:55,804 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=52968.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:21:09,738 INFO [train.py:893] (1/4) Epoch 20, batch 1700, loss[loss=0.1936, simple_loss=0.2546, pruned_loss=0.06631, over 13340.00 frames. ], tot_loss[loss=0.1821, simple_loss=0.2421, pruned_loss=0.0611, over 2658583.31 frames. ], batch size: 118, lr: 6.52e-03, grad_scale: 32.0 2023-04-17 01:21:20,560 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.7441, 5.2260, 5.1834, 5.3269, 5.0702, 5.1386, 5.8067, 5.2901], device='cuda:1'), covar=tensor([0.0683, 0.1028, 0.1998, 0.2150, 0.0766, 0.1433, 0.0690, 0.1021], device='cuda:1'), in_proj_covar=tensor([0.0320, 0.0365, 0.0454, 0.0464, 0.0282, 0.0344, 0.0419, 0.0334], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-17 01:21:38,609 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0000, 2.6943, 2.3349, 1.7203, 1.7643, 2.3701, 2.4301, 2.9531], device='cuda:1'), covar=tensor([0.1035, 0.0403, 0.0748, 0.1742, 0.0275, 0.0625, 0.0837, 0.0241], device='cuda:1'), in_proj_covar=tensor([0.0164, 0.0139, 0.0118, 0.0207, 0.0109, 0.0159, 0.0170, 0.0129], device='cuda:1'), out_proj_covar=tensor([1.2126e-04, 1.0410e-04, 9.2343e-05, 1.5520e-04, 8.0340e-05, 1.2001e-04, 1.2821e-04, 9.5657e-05], device='cuda:1') 2023-04-17 01:21:40,930 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53016.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:21:42,881 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-04-17 01:21:43,048 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.046e+02 2.784e+02 3.291e+02 3.982e+02 7.323e+02, threshold=6.582e+02, percent-clipped=2.0 2023-04-17 01:21:56,322 INFO [train.py:893] (1/4) Epoch 20, batch 1750, loss[loss=0.1929, simple_loss=0.2558, pruned_loss=0.06506, over 13541.00 frames. ], tot_loss[loss=0.1804, simple_loss=0.2403, pruned_loss=0.06026, over 2657580.80 frames. ], batch size: 85, lr: 6.52e-03, grad_scale: 32.0 2023-04-17 01:22:18,609 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53056.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:22:19,484 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8044, 3.8382, 2.7557, 3.4714, 3.8357, 2.5296, 3.4405, 2.6529], device='cuda:1'), covar=tensor([0.0276, 0.0247, 0.1096, 0.0350, 0.0256, 0.1218, 0.0570, 0.1386], device='cuda:1'), in_proj_covar=tensor([0.0157, 0.0170, 0.0176, 0.0204, 0.0134, 0.0158, 0.0158, 0.0158], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 01:22:38,677 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-04-17 01:22:43,001 INFO [train.py:893] (1/4) Epoch 20, batch 1800, loss[loss=0.1999, simple_loss=0.2608, pruned_loss=0.0695, over 13459.00 frames. ], tot_loss[loss=0.1795, simple_loss=0.2396, pruned_loss=0.05966, over 2659115.61 frames. ], batch size: 103, lr: 6.52e-03, grad_scale: 32.0 2023-04-17 01:23:06,739 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53108.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:23:09,680 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.53 vs. limit=5.0 2023-04-17 01:23:15,367 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53117.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:23:15,879 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.144e+02 2.701e+02 3.241e+02 3.784e+02 6.824e+02, threshold=6.483e+02, percent-clipped=1.0 2023-04-17 01:23:30,160 INFO [train.py:893] (1/4) Epoch 20, batch 1850, loss[loss=0.183, simple_loss=0.2406, pruned_loss=0.06271, over 13536.00 frames. ], tot_loss[loss=0.1789, simple_loss=0.2388, pruned_loss=0.05949, over 2659760.12 frames. ], batch size: 85, lr: 6.51e-03, grad_scale: 32.0 2023-04-17 01:23:33,580 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-17 01:23:36,185 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8935, 3.7784, 3.0049, 3.5257, 2.9846, 2.1061, 3.8482, 2.0422], device='cuda:1'), covar=tensor([0.0624, 0.0396, 0.0473, 0.0296, 0.0627, 0.1898, 0.0673, 0.1227], device='cuda:1'), in_proj_covar=tensor([0.0140, 0.0136, 0.0132, 0.0113, 0.0145, 0.0185, 0.0168, 0.0140], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 01:24:16,926 INFO [train.py:893] (1/4) Epoch 20, batch 1900, loss[loss=0.17, simple_loss=0.2333, pruned_loss=0.05334, over 13496.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.2383, pruned_loss=0.05941, over 2660634.03 frames. ], batch size: 93, lr: 6.51e-03, grad_scale: 32.0 2023-04-17 01:24:23,924 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0929, 4.3256, 4.1025, 4.1873, 4.2313, 4.5597, 4.3422, 4.2829], device='cuda:1'), covar=tensor([0.0299, 0.0292, 0.0325, 0.0901, 0.0273, 0.0243, 0.0288, 0.0272], device='cuda:1'), in_proj_covar=tensor([0.0159, 0.0153, 0.0173, 0.0264, 0.0173, 0.0189, 0.0170, 0.0111], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 01:24:40,503 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4044, 4.6695, 4.4408, 4.4330, 4.5182, 4.8583, 4.6522, 4.5622], device='cuda:1'), covar=tensor([0.0344, 0.0303, 0.0331, 0.0985, 0.0305, 0.0235, 0.0340, 0.0213], device='cuda:1'), in_proj_covar=tensor([0.0159, 0.0154, 0.0173, 0.0265, 0.0174, 0.0190, 0.0171, 0.0111], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 01:24:43,020 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53211.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 01:24:46,693 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.96 vs. limit=5.0 2023-04-17 01:24:49,388 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.873e+02 2.617e+02 3.070e+02 3.654e+02 7.055e+02, threshold=6.140e+02, percent-clipped=1.0 2023-04-17 01:25:03,679 INFO [train.py:893] (1/4) Epoch 20, batch 1950, loss[loss=0.193, simple_loss=0.2467, pruned_loss=0.06968, over 13503.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.237, pruned_loss=0.05881, over 2654003.23 frames. ], batch size: 91, lr: 6.51e-03, grad_scale: 32.0 2023-04-17 01:25:10,037 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.63 vs. limit=2.0 2023-04-17 01:25:11,373 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53241.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:25:27,806 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53259.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 01:25:49,069 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-04-17 01:25:50,101 INFO [train.py:893] (1/4) Epoch 20, batch 2000, loss[loss=0.1749, simple_loss=0.2357, pruned_loss=0.05704, over 13568.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.2379, pruned_loss=0.05945, over 2658131.49 frames. ], batch size: 89, lr: 6.50e-03, grad_scale: 32.0 2023-04-17 01:25:56,672 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-17 01:26:08,425 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53302.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:26:10,162 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53304.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:26:20,375 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53315.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:26:23,458 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.898e+02 2.652e+02 3.059e+02 3.675e+02 5.199e+02, threshold=6.119e+02, percent-clipped=0.0 2023-04-17 01:26:37,634 INFO [train.py:893] (1/4) Epoch 20, batch 2050, loss[loss=0.2033, simple_loss=0.261, pruned_loss=0.07279, over 13502.00 frames. ], tot_loss[loss=0.1808, simple_loss=0.2402, pruned_loss=0.06069, over 2659209.89 frames. ], batch size: 93, lr: 6.50e-03, grad_scale: 32.0 2023-04-17 01:26:51,872 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.98 vs. limit=5.0 2023-04-17 01:27:08,031 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53365.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:27:10,467 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=53367.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:27:18,235 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53376.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:27:24,718 INFO [train.py:893] (1/4) Epoch 20, batch 2100, loss[loss=0.1824, simple_loss=0.2375, pruned_loss=0.06368, over 13555.00 frames. ], tot_loss[loss=0.1811, simple_loss=0.2404, pruned_loss=0.06091, over 2651913.34 frames. ], batch size: 78, lr: 6.50e-03, grad_scale: 32.0 2023-04-17 01:27:45,014 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0119, 3.7681, 3.8953, 2.4010, 4.2527, 4.0491, 4.0596, 4.2291], device='cuda:1'), covar=tensor([0.0236, 0.0148, 0.0148, 0.1169, 0.0139, 0.0238, 0.0134, 0.0114], device='cuda:1'), in_proj_covar=tensor([0.0093, 0.0056, 0.0083, 0.0105, 0.0100, 0.0109, 0.0081, 0.0077], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 01:27:49,387 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53408.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:27:52,549 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53412.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:27:57,473 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.055e+02 2.630e+02 3.030e+02 3.715e+02 5.974e+02, threshold=6.059e+02, percent-clipped=0.0 2023-04-17 01:28:08,140 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=53428.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 01:28:12,161 INFO [train.py:893] (1/4) Epoch 20, batch 2150, loss[loss=0.1704, simple_loss=0.2343, pruned_loss=0.0533, over 13538.00 frames. ], tot_loss[loss=0.1804, simple_loss=0.2401, pruned_loss=0.06039, over 2651299.85 frames. ], batch size: 83, lr: 6.50e-03, grad_scale: 32.0 2023-04-17 01:28:19,758 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-04-17 01:28:22,979 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-04-17 01:28:34,421 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53456.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:28:59,755 INFO [train.py:893] (1/4) Epoch 20, batch 2200, loss[loss=0.1911, simple_loss=0.2543, pruned_loss=0.06399, over 13407.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.2392, pruned_loss=0.0597, over 2653410.11 frames. ], batch size: 109, lr: 6.49e-03, grad_scale: 32.0 2023-04-17 01:29:28,060 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.41 vs. limit=5.0 2023-04-17 01:29:32,362 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.983e+02 2.529e+02 2.934e+02 3.600e+02 1.155e+03, threshold=5.867e+02, percent-clipped=3.0 2023-04-17 01:29:42,551 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9385, 2.3122, 1.8400, 3.6951, 4.1181, 3.0777, 4.1130, 3.8938], device='cuda:1'), covar=tensor([0.0127, 0.1292, 0.1388, 0.0146, 0.0157, 0.0668, 0.0128, 0.0133], device='cuda:1'), in_proj_covar=tensor([0.0075, 0.0090, 0.0097, 0.0081, 0.0067, 0.0080, 0.0056, 0.0077], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 01:29:45,638 INFO [train.py:893] (1/4) Epoch 20, batch 2250, loss[loss=0.1616, simple_loss=0.2183, pruned_loss=0.05242, over 13393.00 frames. ], tot_loss[loss=0.1775, simple_loss=0.2374, pruned_loss=0.05886, over 2656249.96 frames. ], batch size: 62, lr: 6.49e-03, grad_scale: 32.0 2023-04-17 01:30:32,420 INFO [train.py:893] (1/4) Epoch 20, batch 2300, loss[loss=0.1727, simple_loss=0.2381, pruned_loss=0.05369, over 13392.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2367, pruned_loss=0.05856, over 2659731.64 frames. ], batch size: 109, lr: 6.49e-03, grad_scale: 32.0 2023-04-17 01:30:44,844 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53597.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:31:05,026 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.562e+02 2.476e+02 2.872e+02 3.310e+02 5.586e+02, threshold=5.744e+02, percent-clipped=0.0 2023-04-17 01:31:19,023 INFO [train.py:893] (1/4) Epoch 20, batch 2350, loss[loss=0.2047, simple_loss=0.2546, pruned_loss=0.07746, over 13533.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2365, pruned_loss=0.05875, over 2651644.63 frames. ], batch size: 85, lr: 6.48e-03, grad_scale: 32.0 2023-04-17 01:31:20,137 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.3839, 2.2983, 4.2507, 3.9617, 4.1929, 3.3007, 3.9034, 3.2278], device='cuda:1'), covar=tensor([0.1790, 0.1530, 0.0098, 0.0213, 0.0197, 0.0584, 0.0227, 0.0671], device='cuda:1'), in_proj_covar=tensor([0.0216, 0.0187, 0.0122, 0.0126, 0.0131, 0.0173, 0.0141, 0.0184], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-17 01:31:42,134 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-17 01:31:43,965 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53660.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:31:54,522 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53671.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:32:05,814 INFO [train.py:893] (1/4) Epoch 20, batch 2400, loss[loss=0.2001, simple_loss=0.2612, pruned_loss=0.06945, over 13554.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2366, pruned_loss=0.05886, over 2656635.98 frames. ], batch size: 78, lr: 6.48e-03, grad_scale: 32.0 2023-04-17 01:32:32,923 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53712.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:32:38,396 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.897e+02 2.663e+02 3.122e+02 3.605e+02 1.210e+03, threshold=6.243e+02, percent-clipped=2.0 2023-04-17 01:32:42,838 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=53723.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 01:32:51,773 INFO [train.py:893] (1/4) Epoch 20, batch 2450, loss[loss=0.1848, simple_loss=0.2499, pruned_loss=0.05983, over 13448.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.237, pruned_loss=0.05882, over 2658932.72 frames. ], batch size: 103, lr: 6.48e-03, grad_scale: 32.0 2023-04-17 01:33:17,557 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53760.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:33:38,964 INFO [train.py:893] (1/4) Epoch 20, batch 2500, loss[loss=0.1784, simple_loss=0.2357, pruned_loss=0.06053, over 13515.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2371, pruned_loss=0.05887, over 2660146.26 frames. ], batch size: 98, lr: 6.47e-03, grad_scale: 32.0 2023-04-17 01:34:11,278 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.985e+02 2.616e+02 2.994e+02 3.505e+02 5.181e+02, threshold=5.987e+02, percent-clipped=0.0 2023-04-17 01:34:26,069 INFO [train.py:893] (1/4) Epoch 20, batch 2550, loss[loss=0.1606, simple_loss=0.213, pruned_loss=0.05413, over 13169.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2375, pruned_loss=0.05898, over 2664231.08 frames. ], batch size: 58, lr: 6.47e-03, grad_scale: 32.0 2023-04-17 01:34:47,470 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.95 vs. limit=5.0 2023-04-17 01:34:48,591 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-17 01:35:11,914 INFO [train.py:893] (1/4) Epoch 20, batch 2600, loss[loss=0.1651, simple_loss=0.2255, pruned_loss=0.05234, over 13369.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2368, pruned_loss=0.05876, over 2658470.67 frames. ], batch size: 73, lr: 6.47e-03, grad_scale: 32.0 2023-04-17 01:35:26,232 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53897.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:35:37,187 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5590, 2.2503, 2.6371, 3.8239, 3.4437, 3.9082, 2.9570, 2.3238], device='cuda:1'), covar=tensor([0.0196, 0.0858, 0.0782, 0.0070, 0.0237, 0.0064, 0.0656, 0.0909], device='cuda:1'), in_proj_covar=tensor([0.0115, 0.0144, 0.0162, 0.0095, 0.0115, 0.0092, 0.0162, 0.0144], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 01:35:39,507 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.8412, 2.6606, 2.2743, 1.4925, 1.5957, 2.0919, 2.3633, 2.8580], device='cuda:1'), covar=tensor([0.0938, 0.0316, 0.0748, 0.1648, 0.0233, 0.0544, 0.0729, 0.0258], device='cuda:1'), in_proj_covar=tensor([0.0165, 0.0140, 0.0118, 0.0207, 0.0109, 0.0157, 0.0171, 0.0128], device='cuda:1'), out_proj_covar=tensor([1.2206e-04, 1.0475e-04, 9.2780e-05, 1.5464e-04, 7.9780e-05, 1.1889e-04, 1.2883e-04, 9.5312e-05], device='cuda:1') 2023-04-17 01:35:43,018 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.831e+02 2.669e+02 3.176e+02 3.708e+02 8.423e+02, threshold=6.351e+02, percent-clipped=2.0 2023-04-17 01:35:54,208 INFO [train.py:893] (1/4) Epoch 20, batch 2650, loss[loss=0.1893, simple_loss=0.2537, pruned_loss=0.06246, over 13401.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.2377, pruned_loss=0.05935, over 2661817.22 frames. ], batch size: 113, lr: 6.47e-03, grad_scale: 32.0 2023-04-17 01:36:03,759 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=53945.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:36:15,732 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53960.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:36:23,990 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=53971.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:36:51,753 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-17 01:37:02,396 INFO [train.py:893] (1/4) Epoch 21, batch 0, loss[loss=0.1669, simple_loss=0.2269, pruned_loss=0.05345, over 13520.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2269, pruned_loss=0.05345, over 13520.00 frames. ], batch size: 72, lr: 6.30e-03, grad_scale: 32.0 2023-04-17 01:37:02,396 INFO [train.py:918] (1/4) Computing validation loss 2023-04-17 01:37:21,057 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.2020, 5.2536, 5.2799, 5.0947, 5.6022, 5.1900, 5.5241, 5.5508], device='cuda:1'), covar=tensor([0.0358, 0.0473, 0.0646, 0.0572, 0.0468, 0.0710, 0.0469, 0.0415], device='cuda:1'), in_proj_covar=tensor([0.0281, 0.0291, 0.0299, 0.0223, 0.0422, 0.0335, 0.0271, 0.0296], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 01:37:25,153 INFO [train.py:927] (1/4) Epoch 21, validation: loss=0.1376, simple_loss=0.1969, pruned_loss=0.03912, over 2446609.00 frames. 2023-04-17 01:37:25,154 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12831MB 2023-04-17 01:37:54,555 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=54008.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:37:54,647 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8974, 4.1267, 3.9064, 3.9949, 4.0834, 4.3306, 4.1721, 3.9082], device='cuda:1'), covar=tensor([0.0280, 0.0268, 0.0341, 0.0770, 0.0259, 0.0221, 0.0268, 0.0425], device='cuda:1'), in_proj_covar=tensor([0.0160, 0.0154, 0.0173, 0.0262, 0.0174, 0.0189, 0.0170, 0.0110], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 01:38:03,425 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.635e+02 2.723e+02 3.218e+02 3.819e+02 6.582e+02, threshold=6.437e+02, percent-clipped=1.0 2023-04-17 01:38:04,495 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=54019.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:38:08,776 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=54023.0, num_to_drop=1, layers_to_drop={2} 2023-04-17 01:38:16,638 INFO [train.py:893] (1/4) Epoch 21, batch 50, loss[loss=0.165, simple_loss=0.2122, pruned_loss=0.05891, over 12487.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2308, pruned_loss=0.05843, over 597927.09 frames. ], batch size: 51, lr: 6.30e-03, grad_scale: 32.0 2023-04-17 01:38:24,610 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5789, 2.8436, 2.7151, 4.5140, 4.9289, 3.6439, 4.8380, 4.6140], device='cuda:1'), covar=tensor([0.0081, 0.0786, 0.0761, 0.0077, 0.0062, 0.0407, 0.0069, 0.0067], device='cuda:1'), in_proj_covar=tensor([0.0075, 0.0091, 0.0096, 0.0081, 0.0066, 0.0080, 0.0055, 0.0076], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 01:38:39,938 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-17 01:38:39,938 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-17 01:38:39,939 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-17 01:38:39,949 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-17 01:38:39,967 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-17 01:38:39,979 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-17 01:38:39,997 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-17 01:38:53,037 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=54071.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:38:55,751 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2957, 2.5933, 2.2862, 4.2152, 4.6459, 3.5053, 4.5340, 4.3506], device='cuda:1'), covar=tensor([0.0096, 0.0872, 0.0903, 0.0082, 0.0063, 0.0410, 0.0077, 0.0076], device='cuda:1'), in_proj_covar=tensor([0.0074, 0.0090, 0.0096, 0.0081, 0.0066, 0.0080, 0.0055, 0.0076], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 01:39:02,797 INFO [train.py:893] (1/4) Epoch 21, batch 100, loss[loss=0.18, simple_loss=0.2459, pruned_loss=0.05708, over 13413.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2334, pruned_loss=0.06003, over 1060385.53 frames. ], batch size: 95, lr: 6.30e-03, grad_scale: 32.0 2023-04-17 01:39:08,509 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-04-17 01:39:24,012 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.57 vs. limit=5.0 2023-04-17 01:39:36,883 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.038e+02 2.853e+02 3.270e+02 3.963e+02 8.282e+02, threshold=6.541e+02, percent-clipped=1.0 2023-04-17 01:39:43,444 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-04-17 01:39:49,418 INFO [train.py:893] (1/4) Epoch 21, batch 150, loss[loss=0.1593, simple_loss=0.2195, pruned_loss=0.04948, over 13539.00 frames. ], tot_loss[loss=0.18, simple_loss=0.2365, pruned_loss=0.06179, over 1401607.73 frames. ], batch size: 76, lr: 6.30e-03, grad_scale: 32.0 2023-04-17 01:40:36,877 INFO [train.py:893] (1/4) Epoch 21, batch 200, loss[loss=0.2091, simple_loss=0.2607, pruned_loss=0.07875, over 12035.00 frames. ], tot_loss[loss=0.1816, simple_loss=0.2386, pruned_loss=0.06232, over 1672483.24 frames. ], batch size: 157, lr: 6.29e-03, grad_scale: 32.0 2023-04-17 01:40:38,919 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7726, 3.6409, 2.9980, 3.3124, 3.0136, 2.0466, 3.7395, 2.0514], device='cuda:1'), covar=tensor([0.0696, 0.0597, 0.0447, 0.0396, 0.0735, 0.2129, 0.0892, 0.1370], device='cuda:1'), in_proj_covar=tensor([0.0140, 0.0137, 0.0132, 0.0115, 0.0146, 0.0187, 0.0171, 0.0141], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 01:41:10,112 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.703e+02 3.137e+02 3.628e+02 8.630e+02, threshold=6.273e+02, percent-clipped=1.0 2023-04-17 01:41:23,276 INFO [train.py:893] (1/4) Epoch 21, batch 250, loss[loss=0.1974, simple_loss=0.2546, pruned_loss=0.0701, over 13561.00 frames. ], tot_loss[loss=0.1831, simple_loss=0.2404, pruned_loss=0.06291, over 1894083.13 frames. ], batch size: 89, lr: 6.29e-03, grad_scale: 32.0 2023-04-17 01:41:27,281 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-04-17 01:42:05,911 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0208, 3.9023, 4.0286, 2.3809, 4.3421, 4.1175, 4.0950, 4.3637], device='cuda:1'), covar=tensor([0.0248, 0.0160, 0.0127, 0.1190, 0.0152, 0.0249, 0.0149, 0.0085], device='cuda:1'), in_proj_covar=tensor([0.0092, 0.0056, 0.0082, 0.0104, 0.0098, 0.0109, 0.0080, 0.0075], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 01:42:09,166 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2650, 4.6881, 4.4753, 4.4356, 4.5096, 4.3003, 4.7495, 4.7579], device='cuda:1'), covar=tensor([0.0203, 0.0224, 0.0206, 0.0324, 0.0236, 0.0255, 0.0236, 0.0224], device='cuda:1'), in_proj_covar=tensor([0.0136, 0.0197, 0.0158, 0.0178, 0.0144, 0.0193, 0.0134, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 01:42:09,744 INFO [train.py:893] (1/4) Epoch 21, batch 300, loss[loss=0.1886, simple_loss=0.2451, pruned_loss=0.06606, over 13366.00 frames. ], tot_loss[loss=0.1831, simple_loss=0.2407, pruned_loss=0.06275, over 2065214.94 frames. ], batch size: 77, lr: 6.29e-03, grad_scale: 32.0 2023-04-17 01:42:15,732 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.0545, 4.5825, 4.4933, 4.5552, 4.2097, 4.3914, 5.0531, 4.6370], device='cuda:1'), covar=tensor([0.0744, 0.1178, 0.2210, 0.2604, 0.1149, 0.1692, 0.0914, 0.1177], device='cuda:1'), in_proj_covar=tensor([0.0326, 0.0371, 0.0462, 0.0469, 0.0289, 0.0346, 0.0427, 0.0341], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-17 01:42:43,127 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.688e+02 2.678e+02 3.033e+02 3.503e+02 6.854e+02, threshold=6.067e+02, percent-clipped=2.0 2023-04-17 01:42:55,536 INFO [train.py:893] (1/4) Epoch 21, batch 350, loss[loss=0.1621, simple_loss=0.2145, pruned_loss=0.05486, over 13352.00 frames. ], tot_loss[loss=0.1841, simple_loss=0.2416, pruned_loss=0.06325, over 2195175.42 frames. ], batch size: 67, lr: 6.28e-03, grad_scale: 32.0 2023-04-17 01:43:06,832 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1677, 4.5245, 4.2822, 4.2627, 4.3187, 4.6483, 4.4708, 4.3221], device='cuda:1'), covar=tensor([0.0312, 0.0277, 0.0286, 0.0986, 0.0281, 0.0218, 0.0273, 0.0269], device='cuda:1'), in_proj_covar=tensor([0.0161, 0.0156, 0.0175, 0.0264, 0.0174, 0.0190, 0.0170, 0.0111], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 01:43:42,898 INFO [train.py:893] (1/4) Epoch 21, batch 400, loss[loss=0.1684, simple_loss=0.229, pruned_loss=0.05392, over 13466.00 frames. ], tot_loss[loss=0.1832, simple_loss=0.2414, pruned_loss=0.06247, over 2301092.30 frames. ], batch size: 79, lr: 6.28e-03, grad_scale: 32.0 2023-04-17 01:44:06,089 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.67 vs. limit=2.0 2023-04-17 01:44:17,085 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.854e+02 2.635e+02 3.018e+02 3.787e+02 7.372e+02, threshold=6.036e+02, percent-clipped=3.0 2023-04-17 01:44:25,678 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9808, 4.4200, 4.1509, 4.2061, 4.1449, 4.0532, 4.4261, 4.4794], device='cuda:1'), covar=tensor([0.0227, 0.0224, 0.0244, 0.0337, 0.0280, 0.0299, 0.0278, 0.0249], device='cuda:1'), in_proj_covar=tensor([0.0137, 0.0197, 0.0160, 0.0178, 0.0145, 0.0195, 0.0134, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 01:44:25,926 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.61 vs. limit=2.0 2023-04-17 01:44:29,299 INFO [train.py:893] (1/4) Epoch 21, batch 450, loss[loss=0.1884, simple_loss=0.2473, pruned_loss=0.06473, over 13416.00 frames. ], tot_loss[loss=0.1845, simple_loss=0.2431, pruned_loss=0.063, over 2383963.10 frames. ], batch size: 113, lr: 6.28e-03, grad_scale: 32.0 2023-04-17 01:44:41,093 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3920, 3.2067, 3.9640, 2.7647, 2.5351, 2.6326, 4.2192, 4.2637], device='cuda:1'), covar=tensor([0.1343, 0.1817, 0.0378, 0.1800, 0.1659, 0.1719, 0.0285, 0.0245], device='cuda:1'), in_proj_covar=tensor([0.0246, 0.0265, 0.0193, 0.0222, 0.0217, 0.0182, 0.0204, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-17 01:44:53,164 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-17 01:44:56,480 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9034, 4.0033, 3.9277, 2.3446, 4.3106, 4.0544, 4.0815, 4.3425], device='cuda:1'), covar=tensor([0.0388, 0.0167, 0.0198, 0.1483, 0.0230, 0.0357, 0.0198, 0.0158], device='cuda:1'), in_proj_covar=tensor([0.0091, 0.0055, 0.0082, 0.0103, 0.0098, 0.0108, 0.0079, 0.0075], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 01:45:16,437 INFO [train.py:893] (1/4) Epoch 21, batch 500, loss[loss=0.1659, simple_loss=0.2279, pruned_loss=0.05196, over 13530.00 frames. ], tot_loss[loss=0.1847, simple_loss=0.2436, pruned_loss=0.06289, over 2440517.26 frames. ], batch size: 72, lr: 6.28e-03, grad_scale: 32.0 2023-04-17 01:45:34,906 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0028, 3.8508, 3.9968, 2.3765, 4.1792, 4.0522, 4.0195, 4.2923], device='cuda:1'), covar=tensor([0.0228, 0.0150, 0.0116, 0.1143, 0.0154, 0.0230, 0.0133, 0.0091], device='cuda:1'), in_proj_covar=tensor([0.0092, 0.0056, 0.0083, 0.0104, 0.0099, 0.0109, 0.0080, 0.0075], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 01:45:49,142 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.865e+02 2.575e+02 3.052e+02 3.653e+02 5.536e+02, threshold=6.104e+02, percent-clipped=0.0 2023-04-17 01:46:02,459 INFO [train.py:893] (1/4) Epoch 21, batch 550, loss[loss=0.184, simple_loss=0.2446, pruned_loss=0.06167, over 13194.00 frames. ], tot_loss[loss=0.1843, simple_loss=0.2435, pruned_loss=0.06256, over 2489320.27 frames. ], batch size: 132, lr: 6.27e-03, grad_scale: 32.0 2023-04-17 01:46:49,285 INFO [train.py:893] (1/4) Epoch 21, batch 600, loss[loss=0.1843, simple_loss=0.2478, pruned_loss=0.0604, over 13569.00 frames. ], tot_loss[loss=0.183, simple_loss=0.2421, pruned_loss=0.06197, over 2529210.18 frames. ], batch size: 89, lr: 6.27e-03, grad_scale: 32.0 2023-04-17 01:46:58,025 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7111, 3.9258, 3.7758, 3.7844, 3.8710, 3.7257, 3.9606, 4.0120], device='cuda:1'), covar=tensor([0.0217, 0.0273, 0.0232, 0.0325, 0.0237, 0.0301, 0.0287, 0.0276], device='cuda:1'), in_proj_covar=tensor([0.0135, 0.0195, 0.0158, 0.0177, 0.0144, 0.0193, 0.0133, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 01:46:59,953 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6407, 3.4507, 4.2107, 2.9406, 2.8075, 2.8469, 4.5121, 4.5685], device='cuda:1'), covar=tensor([0.1251, 0.1759, 0.0386, 0.1863, 0.1620, 0.1587, 0.0270, 0.0330], device='cuda:1'), in_proj_covar=tensor([0.0246, 0.0265, 0.0193, 0.0222, 0.0217, 0.0182, 0.0205, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 01:47:23,765 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.860e+02 2.541e+02 2.955e+02 3.597e+02 5.812e+02, threshold=5.910e+02, percent-clipped=0.0 2023-04-17 01:47:37,095 INFO [train.py:893] (1/4) Epoch 21, batch 650, loss[loss=0.1589, simple_loss=0.224, pruned_loss=0.04695, over 13252.00 frames. ], tot_loss[loss=0.1813, simple_loss=0.2402, pruned_loss=0.06119, over 2551830.49 frames. ], batch size: 124, lr: 6.27e-03, grad_scale: 32.0 2023-04-17 01:48:07,402 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54664.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 01:48:23,877 INFO [train.py:893] (1/4) Epoch 21, batch 700, loss[loss=0.196, simple_loss=0.2526, pruned_loss=0.06972, over 13222.00 frames. ], tot_loss[loss=0.1801, simple_loss=0.2394, pruned_loss=0.06045, over 2578033.52 frames. ], batch size: 117, lr: 6.26e-03, grad_scale: 32.0 2023-04-17 01:48:56,680 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9268, 2.7752, 2.5018, 1.8638, 1.8727, 2.4438, 2.5741, 3.0711], device='cuda:1'), covar=tensor([0.0999, 0.0376, 0.0620, 0.1621, 0.0328, 0.0480, 0.0718, 0.0208], device='cuda:1'), in_proj_covar=tensor([0.0165, 0.0141, 0.0119, 0.0208, 0.0109, 0.0158, 0.0170, 0.0129], device='cuda:1'), out_proj_covar=tensor([1.2203e-04, 1.0574e-04, 9.3181e-05, 1.5555e-04, 7.9710e-05, 1.1953e-04, 1.2853e-04, 9.5413e-05], device='cuda:1') 2023-04-17 01:48:57,157 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.970e+02 2.709e+02 2.989e+02 3.684e+02 5.126e+02, threshold=5.978e+02, percent-clipped=0.0 2023-04-17 01:49:04,932 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54725.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 01:49:11,223 INFO [train.py:893] (1/4) Epoch 21, batch 750, loss[loss=0.1757, simple_loss=0.2399, pruned_loss=0.05577, over 13423.00 frames. ], tot_loss[loss=0.1806, simple_loss=0.2393, pruned_loss=0.06094, over 2596548.67 frames. ], batch size: 95, lr: 6.26e-03, grad_scale: 32.0 2023-04-17 01:49:54,966 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4147, 4.1800, 4.4287, 4.3932, 4.6689, 4.2554, 4.6908, 4.6256], device='cuda:1'), covar=tensor([0.0449, 0.0582, 0.0571, 0.0498, 0.0544, 0.0723, 0.0426, 0.0433], device='cuda:1'), in_proj_covar=tensor([0.0284, 0.0293, 0.0302, 0.0222, 0.0427, 0.0337, 0.0275, 0.0295], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 01:49:56,450 INFO [train.py:893] (1/4) Epoch 21, batch 800, loss[loss=0.1785, simple_loss=0.2354, pruned_loss=0.06082, over 11849.00 frames. ], tot_loss[loss=0.1816, simple_loss=0.2405, pruned_loss=0.06134, over 2611125.78 frames. ], batch size: 158, lr: 6.26e-03, grad_scale: 64.0 2023-04-17 01:50:31,163 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.910e+02 2.667e+02 3.203e+02 3.678e+02 7.837e+02, threshold=6.406e+02, percent-clipped=2.0 2023-04-17 01:50:43,694 INFO [train.py:893] (1/4) Epoch 21, batch 850, loss[loss=0.1941, simple_loss=0.231, pruned_loss=0.07867, over 12546.00 frames. ], tot_loss[loss=0.183, simple_loss=0.2418, pruned_loss=0.06213, over 2622913.89 frames. ], batch size: 51, lr: 6.26e-03, grad_scale: 64.0 2023-04-17 01:51:12,053 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=54862.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:51:30,857 INFO [train.py:893] (1/4) Epoch 21, batch 900, loss[loss=0.2045, simple_loss=0.2523, pruned_loss=0.07835, over 13170.00 frames. ], tot_loss[loss=0.1832, simple_loss=0.2414, pruned_loss=0.06251, over 2626522.99 frames. ], batch size: 58, lr: 6.25e-03, grad_scale: 32.0 2023-04-17 01:51:39,372 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4596, 4.2333, 4.3558, 2.9412, 4.7907, 4.4733, 4.4932, 4.6871], device='cuda:1'), covar=tensor([0.0215, 0.0127, 0.0142, 0.0973, 0.0121, 0.0259, 0.0130, 0.0111], device='cuda:1'), in_proj_covar=tensor([0.0092, 0.0055, 0.0082, 0.0102, 0.0098, 0.0108, 0.0079, 0.0074], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 01:52:00,551 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-17 01:52:05,572 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.031e+02 2.828e+02 3.269e+02 3.911e+02 9.881e+02, threshold=6.537e+02, percent-clipped=2.0 2023-04-17 01:52:09,248 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=54923.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 01:52:17,033 INFO [train.py:893] (1/4) Epoch 21, batch 950, loss[loss=0.1787, simple_loss=0.2401, pruned_loss=0.05867, over 13368.00 frames. ], tot_loss[loss=0.1825, simple_loss=0.2404, pruned_loss=0.06229, over 2634677.11 frames. ], batch size: 73, lr: 6.25e-03, grad_scale: 32.0 2023-04-17 01:52:23,838 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.3785, 4.9034, 4.8665, 4.8946, 4.6275, 4.7399, 5.3593, 4.8363], device='cuda:1'), covar=tensor([0.0740, 0.1216, 0.2021, 0.2455, 0.0957, 0.1553, 0.0859, 0.1071], device='cuda:1'), in_proj_covar=tensor([0.0325, 0.0369, 0.0459, 0.0464, 0.0290, 0.0346, 0.0425, 0.0341], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-17 01:53:02,991 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1501, 4.6125, 4.3851, 4.4109, 4.4370, 4.2751, 4.6883, 4.7168], device='cuda:1'), covar=tensor([0.0230, 0.0223, 0.0180, 0.0331, 0.0247, 0.0266, 0.0269, 0.0232], device='cuda:1'), in_proj_covar=tensor([0.0138, 0.0199, 0.0160, 0.0179, 0.0147, 0.0196, 0.0134, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 01:53:03,583 INFO [train.py:893] (1/4) Epoch 21, batch 1000, loss[loss=0.1653, simple_loss=0.2229, pruned_loss=0.05382, over 13495.00 frames. ], tot_loss[loss=0.1807, simple_loss=0.2383, pruned_loss=0.06149, over 2642288.27 frames. ], batch size: 81, lr: 6.25e-03, grad_scale: 32.0 2023-04-17 01:53:32,256 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4470, 4.6262, 3.3121, 4.2248, 4.4410, 2.9035, 4.0708, 3.1828], device='cuda:1'), covar=tensor([0.0284, 0.0255, 0.0922, 0.0380, 0.0202, 0.1103, 0.0370, 0.1174], device='cuda:1'), in_proj_covar=tensor([0.0159, 0.0175, 0.0176, 0.0210, 0.0136, 0.0160, 0.0161, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 01:53:38,541 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.568e+02 2.603e+02 3.088e+02 3.704e+02 6.747e+02, threshold=6.176e+02, percent-clipped=2.0 2023-04-17 01:53:39,631 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55020.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 01:53:50,170 INFO [train.py:893] (1/4) Epoch 21, batch 1050, loss[loss=0.1912, simple_loss=0.2552, pruned_loss=0.06359, over 13427.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.2363, pruned_loss=0.06051, over 2642992.62 frames. ], batch size: 95, lr: 6.24e-03, grad_scale: 32.0 2023-04-17 01:54:36,880 INFO [train.py:893] (1/4) Epoch 21, batch 1100, loss[loss=0.1662, simple_loss=0.232, pruned_loss=0.0502, over 13204.00 frames. ], tot_loss[loss=0.1783, simple_loss=0.237, pruned_loss=0.05977, over 2644987.03 frames. ], batch size: 132, lr: 6.24e-03, grad_scale: 32.0 2023-04-17 01:55:11,817 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.920e+02 2.491e+02 2.960e+02 3.530e+02 7.455e+02, threshold=5.920e+02, percent-clipped=4.0 2023-04-17 01:55:22,740 INFO [train.py:893] (1/4) Epoch 21, batch 1150, loss[loss=0.2037, simple_loss=0.2599, pruned_loss=0.07372, over 11561.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.2376, pruned_loss=0.05971, over 2648387.27 frames. ], batch size: 157, lr: 6.24e-03, grad_scale: 32.0 2023-04-17 01:55:39,462 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55148.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:56:07,847 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0244, 2.7130, 2.4347, 1.8128, 1.7378, 2.4096, 2.5468, 2.9813], device='cuda:1'), covar=tensor([0.0800, 0.0308, 0.0579, 0.1427, 0.0267, 0.0448, 0.0629, 0.0233], device='cuda:1'), in_proj_covar=tensor([0.0163, 0.0139, 0.0120, 0.0207, 0.0110, 0.0157, 0.0169, 0.0129], device='cuda:1'), out_proj_covar=tensor([1.2077e-04, 1.0420e-04, 9.3980e-05, 1.5458e-04, 8.0178e-05, 1.1903e-04, 1.2794e-04, 9.5388e-05], device='cuda:1') 2023-04-17 01:56:10,685 INFO [train.py:893] (1/4) Epoch 21, batch 1200, loss[loss=0.1683, simple_loss=0.2194, pruned_loss=0.05863, over 12623.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.2379, pruned_loss=0.05946, over 2650337.70 frames. ], batch size: 51, lr: 6.24e-03, grad_scale: 16.0 2023-04-17 01:56:36,167 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55209.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:56:36,801 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-17 01:56:37,961 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55211.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:56:44,318 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55218.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 01:56:45,831 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.729e+02 2.461e+02 2.911e+02 3.858e+02 6.055e+02, threshold=5.823e+02, percent-clipped=2.0 2023-04-17 01:56:47,660 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-17 01:56:57,269 INFO [train.py:893] (1/4) Epoch 21, batch 1250, loss[loss=0.1928, simple_loss=0.2452, pruned_loss=0.07018, over 13452.00 frames. ], tot_loss[loss=0.1793, simple_loss=0.2387, pruned_loss=0.06, over 2654032.42 frames. ], batch size: 79, lr: 6.23e-03, grad_scale: 16.0 2023-04-17 01:57:34,456 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55272.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 01:57:43,313 INFO [train.py:893] (1/4) Epoch 21, batch 1300, loss[loss=0.1896, simple_loss=0.2447, pruned_loss=0.06725, over 13491.00 frames. ], tot_loss[loss=0.1798, simple_loss=0.2393, pruned_loss=0.06018, over 2658787.69 frames. ], batch size: 93, lr: 6.23e-03, grad_scale: 16.0 2023-04-17 01:58:17,914 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.936e+02 2.608e+02 3.169e+02 3.646e+02 7.659e+02, threshold=6.337e+02, percent-clipped=2.0 2023-04-17 01:58:18,242 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55320.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 01:58:30,156 INFO [train.py:893] (1/4) Epoch 21, batch 1350, loss[loss=0.2088, simple_loss=0.2643, pruned_loss=0.07665, over 13445.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.24, pruned_loss=0.06047, over 2662060.71 frames. ], batch size: 103, lr: 6.23e-03, grad_scale: 16.0 2023-04-17 01:59:04,318 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55368.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 01:59:16,582 INFO [train.py:893] (1/4) Epoch 21, batch 1400, loss[loss=0.1878, simple_loss=0.2496, pruned_loss=0.06299, over 13237.00 frames. ], tot_loss[loss=0.18, simple_loss=0.2393, pruned_loss=0.06032, over 2659203.17 frames. ], batch size: 117, lr: 6.22e-03, grad_scale: 16.0 2023-04-17 01:59:52,765 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.807e+02 2.623e+02 2.979e+02 3.447e+02 8.566e+02, threshold=5.958e+02, percent-clipped=5.0 2023-04-17 02:00:03,884 INFO [train.py:893] (1/4) Epoch 21, batch 1450, loss[loss=0.172, simple_loss=0.2308, pruned_loss=0.05658, over 13505.00 frames. ], tot_loss[loss=0.1788, simple_loss=0.2382, pruned_loss=0.05971, over 2665205.21 frames. ], batch size: 81, lr: 6.22e-03, grad_scale: 16.0 2023-04-17 02:00:15,781 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.7127, 4.4813, 4.7321, 4.6840, 4.9925, 4.5155, 5.0415, 4.9811], device='cuda:1'), covar=tensor([0.0429, 0.0586, 0.0694, 0.0535, 0.0560, 0.0874, 0.0418, 0.0472], device='cuda:1'), in_proj_covar=tensor([0.0285, 0.0294, 0.0304, 0.0222, 0.0430, 0.0340, 0.0276, 0.0296], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 02:00:31,104 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3497, 3.7724, 3.5088, 4.1300, 2.1841, 3.1171, 3.8747, 2.3709], device='cuda:1'), covar=tensor([0.0126, 0.0464, 0.0771, 0.0447, 0.1739, 0.0917, 0.0564, 0.1706], device='cuda:1'), in_proj_covar=tensor([0.0127, 0.0180, 0.0204, 0.0240, 0.0179, 0.0197, 0.0176, 0.0174], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 02:00:47,870 INFO [train.py:893] (1/4) Epoch 21, batch 1500, loss[loss=0.1599, simple_loss=0.2189, pruned_loss=0.05041, over 13486.00 frames. ], tot_loss[loss=0.1798, simple_loss=0.2394, pruned_loss=0.06008, over 2668873.17 frames. ], batch size: 70, lr: 6.22e-03, grad_scale: 16.0 2023-04-17 02:00:48,977 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8592, 4.2595, 4.0952, 4.1051, 4.1522, 3.9942, 4.3801, 4.3686], device='cuda:1'), covar=tensor([0.0280, 0.0276, 0.0209, 0.0365, 0.0233, 0.0280, 0.0219, 0.0261], device='cuda:1'), in_proj_covar=tensor([0.0138, 0.0201, 0.0161, 0.0180, 0.0149, 0.0196, 0.0134, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 02:01:08,545 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55504.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:01:21,950 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55518.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 02:01:23,376 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.154e+02 2.733e+02 3.047e+02 3.591e+02 6.078e+02, threshold=6.094e+02, percent-clipped=2.0 2023-04-17 02:01:34,156 INFO [train.py:893] (1/4) Epoch 21, batch 1550, loss[loss=0.157, simple_loss=0.2122, pruned_loss=0.0509, over 13429.00 frames. ], tot_loss[loss=0.1812, simple_loss=0.2408, pruned_loss=0.06082, over 2665182.33 frames. ], batch size: 62, lr: 6.22e-03, grad_scale: 16.0 2023-04-17 02:01:52,391 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4100, 2.0527, 2.6349, 3.7743, 3.4701, 3.8475, 3.1462, 2.1220], device='cuda:1'), covar=tensor([0.0252, 0.1126, 0.0799, 0.0079, 0.0237, 0.0065, 0.0586, 0.1042], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0148, 0.0165, 0.0098, 0.0117, 0.0094, 0.0163, 0.0144], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 02:02:05,319 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55566.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:02:06,942 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=55567.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:02:20,732 INFO [train.py:893] (1/4) Epoch 21, batch 1600, loss[loss=0.1784, simple_loss=0.2482, pruned_loss=0.05428, over 13248.00 frames. ], tot_loss[loss=0.1809, simple_loss=0.2408, pruned_loss=0.06047, over 2665640.08 frames. ], batch size: 117, lr: 6.21e-03, grad_scale: 16.0 2023-04-17 02:02:56,302 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.917e+02 2.614e+02 2.838e+02 3.865e+02 7.437e+02, threshold=5.675e+02, percent-clipped=3.0 2023-04-17 02:03:07,349 INFO [train.py:893] (1/4) Epoch 21, batch 1650, loss[loss=0.1985, simple_loss=0.2499, pruned_loss=0.07359, over 13561.00 frames. ], tot_loss[loss=0.18, simple_loss=0.2403, pruned_loss=0.0598, over 2665491.35 frames. ], batch size: 72, lr: 6.21e-03, grad_scale: 16.0 2023-04-17 02:03:12,882 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-04-17 02:03:54,308 INFO [train.py:893] (1/4) Epoch 21, batch 1700, loss[loss=0.1841, simple_loss=0.2492, pruned_loss=0.05944, over 13454.00 frames. ], tot_loss[loss=0.1802, simple_loss=0.2408, pruned_loss=0.05975, over 2663372.78 frames. ], batch size: 103, lr: 6.21e-03, grad_scale: 16.0 2023-04-17 02:04:20,606 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9470, 3.7820, 3.9487, 2.4108, 4.1403, 3.9815, 3.9567, 4.1461], device='cuda:1'), covar=tensor([0.0218, 0.0134, 0.0127, 0.1094, 0.0119, 0.0202, 0.0112, 0.0082], device='cuda:1'), in_proj_covar=tensor([0.0091, 0.0054, 0.0081, 0.0101, 0.0097, 0.0107, 0.0078, 0.0074], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 02:04:20,661 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6894, 3.4007, 2.7345, 2.9629, 2.7788, 1.9303, 3.4465, 2.0381], device='cuda:1'), covar=tensor([0.0728, 0.0619, 0.0536, 0.0546, 0.0733, 0.2213, 0.0994, 0.1429], device='cuda:1'), in_proj_covar=tensor([0.0141, 0.0140, 0.0136, 0.0116, 0.0147, 0.0187, 0.0173, 0.0142], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 02:04:29,421 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.606e+02 2.727e+02 3.146e+02 3.947e+02 9.582e+02, threshold=6.292e+02, percent-clipped=5.0 2023-04-17 02:04:40,169 INFO [train.py:893] (1/4) Epoch 21, batch 1750, loss[loss=0.1963, simple_loss=0.257, pruned_loss=0.06781, over 13342.00 frames. ], tot_loss[loss=0.1786, simple_loss=0.2394, pruned_loss=0.05893, over 2664818.83 frames. ], batch size: 118, lr: 6.21e-03, grad_scale: 16.0 2023-04-17 02:04:55,973 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55748.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:04:58,568 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-04-17 02:05:07,699 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.79 vs. limit=5.0 2023-04-17 02:05:25,985 INFO [train.py:893] (1/4) Epoch 21, batch 1800, loss[loss=0.1789, simple_loss=0.2415, pruned_loss=0.05817, over 13515.00 frames. ], tot_loss[loss=0.1783, simple_loss=0.239, pruned_loss=0.05879, over 2665927.42 frames. ], batch size: 98, lr: 6.20e-03, grad_scale: 16.0 2023-04-17 02:05:44,141 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.01 vs. limit=2.0 2023-04-17 02:05:46,217 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55804.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:05:50,434 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55809.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:06:00,652 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.879e+02 2.483e+02 2.815e+02 3.302e+02 6.184e+02, threshold=5.630e+02, percent-clipped=0.0 2023-04-17 02:06:12,448 INFO [train.py:893] (1/4) Epoch 21, batch 1850, loss[loss=0.2054, simple_loss=0.255, pruned_loss=0.07793, over 13484.00 frames. ], tot_loss[loss=0.1776, simple_loss=0.2381, pruned_loss=0.05852, over 2666644.41 frames. ], batch size: 93, lr: 6.20e-03, grad_scale: 16.0 2023-04-17 02:06:15,728 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-17 02:06:21,760 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.37 vs. limit=5.0 2023-04-17 02:06:26,658 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=55848.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:06:30,477 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55852.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:06:44,835 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=55867.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:06:58,108 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5647, 3.3140, 4.0716, 2.9211, 2.6437, 2.8410, 4.3554, 4.4558], device='cuda:1'), covar=tensor([0.1257, 0.1758, 0.0371, 0.1714, 0.1612, 0.1371, 0.0312, 0.0219], device='cuda:1'), in_proj_covar=tensor([0.0245, 0.0266, 0.0193, 0.0224, 0.0217, 0.0181, 0.0207, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-17 02:06:58,578 INFO [train.py:893] (1/4) Epoch 21, batch 1900, loss[loss=0.1843, simple_loss=0.243, pruned_loss=0.06277, over 13512.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.2375, pruned_loss=0.05866, over 2664619.70 frames. ], batch size: 81, lr: 6.20e-03, grad_scale: 16.0 2023-04-17 02:07:15,888 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2023-04-17 02:07:22,673 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-04-17 02:07:23,875 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=55909.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:07:29,531 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=55915.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:07:33,637 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.957e+02 2.684e+02 2.952e+02 3.656e+02 5.024e+02, threshold=5.905e+02, percent-clipped=0.0 2023-04-17 02:07:45,376 INFO [train.py:893] (1/4) Epoch 21, batch 1950, loss[loss=0.1927, simple_loss=0.248, pruned_loss=0.06868, over 13514.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.237, pruned_loss=0.05817, over 2666834.75 frames. ], batch size: 83, lr: 6.19e-03, grad_scale: 16.0 2023-04-17 02:08:08,212 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7612, 2.7619, 2.4630, 1.7035, 1.8719, 2.4130, 2.4615, 3.0314], device='cuda:1'), covar=tensor([0.1062, 0.0385, 0.0724, 0.1724, 0.0381, 0.0613, 0.0828, 0.0240], device='cuda:1'), in_proj_covar=tensor([0.0163, 0.0141, 0.0120, 0.0207, 0.0112, 0.0160, 0.0170, 0.0128], device='cuda:1'), out_proj_covar=tensor([1.2107e-04, 1.0545e-04, 9.3842e-05, 1.5431e-04, 8.2269e-05, 1.2108e-04, 1.2864e-04, 9.4864e-05], device='cuda:1') 2023-04-17 02:08:31,987 INFO [train.py:893] (1/4) Epoch 21, batch 2000, loss[loss=0.2207, simple_loss=0.2705, pruned_loss=0.08546, over 11808.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.239, pruned_loss=0.05894, over 2665757.84 frames. ], batch size: 157, lr: 6.19e-03, grad_scale: 16.0 2023-04-17 02:08:39,735 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-17 02:09:11,781 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5490, 3.3765, 4.0862, 2.8454, 2.6532, 2.7031, 4.4941, 4.5833], device='cuda:1'), covar=tensor([0.1233, 0.1761, 0.0429, 0.1900, 0.1788, 0.1628, 0.0257, 0.0210], device='cuda:1'), in_proj_covar=tensor([0.0247, 0.0267, 0.0194, 0.0226, 0.0219, 0.0182, 0.0209, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 02:09:13,048 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.874e+02 2.480e+02 2.958e+02 3.514e+02 5.155e+02, threshold=5.917e+02, percent-clipped=0.0 2023-04-17 02:09:23,723 INFO [train.py:893] (1/4) Epoch 21, batch 2050, loss[loss=0.1568, simple_loss=0.2203, pruned_loss=0.04668, over 13445.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.241, pruned_loss=0.06002, over 2659662.57 frames. ], batch size: 95, lr: 6.19e-03, grad_scale: 16.0 2023-04-17 02:10:09,198 INFO [train.py:893] (1/4) Epoch 21, batch 2100, loss[loss=0.1894, simple_loss=0.2563, pruned_loss=0.06121, over 13523.00 frames. ], tot_loss[loss=0.1801, simple_loss=0.2403, pruned_loss=0.05998, over 2657765.54 frames. ], batch size: 98, lr: 6.19e-03, grad_scale: 16.0 2023-04-17 02:10:29,015 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56104.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:10:41,443 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-04-17 02:10:43,842 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.062e+02 2.607e+02 3.023e+02 3.462e+02 7.013e+02, threshold=6.046e+02, percent-clipped=1.0 2023-04-17 02:10:54,790 INFO [train.py:893] (1/4) Epoch 21, batch 2150, loss[loss=0.1888, simple_loss=0.2526, pruned_loss=0.06251, over 13536.00 frames. ], tot_loss[loss=0.1795, simple_loss=0.2401, pruned_loss=0.05945, over 2657775.79 frames. ], batch size: 85, lr: 6.18e-03, grad_scale: 16.0 2023-04-17 02:11:06,126 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.6028, 2.6050, 2.5091, 1.5988, 1.5612, 2.2286, 2.1870, 2.8331], device='cuda:1'), covar=tensor([0.1091, 0.0324, 0.0502, 0.1760, 0.0247, 0.0622, 0.0914, 0.0214], device='cuda:1'), in_proj_covar=tensor([0.0164, 0.0143, 0.0120, 0.0209, 0.0113, 0.0162, 0.0172, 0.0129], device='cuda:1'), out_proj_covar=tensor([1.2183e-04, 1.0723e-04, 9.4344e-05, 1.5597e-04, 8.2741e-05, 1.2255e-04, 1.3029e-04, 9.5471e-05], device='cuda:1') 2023-04-17 02:11:06,861 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56145.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:11:07,964 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.82 vs. limit=2.0 2023-04-17 02:11:40,727 INFO [train.py:893] (1/4) Epoch 21, batch 2200, loss[loss=0.1857, simple_loss=0.2361, pruned_loss=0.06767, over 13495.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2383, pruned_loss=0.05859, over 2657033.15 frames. ], batch size: 70, lr: 6.18e-03, grad_scale: 16.0 2023-04-17 02:11:41,842 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2435, 2.7393, 2.2112, 4.1157, 4.6279, 3.5031, 4.5188, 4.2900], device='cuda:1'), covar=tensor([0.0092, 0.0891, 0.1041, 0.0103, 0.0075, 0.0426, 0.0086, 0.0095], device='cuda:1'), in_proj_covar=tensor([0.0076, 0.0090, 0.0097, 0.0081, 0.0067, 0.0081, 0.0056, 0.0077], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 02:12:01,650 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56204.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:12:03,503 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56206.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:12:15,681 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.840e+02 2.445e+02 3.046e+02 3.576e+02 5.421e+02, threshold=6.093e+02, percent-clipped=0.0 2023-04-17 02:12:26,666 INFO [train.py:893] (1/4) Epoch 21, batch 2250, loss[loss=0.1587, simple_loss=0.2165, pruned_loss=0.05045, over 13153.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2369, pruned_loss=0.05832, over 2658383.69 frames. ], batch size: 58, lr: 6.18e-03, grad_scale: 16.0 2023-04-17 02:12:45,594 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.6658, 2.2490, 2.1084, 1.3302, 1.8344, 1.9533, 2.0668, 2.4245], device='cuda:1'), covar=tensor([0.0720, 0.0254, 0.0575, 0.1480, 0.0194, 0.0432, 0.0552, 0.0198], device='cuda:1'), in_proj_covar=tensor([0.0166, 0.0143, 0.0121, 0.0210, 0.0114, 0.0163, 0.0172, 0.0129], device='cuda:1'), out_proj_covar=tensor([1.2261e-04, 1.0728e-04, 9.4940e-05, 1.5669e-04, 8.3544e-05, 1.2292e-04, 1.3045e-04, 9.5517e-05], device='cuda:1') 2023-04-17 02:13:12,795 INFO [train.py:893] (1/4) Epoch 21, batch 2300, loss[loss=0.1962, simple_loss=0.2567, pruned_loss=0.06785, over 13550.00 frames. ], tot_loss[loss=0.1758, simple_loss=0.2364, pruned_loss=0.05763, over 2658658.73 frames. ], batch size: 87, lr: 6.18e-03, grad_scale: 16.0 2023-04-17 02:13:48,866 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.697e+02 2.453e+02 2.876e+02 3.418e+02 5.342e+02, threshold=5.751e+02, percent-clipped=0.0 2023-04-17 02:13:59,954 INFO [train.py:893] (1/4) Epoch 21, batch 2350, loss[loss=0.1697, simple_loss=0.2289, pruned_loss=0.05521, over 13513.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2364, pruned_loss=0.05765, over 2661276.46 frames. ], batch size: 91, lr: 6.17e-03, grad_scale: 16.0 2023-04-17 02:14:21,861 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-17 02:14:45,690 INFO [train.py:893] (1/4) Epoch 21, batch 2400, loss[loss=0.1602, simple_loss=0.2226, pruned_loss=0.04888, over 13370.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2359, pruned_loss=0.05746, over 2662482.51 frames. ], batch size: 73, lr: 6.17e-03, grad_scale: 16.0 2023-04-17 02:15:06,333 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56404.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:15:06,474 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3556, 3.7009, 3.6018, 4.0612, 2.1494, 3.2241, 3.8899, 2.1830], device='cuda:1'), covar=tensor([0.0145, 0.0445, 0.0785, 0.0512, 0.1738, 0.0952, 0.0549, 0.1971], device='cuda:1'), in_proj_covar=tensor([0.0130, 0.0186, 0.0209, 0.0248, 0.0184, 0.0201, 0.0183, 0.0181], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 02:15:19,926 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.778e+02 2.549e+02 3.033e+02 3.703e+02 6.440e+02, threshold=6.067e+02, percent-clipped=2.0 2023-04-17 02:15:31,145 INFO [train.py:893] (1/4) Epoch 21, batch 2450, loss[loss=0.2092, simple_loss=0.2653, pruned_loss=0.07653, over 13362.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2363, pruned_loss=0.0578, over 2666541.01 frames. ], batch size: 109, lr: 6.17e-03, grad_scale: 16.0 2023-04-17 02:15:49,579 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56452.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:15:58,410 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2950, 3.5866, 3.5579, 3.9643, 2.2583, 3.0655, 3.7435, 2.1815], device='cuda:1'), covar=tensor([0.0138, 0.0529, 0.0760, 0.0545, 0.1595, 0.1057, 0.0655, 0.1859], device='cuda:1'), in_proj_covar=tensor([0.0131, 0.0186, 0.0210, 0.0249, 0.0184, 0.0202, 0.0183, 0.0181], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 02:16:05,304 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 2023-04-17 02:16:18,239 INFO [train.py:893] (1/4) Epoch 21, batch 2500, loss[loss=0.1856, simple_loss=0.2337, pruned_loss=0.06875, over 11794.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2361, pruned_loss=0.05782, over 2661577.75 frames. ], batch size: 157, lr: 6.16e-03, grad_scale: 16.0 2023-04-17 02:16:36,507 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56501.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:16:39,094 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56504.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:16:54,755 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.811e+02 2.337e+02 2.785e+02 3.508e+02 6.216e+02, threshold=5.571e+02, percent-clipped=2.0 2023-04-17 02:16:57,563 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56523.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 02:16:58,500 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.47 vs. limit=5.0 2023-04-17 02:17:05,612 INFO [train.py:893] (1/4) Epoch 21, batch 2550, loss[loss=0.1738, simple_loss=0.2398, pruned_loss=0.05392, over 13489.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2358, pruned_loss=0.05738, over 2665023.42 frames. ], batch size: 81, lr: 6.16e-03, grad_scale: 16.0 2023-04-17 02:17:24,986 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56552.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:17:28,212 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-17 02:17:39,173 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1685, 2.5323, 2.1470, 4.0321, 4.4817, 3.3455, 4.3566, 4.1962], device='cuda:1'), covar=tensor([0.0101, 0.0935, 0.1051, 0.0117, 0.0087, 0.0514, 0.0101, 0.0088], device='cuda:1'), in_proj_covar=tensor([0.0076, 0.0090, 0.0097, 0.0082, 0.0068, 0.0082, 0.0056, 0.0077], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 02:17:52,124 INFO [train.py:893] (1/4) Epoch 21, batch 2600, loss[loss=0.1865, simple_loss=0.2489, pruned_loss=0.0621, over 13415.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.236, pruned_loss=0.05775, over 2666430.92 frames. ], batch size: 95, lr: 6.16e-03, grad_scale: 16.0 2023-04-17 02:17:54,878 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56584.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 02:18:18,335 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.98 vs. limit=2.0 2023-04-17 02:18:25,356 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.056e+02 2.625e+02 3.073e+02 3.881e+02 8.631e+02, threshold=6.146e+02, percent-clipped=4.0 2023-04-17 02:18:34,427 INFO [train.py:893] (1/4) Epoch 21, batch 2650, loss[loss=0.1753, simple_loss=0.2365, pruned_loss=0.0571, over 13553.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2365, pruned_loss=0.05807, over 2664454.44 frames. ], batch size: 87, lr: 6.16e-03, grad_scale: 16.0 2023-04-17 02:19:33,259 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-17 02:19:41,993 INFO [train.py:893] (1/4) Epoch 22, batch 0, loss[loss=0.1788, simple_loss=0.2361, pruned_loss=0.06073, over 13518.00 frames. ], tot_loss[loss=0.1788, simple_loss=0.2361, pruned_loss=0.06073, over 13518.00 frames. ], batch size: 91, lr: 6.01e-03, grad_scale: 16.0 2023-04-17 02:19:41,994 INFO [train.py:918] (1/4) Computing validation loss 2023-04-17 02:20:04,055 INFO [train.py:927] (1/4) Epoch 22, validation: loss=0.136, simple_loss=0.1958, pruned_loss=0.03815, over 2446609.00 frames. 2023-04-17 02:20:04,056 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12831MB 2023-04-17 02:20:24,525 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.3844, 4.8885, 4.8409, 4.9062, 4.7171, 4.7299, 5.3691, 4.9055], device='cuda:1'), covar=tensor([0.0732, 0.1287, 0.2195, 0.2728, 0.1024, 0.1626, 0.0985, 0.1319], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0376, 0.0469, 0.0476, 0.0293, 0.0349, 0.0435, 0.0345], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 02:20:40,793 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.028e+02 2.714e+02 3.125e+02 3.595e+02 7.185e+02, threshold=6.251e+02, percent-clipped=1.0 2023-04-17 02:20:52,110 INFO [train.py:893] (1/4) Epoch 22, batch 50, loss[loss=0.1547, simple_loss=0.2186, pruned_loss=0.04541, over 13541.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2323, pruned_loss=0.0575, over 599642.96 frames. ], batch size: 85, lr: 6.01e-03, grad_scale: 16.0 2023-04-17 02:21:14,868 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6530, 3.3388, 4.1548, 2.9447, 2.7523, 2.8633, 4.4009, 4.4622], device='cuda:1'), covar=tensor([0.1226, 0.1738, 0.0373, 0.1753, 0.1526, 0.1553, 0.0295, 0.0283], device='cuda:1'), in_proj_covar=tensor([0.0245, 0.0265, 0.0194, 0.0223, 0.0216, 0.0181, 0.0208, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-17 02:21:15,345 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-17 02:21:15,346 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-17 02:21:15,346 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-17 02:21:15,355 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-17 02:21:15,363 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-17 02:21:15,384 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-17 02:21:15,402 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-17 02:21:21,676 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-04-17 02:21:37,889 INFO [train.py:893] (1/4) Epoch 22, batch 100, loss[loss=0.1655, simple_loss=0.2322, pruned_loss=0.04941, over 13522.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2338, pruned_loss=0.05841, over 1060306.70 frames. ], batch size: 98, lr: 6.00e-03, grad_scale: 16.0 2023-04-17 02:21:58,569 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=56801.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:22:00,234 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56803.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:22:07,044 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5409, 2.2980, 2.7746, 3.9663, 3.5813, 4.0319, 3.1952, 2.5583], device='cuda:1'), covar=tensor([0.0251, 0.0921, 0.0782, 0.0069, 0.0235, 0.0054, 0.0612, 0.0844], device='cuda:1'), in_proj_covar=tensor([0.0118, 0.0149, 0.0167, 0.0099, 0.0119, 0.0096, 0.0167, 0.0146], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 02:22:15,149 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.959e+02 2.858e+02 3.303e+02 3.890e+02 1.630e+03, threshold=6.606e+02, percent-clipped=4.0 2023-04-17 02:22:26,756 INFO [train.py:893] (1/4) Epoch 22, batch 150, loss[loss=0.1804, simple_loss=0.2351, pruned_loss=0.0628, over 11898.00 frames. ], tot_loss[loss=0.1788, simple_loss=0.237, pruned_loss=0.06033, over 1411646.65 frames. ], batch size: 157, lr: 6.00e-03, grad_scale: 16.0 2023-04-17 02:22:42,780 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=56849.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:22:57,962 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56864.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:23:04,784 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5299, 2.4525, 2.6935, 4.0100, 3.6302, 4.0682, 3.1548, 2.5806], device='cuda:1'), covar=tensor([0.0244, 0.0906, 0.0840, 0.0058, 0.0241, 0.0054, 0.0647, 0.0868], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0151, 0.0168, 0.0100, 0.0120, 0.0096, 0.0168, 0.0147], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 02:23:11,271 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=56879.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 02:23:12,698 INFO [train.py:893] (1/4) Epoch 22, batch 200, loss[loss=0.1552, simple_loss=0.217, pruned_loss=0.04671, over 13529.00 frames. ], tot_loss[loss=0.1801, simple_loss=0.2382, pruned_loss=0.06105, over 1681482.02 frames. ], batch size: 72, lr: 6.00e-03, grad_scale: 16.0 2023-04-17 02:23:28,659 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.89 vs. limit=5.0 2023-04-17 02:23:35,995 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.8273, 4.9458, 4.7898, 4.7814, 4.8610, 5.1472, 4.9791, 4.7923], device='cuda:1'), covar=tensor([0.0245, 0.0232, 0.0230, 0.0764, 0.0224, 0.0200, 0.0218, 0.0182], device='cuda:1'), in_proj_covar=tensor([0.0160, 0.0157, 0.0174, 0.0263, 0.0173, 0.0191, 0.0171, 0.0110], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 02:23:38,641 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56908.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:23:44,008 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.82 vs. limit=5.0 2023-04-17 02:23:49,167 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.056e+02 2.667e+02 3.208e+02 3.785e+02 6.289e+02, threshold=6.417e+02, percent-clipped=0.0 2023-04-17 02:24:00,157 INFO [train.py:893] (1/4) Epoch 22, batch 250, loss[loss=0.1654, simple_loss=0.2196, pruned_loss=0.05564, over 13387.00 frames. ], tot_loss[loss=0.1806, simple_loss=0.2386, pruned_loss=0.06126, over 1891463.83 frames. ], batch size: 62, lr: 6.00e-03, grad_scale: 16.0 2023-04-17 02:24:34,445 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=56969.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:24:44,804 INFO [train.py:893] (1/4) Epoch 22, batch 300, loss[loss=0.1823, simple_loss=0.2435, pruned_loss=0.06056, over 13489.00 frames. ], tot_loss[loss=0.1799, simple_loss=0.2387, pruned_loss=0.06053, over 2064504.06 frames. ], batch size: 81, lr: 5.99e-03, grad_scale: 16.0 2023-04-17 02:24:56,525 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56993.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:24:58,185 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=56995.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:25:22,195 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.865e+02 2.638e+02 3.070e+02 3.794e+02 6.300e+02, threshold=6.141e+02, percent-clipped=0.0 2023-04-17 02:25:31,433 INFO [train.py:893] (1/4) Epoch 22, batch 350, loss[loss=0.1474, simple_loss=0.2114, pruned_loss=0.04168, over 13349.00 frames. ], tot_loss[loss=0.1801, simple_loss=0.2389, pruned_loss=0.06067, over 2199471.50 frames. ], batch size: 73, lr: 5.99e-03, grad_scale: 16.0 2023-04-17 02:25:35,002 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0197, 3.8559, 3.9062, 2.3242, 4.2372, 4.0693, 4.0389, 4.2569], device='cuda:1'), covar=tensor([0.0220, 0.0143, 0.0145, 0.1235, 0.0131, 0.0228, 0.0119, 0.0106], device='cuda:1'), in_proj_covar=tensor([0.0093, 0.0056, 0.0083, 0.0104, 0.0100, 0.0109, 0.0081, 0.0077], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 02:25:53,972 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57054.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:25:55,631 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57056.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:26:18,063 INFO [train.py:893] (1/4) Epoch 22, batch 400, loss[loss=0.2052, simple_loss=0.2621, pruned_loss=0.07412, over 13534.00 frames. ], tot_loss[loss=0.1803, simple_loss=0.2391, pruned_loss=0.06075, over 2302716.65 frames. ], batch size: 83, lr: 5.99e-03, grad_scale: 16.0 2023-04-17 02:26:22,141 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8110, 3.9537, 3.0476, 2.6900, 2.8300, 2.4318, 4.0820, 2.3183], device='cuda:1'), covar=tensor([0.1807, 0.0413, 0.1298, 0.2317, 0.0919, 0.3428, 0.0281, 0.4123], device='cuda:1'), in_proj_covar=tensor([0.0315, 0.0290, 0.0316, 0.0332, 0.0260, 0.0329, 0.0213, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 02:26:56,971 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.094e+02 2.625e+02 3.015e+02 3.405e+02 5.837e+02, threshold=6.030e+02, percent-clipped=0.0 2023-04-17 02:27:06,203 INFO [train.py:893] (1/4) Epoch 22, batch 450, loss[loss=0.1916, simple_loss=0.2519, pruned_loss=0.06566, over 13268.00 frames. ], tot_loss[loss=0.1818, simple_loss=0.2409, pruned_loss=0.06133, over 2384824.40 frames. ], batch size: 124, lr: 5.99e-03, grad_scale: 16.0 2023-04-17 02:27:10,004 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8364, 4.1926, 4.0205, 4.0032, 4.0887, 3.9237, 4.2681, 4.2712], device='cuda:1'), covar=tensor([0.0258, 0.0245, 0.0246, 0.0346, 0.0280, 0.0253, 0.0245, 0.0268], device='cuda:1'), in_proj_covar=tensor([0.0136, 0.0196, 0.0162, 0.0178, 0.0150, 0.0194, 0.0134, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 02:27:33,692 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57159.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:27:34,322 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-17 02:27:54,306 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57179.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 02:27:54,630 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.56 vs. limit=5.0 2023-04-17 02:27:55,699 INFO [train.py:893] (1/4) Epoch 22, batch 500, loss[loss=0.1531, simple_loss=0.2098, pruned_loss=0.04821, over 13138.00 frames. ], tot_loss[loss=0.182, simple_loss=0.2412, pruned_loss=0.06139, over 2444745.09 frames. ], batch size: 58, lr: 5.98e-03, grad_scale: 32.0 2023-04-17 02:28:00,255 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57186.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:28:10,440 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57197.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:28:25,189 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.3440, 4.8568, 4.8364, 4.8438, 4.6357, 4.6583, 5.3304, 4.8634], device='cuda:1'), covar=tensor([0.0593, 0.1057, 0.1781, 0.2328, 0.0965, 0.1509, 0.0751, 0.1137], device='cuda:1'), in_proj_covar=tensor([0.0336, 0.0380, 0.0472, 0.0478, 0.0298, 0.0350, 0.0435, 0.0350], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 02:28:32,507 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.660e+02 2.678e+02 3.189e+02 3.924e+02 6.085e+02, threshold=6.377e+02, percent-clipped=1.0 2023-04-17 02:28:39,549 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57227.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 02:28:42,451 INFO [train.py:893] (1/4) Epoch 22, batch 550, loss[loss=0.166, simple_loss=0.2226, pruned_loss=0.05475, over 13340.00 frames. ], tot_loss[loss=0.1818, simple_loss=0.2413, pruned_loss=0.06115, over 2489149.43 frames. ], batch size: 67, lr: 5.98e-03, grad_scale: 32.0 2023-04-17 02:28:56,670 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57244.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:28:59,151 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57247.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:29:09,585 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57258.0, num_to_drop=1, layers_to_drop={2} 2023-04-17 02:29:14,648 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57264.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:29:30,574 INFO [train.py:893] (1/4) Epoch 22, batch 600, loss[loss=0.1853, simple_loss=0.2388, pruned_loss=0.06591, over 13377.00 frames. ], tot_loss[loss=0.1801, simple_loss=0.2392, pruned_loss=0.06046, over 2527256.75 frames. ], batch size: 118, lr: 5.98e-03, grad_scale: 32.0 2023-04-17 02:29:54,877 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57305.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:30:07,436 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.837e+02 2.577e+02 3.083e+02 3.651e+02 6.901e+02, threshold=6.166e+02, percent-clipped=2.0 2023-04-17 02:30:18,844 INFO [train.py:893] (1/4) Epoch 22, batch 650, loss[loss=0.1827, simple_loss=0.2427, pruned_loss=0.06133, over 13482.00 frames. ], tot_loss[loss=0.1804, simple_loss=0.2394, pruned_loss=0.06067, over 2546322.13 frames. ], batch size: 81, lr: 5.98e-03, grad_scale: 32.0 2023-04-17 02:30:35,140 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57349.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:30:36,859 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57351.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:30:52,809 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4257, 2.5123, 2.6755, 3.9021, 3.5369, 3.9693, 3.0540, 2.5848], device='cuda:1'), covar=tensor([0.0326, 0.0885, 0.0826, 0.0064, 0.0272, 0.0060, 0.0650, 0.0839], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0152, 0.0169, 0.0101, 0.0122, 0.0098, 0.0171, 0.0149], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 02:31:05,286 INFO [train.py:893] (1/4) Epoch 22, batch 700, loss[loss=0.1816, simple_loss=0.2347, pruned_loss=0.06427, over 13446.00 frames. ], tot_loss[loss=0.1788, simple_loss=0.2383, pruned_loss=0.05963, over 2575102.02 frames. ], batch size: 65, lr: 5.97e-03, grad_scale: 32.0 2023-04-17 02:31:22,433 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57398.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:31:24,574 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2269, 4.3432, 2.9980, 3.9706, 4.2103, 2.7705, 3.8270, 3.0058], device='cuda:1'), covar=tensor([0.0275, 0.0203, 0.1015, 0.0389, 0.0236, 0.1167, 0.0480, 0.1145], device='cuda:1'), in_proj_covar=tensor([0.0160, 0.0175, 0.0177, 0.0217, 0.0139, 0.0159, 0.0162, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 02:31:43,046 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.065e+02 2.787e+02 3.172e+02 3.577e+02 6.833e+02, threshold=6.345e+02, percent-clipped=1.0 2023-04-17 02:31:53,817 INFO [train.py:893] (1/4) Epoch 22, batch 750, loss[loss=0.1835, simple_loss=0.2435, pruned_loss=0.06172, over 13540.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.2376, pruned_loss=0.05959, over 2594291.00 frames. ], batch size: 87, lr: 5.97e-03, grad_scale: 32.0 2023-04-17 02:32:05,977 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.4678, 4.9820, 4.8904, 4.9474, 4.7686, 4.8292, 5.4394, 4.9981], device='cuda:1'), covar=tensor([0.0742, 0.1265, 0.2055, 0.2471, 0.0998, 0.1503, 0.0962, 0.1322], device='cuda:1'), in_proj_covar=tensor([0.0333, 0.0379, 0.0473, 0.0475, 0.0295, 0.0349, 0.0435, 0.0349], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 02:32:17,217 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0837, 2.7174, 2.7818, 3.0373, 2.4718, 3.1385, 3.0706, 2.6671], device='cuda:1'), covar=tensor([0.0085, 0.0199, 0.0149, 0.0200, 0.0205, 0.0116, 0.0156, 0.0157], device='cuda:1'), in_proj_covar=tensor([0.0106, 0.0116, 0.0123, 0.0122, 0.0131, 0.0108, 0.0107, 0.0105], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 02:32:21,060 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57459.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:32:21,131 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57459.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:32:23,661 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57462.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:32:41,327 INFO [train.py:893] (1/4) Epoch 22, batch 800, loss[loss=0.1879, simple_loss=0.2456, pruned_loss=0.06511, over 13531.00 frames. ], tot_loss[loss=0.1792, simple_loss=0.2387, pruned_loss=0.05984, over 2613116.38 frames. ], batch size: 85, lr: 5.97e-03, grad_scale: 32.0 2023-04-17 02:32:50,672 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=2.65 vs. limit=5.0 2023-04-17 02:32:54,703 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57495.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:33:04,183 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3259, 2.3916, 2.8115, 3.8226, 3.4693, 3.8829, 3.0258, 2.5214], device='cuda:1'), covar=tensor([0.0315, 0.0749, 0.0645, 0.0060, 0.0252, 0.0060, 0.0547, 0.0774], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0151, 0.0168, 0.0101, 0.0121, 0.0098, 0.0169, 0.0148], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 02:33:05,894 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57507.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:33:18,175 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.840e+02 2.750e+02 3.203e+02 3.790e+02 7.291e+02, threshold=6.405e+02, percent-clipped=0.0 2023-04-17 02:33:21,140 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57523.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:33:28,397 INFO [train.py:893] (1/4) Epoch 22, batch 850, loss[loss=0.1595, simple_loss=0.2226, pruned_loss=0.04818, over 13476.00 frames. ], tot_loss[loss=0.1797, simple_loss=0.2393, pruned_loss=0.06006, over 2622929.27 frames. ], batch size: 79, lr: 5.97e-03, grad_scale: 32.0 2023-04-17 02:33:38,760 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57542.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:33:49,704 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57553.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 02:33:52,274 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57556.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:33:59,861 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57564.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:34:15,654 INFO [train.py:893] (1/4) Epoch 22, batch 900, loss[loss=0.2057, simple_loss=0.2482, pruned_loss=0.08162, over 11983.00 frames. ], tot_loss[loss=0.1805, simple_loss=0.2396, pruned_loss=0.06066, over 2628431.55 frames. ], batch size: 157, lr: 5.96e-03, grad_scale: 32.0 2023-04-17 02:34:33,473 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57600.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:34:45,142 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57612.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:34:48,275 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-17 02:34:52,399 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.880e+02 2.743e+02 3.191e+02 3.873e+02 7.153e+02, threshold=6.382e+02, percent-clipped=2.0 2023-04-17 02:35:02,807 INFO [train.py:893] (1/4) Epoch 22, batch 950, loss[loss=0.1665, simple_loss=0.2254, pruned_loss=0.05378, over 13514.00 frames. ], tot_loss[loss=0.1796, simple_loss=0.2385, pruned_loss=0.06033, over 2635137.09 frames. ], batch size: 76, lr: 5.96e-03, grad_scale: 32.0 2023-04-17 02:35:20,505 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57649.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:35:22,051 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57651.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:35:37,191 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-04-17 02:35:41,291 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7424, 3.5028, 4.2995, 2.9862, 2.8761, 2.8868, 4.6043, 4.7002], device='cuda:1'), covar=tensor([0.1215, 0.1714, 0.0315, 0.1762, 0.1524, 0.1537, 0.0246, 0.0228], device='cuda:1'), in_proj_covar=tensor([0.0250, 0.0269, 0.0196, 0.0228, 0.0219, 0.0184, 0.0212, 0.0182], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 02:35:50,157 INFO [train.py:893] (1/4) Epoch 22, batch 1000, loss[loss=0.1791, simple_loss=0.2376, pruned_loss=0.06028, over 13544.00 frames. ], tot_loss[loss=0.1786, simple_loss=0.2371, pruned_loss=0.06004, over 2640327.30 frames. ], batch size: 87, lr: 5.96e-03, grad_scale: 32.0 2023-04-17 02:36:05,536 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57697.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:36:07,227 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57699.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:36:25,800 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.668e+02 2.772e+02 3.188e+02 3.825e+02 8.358e+02, threshold=6.376e+02, percent-clipped=3.0 2023-04-17 02:36:36,824 INFO [train.py:893] (1/4) Epoch 22, batch 1050, loss[loss=0.1782, simple_loss=0.2357, pruned_loss=0.06033, over 13347.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2356, pruned_loss=0.05893, over 2643403.49 frames. ], batch size: 73, lr: 5.96e-03, grad_scale: 32.0 2023-04-17 02:36:38,952 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=57733.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:36:58,848 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57754.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:37:23,701 INFO [train.py:893] (1/4) Epoch 22, batch 1100, loss[loss=0.1619, simple_loss=0.224, pruned_loss=0.04986, over 13552.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.2361, pruned_loss=0.05876, over 2652235.24 frames. ], batch size: 87, lr: 5.95e-03, grad_scale: 32.0 2023-04-17 02:37:36,355 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=57794.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:37:58,288 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57818.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:38:00,595 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.815e+02 2.427e+02 2.790e+02 3.332e+02 6.783e+02, threshold=5.579e+02, percent-clipped=1.0 2023-04-17 02:38:10,813 INFO [train.py:893] (1/4) Epoch 22, batch 1150, loss[loss=0.1572, simple_loss=0.2298, pruned_loss=0.04231, over 13482.00 frames. ], tot_loss[loss=0.1765, simple_loss=0.2363, pruned_loss=0.05835, over 2654608.91 frames. ], batch size: 81, lr: 5.95e-03, grad_scale: 32.0 2023-04-17 02:38:21,777 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57842.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:38:30,236 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=57851.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:38:31,985 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57853.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 02:38:57,359 INFO [train.py:893] (1/4) Epoch 22, batch 1200, loss[loss=0.201, simple_loss=0.271, pruned_loss=0.06551, over 13522.00 frames. ], tot_loss[loss=0.176, simple_loss=0.236, pruned_loss=0.05802, over 2651259.69 frames. ], batch size: 98, lr: 5.95e-03, grad_scale: 32.0 2023-04-17 02:39:06,938 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57890.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:39:13,503 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-04-17 02:39:17,322 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=57900.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:39:18,047 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57901.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:39:27,340 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-17 02:39:35,634 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.743e+02 2.547e+02 2.982e+02 3.467e+02 6.572e+02, threshold=5.963e+02, percent-clipped=2.0 2023-04-17 02:39:39,044 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-17 02:39:46,595 INFO [train.py:893] (1/4) Epoch 22, batch 1250, loss[loss=0.1568, simple_loss=0.2182, pruned_loss=0.04769, over 13513.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.237, pruned_loss=0.05865, over 2657151.01 frames. ], batch size: 76, lr: 5.94e-03, grad_scale: 32.0 2023-04-17 02:40:01,122 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0360, 4.2055, 3.2441, 2.8888, 2.9949, 2.5664, 4.2825, 2.4717], device='cuda:1'), covar=tensor([0.1677, 0.0303, 0.1165, 0.2023, 0.0852, 0.3138, 0.0249, 0.3874], device='cuda:1'), in_proj_covar=tensor([0.0316, 0.0289, 0.0315, 0.0332, 0.0260, 0.0329, 0.0214, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 02:40:01,703 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=57948.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:40:32,336 INFO [train.py:893] (1/4) Epoch 22, batch 1300, loss[loss=0.1831, simple_loss=0.2356, pruned_loss=0.06529, over 13347.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.2382, pruned_loss=0.05929, over 2654964.34 frames. ], batch size: 67, lr: 5.94e-03, grad_scale: 32.0 2023-04-17 02:41:14,238 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.879e+02 2.746e+02 3.180e+02 3.759e+02 5.713e+02, threshold=6.361e+02, percent-clipped=0.0 2023-04-17 02:41:15,868 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2023-04-17 02:41:24,146 INFO [train.py:893] (1/4) Epoch 22, batch 1350, loss[loss=0.1272, simple_loss=0.1804, pruned_loss=0.03702, over 12639.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2377, pruned_loss=0.05905, over 2652108.29 frames. ], batch size: 51, lr: 5.94e-03, grad_scale: 32.0 2023-04-17 02:41:47,134 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58054.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:42:12,097 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5244, 2.5858, 2.8063, 3.9922, 3.5987, 4.0200, 3.0658, 2.6033], device='cuda:1'), covar=tensor([0.0284, 0.0904, 0.0825, 0.0065, 0.0261, 0.0070, 0.0730, 0.0901], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0153, 0.0168, 0.0101, 0.0120, 0.0098, 0.0169, 0.0149], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 02:42:12,624 INFO [train.py:893] (1/4) Epoch 22, batch 1400, loss[loss=0.1725, simple_loss=0.2397, pruned_loss=0.05263, over 13536.00 frames. ], tot_loss[loss=0.1778, simple_loss=0.2377, pruned_loss=0.05896, over 2650958.85 frames. ], batch size: 83, lr: 5.94e-03, grad_scale: 32.0 2023-04-17 02:42:20,208 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58089.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:42:32,153 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58102.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:42:48,242 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58118.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:42:49,590 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.993e+02 2.607e+02 2.970e+02 3.475e+02 7.566e+02, threshold=5.940e+02, percent-clipped=1.0 2023-04-17 02:42:58,875 INFO [train.py:893] (1/4) Epoch 22, batch 1450, loss[loss=0.1894, simple_loss=0.2487, pruned_loss=0.06507, over 13547.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2369, pruned_loss=0.05878, over 2654590.31 frames. ], batch size: 72, lr: 5.93e-03, grad_scale: 32.0 2023-04-17 02:43:10,133 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58141.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:43:12,676 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58144.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:43:19,040 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58151.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:43:32,643 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58166.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:43:47,422 INFO [train.py:893] (1/4) Epoch 22, batch 1500, loss[loss=0.1765, simple_loss=0.2394, pruned_loss=0.0568, over 13031.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2368, pruned_loss=0.05857, over 2651748.62 frames. ], batch size: 142, lr: 5.93e-03, grad_scale: 32.0 2023-04-17 02:44:04,460 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58199.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:44:08,255 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58202.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:44:10,939 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58205.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:44:19,237 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0089, 2.1124, 2.3708, 3.2816, 3.0248, 3.3068, 2.7597, 2.2774], device='cuda:1'), covar=tensor([0.0278, 0.0730, 0.0711, 0.0096, 0.0271, 0.0078, 0.0512, 0.0796], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0153, 0.0170, 0.0102, 0.0122, 0.0099, 0.0170, 0.0151], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 02:44:23,978 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.863e+02 2.687e+02 3.148e+02 3.819e+02 8.002e+02, threshold=6.297e+02, percent-clipped=6.0 2023-04-17 02:44:34,889 INFO [train.py:893] (1/4) Epoch 22, batch 1550, loss[loss=0.142, simple_loss=0.2023, pruned_loss=0.04088, over 13424.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2366, pruned_loss=0.05833, over 2650104.08 frames. ], batch size: 65, lr: 5.93e-03, grad_scale: 32.0 2023-04-17 02:44:40,380 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1679, 4.0180, 4.1233, 2.5827, 4.4395, 4.2094, 4.2000, 4.4287], device='cuda:1'), covar=tensor([0.0272, 0.0173, 0.0154, 0.1253, 0.0163, 0.0249, 0.0171, 0.0118], device='cuda:1'), in_proj_covar=tensor([0.0094, 0.0057, 0.0085, 0.0104, 0.0100, 0.0110, 0.0082, 0.0077], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 02:45:11,810 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.3045, 4.8283, 4.7299, 4.7985, 4.6066, 4.7013, 5.3287, 4.8938], device='cuda:1'), covar=tensor([0.0703, 0.1236, 0.1992, 0.2448, 0.0917, 0.1399, 0.0806, 0.1033], device='cuda:1'), in_proj_covar=tensor([0.0332, 0.0380, 0.0468, 0.0475, 0.0293, 0.0350, 0.0434, 0.0345], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 02:45:21,745 INFO [train.py:893] (1/4) Epoch 22, batch 1600, loss[loss=0.1772, simple_loss=0.2364, pruned_loss=0.059, over 13030.00 frames. ], tot_loss[loss=0.1779, simple_loss=0.2378, pruned_loss=0.05898, over 2649938.35 frames. ], batch size: 142, lr: 5.93e-03, grad_scale: 32.0 2023-04-17 02:45:22,422 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.78 vs. limit=5.0 2023-04-17 02:45:39,229 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3105, 3.0119, 3.6044, 2.6857, 2.4451, 2.6590, 3.9848, 4.0561], device='cuda:1'), covar=tensor([0.1240, 0.1995, 0.0413, 0.1709, 0.1760, 0.1552, 0.0351, 0.0239], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0270, 0.0197, 0.0228, 0.0220, 0.0185, 0.0212, 0.0183], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 02:45:57,052 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.013e+02 2.646e+02 2.984e+02 3.538e+02 5.266e+02, threshold=5.968e+02, percent-clipped=0.0 2023-04-17 02:46:08,722 INFO [train.py:893] (1/4) Epoch 22, batch 1650, loss[loss=0.1906, simple_loss=0.257, pruned_loss=0.06214, over 13449.00 frames. ], tot_loss[loss=0.1774, simple_loss=0.238, pruned_loss=0.05836, over 2655721.76 frames. ], batch size: 100, lr: 5.92e-03, grad_scale: 32.0 2023-04-17 02:46:38,864 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1366, 4.3184, 3.3004, 2.9125, 3.1457, 2.5826, 4.5302, 2.4951], device='cuda:1'), covar=tensor([0.1771, 0.0351, 0.1276, 0.2262, 0.0881, 0.3558, 0.0243, 0.4263], device='cuda:1'), in_proj_covar=tensor([0.0315, 0.0289, 0.0314, 0.0331, 0.0259, 0.0327, 0.0212, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 02:46:55,369 INFO [train.py:893] (1/4) Epoch 22, batch 1700, loss[loss=0.1905, simple_loss=0.2563, pruned_loss=0.06234, over 13347.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2376, pruned_loss=0.05779, over 2655942.59 frames. ], batch size: 118, lr: 5.92e-03, grad_scale: 32.0 2023-04-17 02:47:04,896 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58389.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:47:34,541 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.954e+02 2.578e+02 3.029e+02 3.680e+02 6.898e+02, threshold=6.059e+02, percent-clipped=3.0 2023-04-17 02:47:44,711 INFO [train.py:893] (1/4) Epoch 22, batch 1750, loss[loss=0.1722, simple_loss=0.2357, pruned_loss=0.05431, over 13479.00 frames. ], tot_loss[loss=0.1746, simple_loss=0.2359, pruned_loss=0.05658, over 2659774.26 frames. ], batch size: 93, lr: 5.92e-03, grad_scale: 32.0 2023-04-17 02:47:49,791 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58437.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:48:05,669 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1895, 2.8857, 3.5879, 2.7479, 2.4316, 2.6065, 3.9193, 3.9468], device='cuda:1'), covar=tensor([0.1377, 0.2066, 0.0389, 0.1627, 0.1678, 0.1539, 0.0274, 0.0305], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0268, 0.0195, 0.0224, 0.0217, 0.0183, 0.0210, 0.0181], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 02:48:33,554 INFO [train.py:893] (1/4) Epoch 22, batch 1800, loss[loss=0.1724, simple_loss=0.2364, pruned_loss=0.05419, over 13490.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2359, pruned_loss=0.05675, over 2663133.61 frames. ], batch size: 110, lr: 5.92e-03, grad_scale: 32.0 2023-04-17 02:48:45,481 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7896, 2.4907, 2.5270, 2.8654, 2.2063, 2.9381, 2.8870, 2.3881], device='cuda:1'), covar=tensor([0.0097, 0.0184, 0.0150, 0.0147, 0.0232, 0.0105, 0.0154, 0.0169], device='cuda:1'), in_proj_covar=tensor([0.0106, 0.0115, 0.0124, 0.0122, 0.0133, 0.0109, 0.0108, 0.0107], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 02:48:47,828 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58497.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:48:50,277 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=58500.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:49:08,653 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.958e+02 2.574e+02 2.968e+02 3.492e+02 5.400e+02, threshold=5.936e+02, percent-clipped=0.0 2023-04-17 02:49:18,718 INFO [train.py:893] (1/4) Epoch 22, batch 1850, loss[loss=0.143, simple_loss=0.209, pruned_loss=0.0385, over 13509.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.235, pruned_loss=0.05621, over 2659197.30 frames. ], batch size: 76, lr: 5.91e-03, grad_scale: 32.0 2023-04-17 02:49:21,272 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-17 02:49:49,418 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1391, 2.8382, 2.8326, 3.0721, 2.3432, 3.2386, 3.0832, 2.7387], device='cuda:1'), covar=tensor([0.0093, 0.0148, 0.0107, 0.0164, 0.0206, 0.0105, 0.0154, 0.0132], device='cuda:1'), in_proj_covar=tensor([0.0105, 0.0115, 0.0123, 0.0121, 0.0132, 0.0109, 0.0107, 0.0106], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 02:50:05,932 INFO [train.py:893] (1/4) Epoch 22, batch 1900, loss[loss=0.1696, simple_loss=0.2279, pruned_loss=0.05565, over 13129.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2346, pruned_loss=0.05636, over 2660213.04 frames. ], batch size: 142, lr: 5.91e-03, grad_scale: 32.0 2023-04-17 02:50:42,961 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.656e+02 2.493e+02 2.844e+02 3.457e+02 5.617e+02, threshold=5.688e+02, percent-clipped=0.0 2023-04-17 02:50:52,233 INFO [train.py:893] (1/4) Epoch 22, batch 1950, loss[loss=0.2216, simple_loss=0.2751, pruned_loss=0.08402, over 13579.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.235, pruned_loss=0.05691, over 2661288.17 frames. ], batch size: 89, lr: 5.91e-03, grad_scale: 32.0 2023-04-17 02:51:23,152 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.03 vs. limit=2.0 2023-04-17 02:51:33,395 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.74 vs. limit=5.0 2023-04-17 02:51:40,608 INFO [train.py:893] (1/4) Epoch 22, batch 2000, loss[loss=0.2051, simple_loss=0.2679, pruned_loss=0.07117, over 13493.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2368, pruned_loss=0.05749, over 2659095.78 frames. ], batch size: 93, lr: 5.91e-03, grad_scale: 32.0 2023-04-17 02:51:40,956 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58681.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:51:43,253 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.4249, 2.1984, 2.1894, 2.4911, 1.8171, 2.4605, 2.4330, 2.0538], device='cuda:1'), covar=tensor([0.0093, 0.0200, 0.0148, 0.0132, 0.0246, 0.0129, 0.0157, 0.0180], device='cuda:1'), in_proj_covar=tensor([0.0105, 0.0115, 0.0124, 0.0122, 0.0133, 0.0109, 0.0108, 0.0106], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 02:51:46,241 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-17 02:52:17,354 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.815e+02 2.571e+02 3.060e+02 3.640e+02 6.569e+02, threshold=6.120e+02, percent-clipped=2.0 2023-04-17 02:52:28,166 INFO [train.py:893] (1/4) Epoch 22, batch 2050, loss[loss=0.165, simple_loss=0.2282, pruned_loss=0.05092, over 13494.00 frames. ], tot_loss[loss=0.1786, simple_loss=0.2393, pruned_loss=0.05898, over 2658532.69 frames. ], batch size: 70, lr: 5.90e-03, grad_scale: 32.0 2023-04-17 02:52:39,446 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58742.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:53:03,160 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-17 02:53:10,153 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58775.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:53:14,977 INFO [train.py:893] (1/4) Epoch 22, batch 2100, loss[loss=0.1634, simple_loss=0.228, pruned_loss=0.04946, over 13478.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2379, pruned_loss=0.05804, over 2663132.70 frames. ], batch size: 81, lr: 5.90e-03, grad_scale: 32.0 2023-04-17 02:53:31,526 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58797.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:53:34,466 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=58800.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:53:45,519 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-04-17 02:53:51,732 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.861e+02 2.599e+02 3.041e+02 3.850e+02 9.204e+02, threshold=6.081e+02, percent-clipped=2.0 2023-04-17 02:54:03,793 INFO [train.py:893] (1/4) Epoch 22, batch 2150, loss[loss=0.1462, simple_loss=0.2146, pruned_loss=0.0389, over 13365.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2373, pruned_loss=0.05734, over 2659614.95 frames. ], batch size: 67, lr: 5.90e-03, grad_scale: 32.0 2023-04-17 02:54:08,905 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58836.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:54:16,267 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58845.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:54:18,772 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=58848.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:54:42,736 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7327, 3.5750, 4.3022, 2.9936, 2.9431, 2.9506, 4.6072, 4.6514], device='cuda:1'), covar=tensor([0.1158, 0.1550, 0.0317, 0.1640, 0.1408, 0.1512, 0.0277, 0.0242], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0272, 0.0198, 0.0228, 0.0220, 0.0183, 0.0214, 0.0182], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 02:54:49,087 INFO [train.py:893] (1/4) Epoch 22, batch 2200, loss[loss=0.1929, simple_loss=0.254, pruned_loss=0.06587, over 13345.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2365, pruned_loss=0.05703, over 2664388.47 frames. ], batch size: 118, lr: 5.90e-03, grad_scale: 32.0 2023-04-17 02:54:51,380 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2023-04-17 02:55:13,854 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=58906.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:55:16,363 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7148, 2.3124, 1.9985, 1.5025, 1.7905, 2.0214, 2.1122, 2.5868], device='cuda:1'), covar=tensor([0.0876, 0.0311, 0.0775, 0.1625, 0.0165, 0.0510, 0.0751, 0.0228], device='cuda:1'), in_proj_covar=tensor([0.0171, 0.0145, 0.0125, 0.0213, 0.0115, 0.0164, 0.0174, 0.0132], device='cuda:1'), out_proj_covar=tensor([1.2647e-04, 1.0896e-04, 9.7419e-05, 1.5849e-04, 8.4248e-05, 1.2362e-04, 1.3194e-04, 9.7350e-05], device='cuda:1') 2023-04-17 02:55:25,992 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.683e+02 2.473e+02 2.857e+02 3.422e+02 4.907e+02, threshold=5.714e+02, percent-clipped=0.0 2023-04-17 02:55:37,285 INFO [train.py:893] (1/4) Epoch 22, batch 2250, loss[loss=0.1451, simple_loss=0.2131, pruned_loss=0.03851, over 13494.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2364, pruned_loss=0.05708, over 2663785.93 frames. ], batch size: 81, lr: 5.89e-03, grad_scale: 32.0 2023-04-17 02:56:10,635 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.4333, 3.3265, 2.6478, 2.9343, 2.6896, 2.0209, 3.4237, 1.8858], device='cuda:1'), covar=tensor([0.0767, 0.0734, 0.0535, 0.0558, 0.0793, 0.2079, 0.1106, 0.1401], device='cuda:1'), in_proj_covar=tensor([0.0144, 0.0144, 0.0136, 0.0118, 0.0150, 0.0191, 0.0178, 0.0146], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 02:56:12,303 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=58967.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:56:24,443 INFO [train.py:893] (1/4) Epoch 22, batch 2300, loss[loss=0.1821, simple_loss=0.2403, pruned_loss=0.06192, over 13190.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2348, pruned_loss=0.05595, over 2664771.78 frames. ], batch size: 132, lr: 5.89e-03, grad_scale: 32.0 2023-04-17 02:57:02,018 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.824e+02 2.500e+02 2.983e+02 3.435e+02 5.453e+02, threshold=5.967e+02, percent-clipped=0.0 2023-04-17 02:57:12,922 INFO [train.py:893] (1/4) Epoch 22, batch 2350, loss[loss=0.1841, simple_loss=0.2398, pruned_loss=0.06414, over 13501.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2344, pruned_loss=0.05578, over 2661643.63 frames. ], batch size: 93, lr: 5.89e-03, grad_scale: 32.0 2023-04-17 02:57:18,297 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59037.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:57:36,445 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-17 02:57:52,605 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.5557, 2.1625, 4.1623, 3.9044, 4.0401, 3.3118, 3.7652, 3.0860], device='cuda:1'), covar=tensor([0.1767, 0.1639, 0.0125, 0.0236, 0.0185, 0.0558, 0.0229, 0.0744], device='cuda:1'), in_proj_covar=tensor([0.0217, 0.0184, 0.0126, 0.0128, 0.0137, 0.0174, 0.0144, 0.0185], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-17 02:57:59,830 INFO [train.py:893] (1/4) Epoch 22, batch 2400, loss[loss=0.1864, simple_loss=0.241, pruned_loss=0.06596, over 13444.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2335, pruned_loss=0.05579, over 2656190.51 frames. ], batch size: 103, lr: 5.89e-03, grad_scale: 32.0 2023-04-17 02:58:03,959 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-04-17 02:58:07,345 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-04-17 02:58:33,878 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.8857, 2.8244, 2.5245, 1.8253, 1.8684, 2.4715, 2.4658, 3.0883], device='cuda:1'), covar=tensor([0.1080, 0.0392, 0.0598, 0.1540, 0.0386, 0.0457, 0.0792, 0.0216], device='cuda:1'), in_proj_covar=tensor([0.0172, 0.0145, 0.0124, 0.0214, 0.0116, 0.0165, 0.0174, 0.0131], device='cuda:1'), out_proj_covar=tensor([1.2703e-04, 1.0924e-04, 9.6953e-05, 1.5847e-04, 8.4549e-05, 1.2448e-04, 1.3169e-04, 9.6634e-05], device='cuda:1') 2023-04-17 02:58:36,855 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.705e+02 2.612e+02 2.930e+02 3.503e+02 4.997e+02, threshold=5.859e+02, percent-clipped=0.0 2023-04-17 02:58:46,950 INFO [train.py:893] (1/4) Epoch 22, batch 2450, loss[loss=0.179, simple_loss=0.2451, pruned_loss=0.05649, over 13497.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2338, pruned_loss=0.05602, over 2658580.77 frames. ], batch size: 93, lr: 5.88e-03, grad_scale: 32.0 2023-04-17 02:58:47,150 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59131.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:59:18,162 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59164.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 02:59:31,498 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1619, 3.9159, 4.1100, 2.7313, 4.4448, 4.1815, 4.1130, 4.3822], device='cuda:1'), covar=tensor([0.0206, 0.0145, 0.0123, 0.0999, 0.0121, 0.0224, 0.0147, 0.0092], device='cuda:1'), in_proj_covar=tensor([0.0093, 0.0057, 0.0085, 0.0105, 0.0101, 0.0111, 0.0082, 0.0077], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 02:59:34,872 INFO [train.py:893] (1/4) Epoch 22, batch 2500, loss[loss=0.1717, simple_loss=0.2345, pruned_loss=0.05442, over 13529.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2343, pruned_loss=0.05636, over 2652071.17 frames. ], batch size: 98, lr: 5.88e-03, grad_scale: 64.0 2023-04-17 02:59:41,040 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0988, 4.2255, 3.2797, 2.8158, 3.0849, 2.5388, 4.3592, 2.4297], device='cuda:1'), covar=tensor([0.1592, 0.0291, 0.1184, 0.2229, 0.0839, 0.3263, 0.0225, 0.4233], device='cuda:1'), in_proj_covar=tensor([0.0314, 0.0291, 0.0318, 0.0334, 0.0261, 0.0330, 0.0214, 0.0368], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 02:59:48,467 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9453, 4.0438, 3.0478, 3.7075, 3.0821, 2.2458, 4.0781, 2.1476], device='cuda:1'), covar=tensor([0.0700, 0.0330, 0.0541, 0.0322, 0.0714, 0.2012, 0.0537, 0.1414], device='cuda:1'), in_proj_covar=tensor([0.0145, 0.0145, 0.0136, 0.0118, 0.0151, 0.0192, 0.0178, 0.0147], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 03:00:12,740 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.906e+02 2.615e+02 2.930e+02 3.639e+02 8.531e+02, threshold=5.860e+02, percent-clipped=2.0 2023-04-17 03:00:16,476 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59225.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:00:22,136 INFO [train.py:893] (1/4) Epoch 22, batch 2550, loss[loss=0.147, simple_loss=0.2165, pruned_loss=0.03878, over 13544.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2343, pruned_loss=0.05616, over 2657509.91 frames. ], batch size: 78, lr: 5.88e-03, grad_scale: 16.0 2023-04-17 03:00:47,170 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-17 03:00:51,300 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59262.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:01:08,889 INFO [train.py:893] (1/4) Epoch 22, batch 2600, loss[loss=0.1626, simple_loss=0.2259, pruned_loss=0.04967, over 13470.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2344, pruned_loss=0.05644, over 2660203.14 frames. ], batch size: 79, lr: 5.88e-03, grad_scale: 16.0 2023-04-17 03:01:44,367 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.931e+02 2.687e+02 3.147e+02 3.617e+02 7.653e+02, threshold=6.293e+02, percent-clipped=2.0 2023-04-17 03:01:44,529 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1303, 4.3005, 4.2374, 4.0742, 4.3408, 3.9572, 4.3591, 4.4606], device='cuda:1'), covar=tensor([0.0296, 0.0524, 0.0320, 0.0502, 0.0334, 0.0453, 0.0550, 0.0394], device='cuda:1'), in_proj_covar=tensor([0.0136, 0.0200, 0.0166, 0.0182, 0.0152, 0.0198, 0.0137, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:01:51,714 INFO [train.py:893] (1/4) Epoch 22, batch 2650, loss[loss=0.1777, simple_loss=0.2328, pruned_loss=0.06128, over 13349.00 frames. ], tot_loss[loss=0.1751, simple_loss=0.2354, pruned_loss=0.05737, over 2658155.79 frames. ], batch size: 67, lr: 5.87e-03, grad_scale: 16.0 2023-04-17 03:01:56,393 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59337.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:02:00,342 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3896, 3.7434, 3.5987, 4.1231, 2.3176, 3.1705, 3.9199, 2.3027], device='cuda:1'), covar=tensor([0.0147, 0.0458, 0.0737, 0.0473, 0.1580, 0.0900, 0.0502, 0.1662], device='cuda:1'), in_proj_covar=tensor([0.0131, 0.0185, 0.0207, 0.0247, 0.0183, 0.0198, 0.0180, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:02:25,072 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6504, 3.6964, 2.6584, 3.3831, 3.6105, 2.4166, 3.2825, 2.5821], device='cuda:1'), covar=tensor([0.0323, 0.0251, 0.1089, 0.0403, 0.0348, 0.1307, 0.0595, 0.1300], device='cuda:1'), in_proj_covar=tensor([0.0161, 0.0179, 0.0179, 0.0220, 0.0142, 0.0161, 0.0163, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:02:49,883 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-17 03:03:00,008 INFO [train.py:893] (1/4) Epoch 23, batch 0, loss[loss=0.1707, simple_loss=0.2314, pruned_loss=0.05506, over 13342.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2314, pruned_loss=0.05506, over 13342.00 frames. ], batch size: 118, lr: 5.74e-03, grad_scale: 16.0 2023-04-17 03:03:00,008 INFO [train.py:918] (1/4) Computing validation loss 2023-04-17 03:03:15,330 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0523, 4.3010, 4.1591, 4.1659, 4.2139, 4.1237, 4.3324, 4.4017], device='cuda:1'), covar=tensor([0.0228, 0.0247, 0.0214, 0.0324, 0.0271, 0.0275, 0.0293, 0.0241], device='cuda:1'), in_proj_covar=tensor([0.0139, 0.0203, 0.0169, 0.0186, 0.0155, 0.0202, 0.0140, 0.0180], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:03:18,050 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1813, 2.3049, 2.7752, 3.4457, 3.1722, 3.4448, 3.0070, 2.5272], device='cuda:1'), covar=tensor([0.0267, 0.0725, 0.0452, 0.0078, 0.0228, 0.0072, 0.0421, 0.0683], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0151, 0.0166, 0.0100, 0.0120, 0.0097, 0.0168, 0.0148], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:03:23,126 INFO [train.py:927] (1/4) Epoch 23, validation: loss=0.135, simple_loss=0.1947, pruned_loss=0.03761, over 2446609.00 frames. 2023-04-17 03:03:23,128 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12831MB 2023-04-17 03:03:26,594 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59385.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:03:54,060 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.72 vs. limit=5.0 2023-04-17 03:04:01,981 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.652e+02 2.605e+02 3.123e+02 3.576e+02 6.770e+02, threshold=6.246e+02, percent-clipped=1.0 2023-04-17 03:04:10,606 INFO [train.py:893] (1/4) Epoch 23, batch 50, loss[loss=0.2063, simple_loss=0.2547, pruned_loss=0.07889, over 11712.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.228, pruned_loss=0.05628, over 588689.83 frames. ], batch size: 157, lr: 5.74e-03, grad_scale: 16.0 2023-04-17 03:04:10,882 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59431.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:04:35,467 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-17 03:04:35,467 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-17 03:04:35,468 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-17 03:04:35,477 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-17 03:04:35,485 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-17 03:04:35,507 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-17 03:04:36,189 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-17 03:04:39,034 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7445, 3.9943, 3.9242, 3.4421, 3.8329, 4.0564, 4.0202, 3.8359], device='cuda:1'), covar=tensor([0.0355, 0.0310, 0.0348, 0.1207, 0.0352, 0.0344, 0.0355, 0.0742], device='cuda:1'), in_proj_covar=tensor([0.0162, 0.0157, 0.0177, 0.0265, 0.0175, 0.0191, 0.0172, 0.0112], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 03:04:55,721 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59479.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:04:57,221 INFO [train.py:893] (1/4) Epoch 23, batch 100, loss[loss=0.1898, simple_loss=0.2458, pruned_loss=0.06687, over 13526.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2336, pruned_loss=0.05888, over 1046397.65 frames. ], batch size: 98, lr: 5.74e-03, grad_scale: 16.0 2023-04-17 03:04:58,382 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9226, 2.1584, 3.7852, 3.6318, 3.5921, 3.0134, 3.4182, 2.9434], device='cuda:1'), covar=tensor([0.2242, 0.1436, 0.0151, 0.0198, 0.0312, 0.0651, 0.0264, 0.0746], device='cuda:1'), in_proj_covar=tensor([0.0219, 0.0186, 0.0127, 0.0129, 0.0138, 0.0177, 0.0144, 0.0187], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-17 03:05:17,323 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59502.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:05:34,909 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59520.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:05:36,338 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.879e+02 2.548e+02 3.039e+02 3.718e+02 1.416e+03, threshold=6.077e+02, percent-clipped=2.0 2023-04-17 03:05:44,646 INFO [train.py:893] (1/4) Epoch 23, batch 150, loss[loss=0.1928, simple_loss=0.2529, pruned_loss=0.06639, over 13445.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2357, pruned_loss=0.05984, over 1406287.01 frames. ], batch size: 103, lr: 5.73e-03, grad_scale: 16.0 2023-04-17 03:06:14,299 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59562.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:06:15,145 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59563.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:06:20,186 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9466, 2.3804, 1.9677, 3.8428, 4.2783, 3.1950, 4.1699, 3.9759], device='cuda:1'), covar=tensor([0.0085, 0.0966, 0.0983, 0.0077, 0.0059, 0.0459, 0.0077, 0.0086], device='cuda:1'), in_proj_covar=tensor([0.0076, 0.0089, 0.0096, 0.0080, 0.0068, 0.0080, 0.0056, 0.0077], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 03:06:21,301 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.77 vs. limit=5.0 2023-04-17 03:06:31,611 INFO [train.py:893] (1/4) Epoch 23, batch 200, loss[loss=0.1699, simple_loss=0.2346, pruned_loss=0.05264, over 13376.00 frames. ], tot_loss[loss=0.179, simple_loss=0.2375, pruned_loss=0.06031, over 1683395.50 frames. ], batch size: 77, lr: 5.73e-03, grad_scale: 16.0 2023-04-17 03:06:50,437 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-04-17 03:06:57,750 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-04-17 03:06:59,079 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59610.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:07:07,021 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 2023-04-17 03:07:11,432 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.051e+02 2.601e+02 3.074e+02 3.611e+02 6.828e+02, threshold=6.148e+02, percent-clipped=4.0 2023-04-17 03:07:18,927 INFO [train.py:893] (1/4) Epoch 23, batch 250, loss[loss=0.1823, simple_loss=0.2302, pruned_loss=0.06717, over 13439.00 frames. ], tot_loss[loss=0.179, simple_loss=0.2377, pruned_loss=0.06012, over 1901360.32 frames. ], batch size: 65, lr: 5.73e-03, grad_scale: 16.0 2023-04-17 03:07:26,214 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-04-17 03:07:54,885 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5748, 3.9721, 3.7953, 4.3442, 2.5068, 3.3116, 4.1827, 2.4726], device='cuda:1'), covar=tensor([0.0219, 0.0485, 0.0773, 0.0564, 0.1626, 0.1013, 0.0412, 0.1862], device='cuda:1'), in_proj_covar=tensor([0.0133, 0.0186, 0.0208, 0.0248, 0.0183, 0.0199, 0.0183, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:08:06,536 INFO [train.py:893] (1/4) Epoch 23, batch 300, loss[loss=0.1443, simple_loss=0.1985, pruned_loss=0.04508, over 12469.00 frames. ], tot_loss[loss=0.178, simple_loss=0.2371, pruned_loss=0.05941, over 2060699.75 frames. ], batch size: 50, lr: 5.73e-03, grad_scale: 16.0 2023-04-17 03:08:45,230 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.995e+02 2.544e+02 2.931e+02 3.487e+02 6.573e+02, threshold=5.861e+02, percent-clipped=2.0 2023-04-17 03:08:52,893 INFO [train.py:893] (1/4) Epoch 23, batch 350, loss[loss=0.1796, simple_loss=0.2292, pruned_loss=0.06504, over 13370.00 frames. ], tot_loss[loss=0.1783, simple_loss=0.2375, pruned_loss=0.05956, over 2192517.10 frames. ], batch size: 73, lr: 5.72e-03, grad_scale: 16.0 2023-04-17 03:08:59,466 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-04-17 03:09:06,679 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.66 vs. limit=2.0 2023-04-17 03:09:40,695 INFO [train.py:893] (1/4) Epoch 23, batch 400, loss[loss=0.1789, simple_loss=0.2466, pruned_loss=0.05563, over 13379.00 frames. ], tot_loss[loss=0.1782, simple_loss=0.2377, pruned_loss=0.05929, over 2294855.47 frames. ], batch size: 109, lr: 5.72e-03, grad_scale: 16.0 2023-04-17 03:10:04,475 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.87 vs. limit=5.0 2023-04-17 03:10:09,993 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6247, 2.6691, 2.8533, 4.0774, 3.6489, 4.1003, 3.2355, 2.6071], device='cuda:1'), covar=tensor([0.0252, 0.0799, 0.0738, 0.0056, 0.0237, 0.0056, 0.0622, 0.0847], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0152, 0.0170, 0.0102, 0.0121, 0.0099, 0.0171, 0.0150], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:10:16,677 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=59820.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:10:17,669 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5522, 3.4629, 4.0395, 2.8694, 2.4982, 2.6882, 4.3661, 4.4892], device='cuda:1'), covar=tensor([0.1102, 0.1628, 0.0398, 0.1742, 0.1766, 0.1630, 0.0267, 0.0272], device='cuda:1'), in_proj_covar=tensor([0.0247, 0.0267, 0.0195, 0.0223, 0.0218, 0.0180, 0.0211, 0.0181], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002], device='cuda:1') 2023-04-17 03:10:18,076 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.855e+02 2.576e+02 2.992e+02 3.488e+02 5.440e+02, threshold=5.985e+02, percent-clipped=0.0 2023-04-17 03:10:26,412 INFO [train.py:893] (1/4) Epoch 23, batch 450, loss[loss=0.1879, simple_loss=0.2501, pruned_loss=0.06289, over 13453.00 frames. ], tot_loss[loss=0.1795, simple_loss=0.2394, pruned_loss=0.05982, over 2377238.83 frames. ], batch size: 100, lr: 5.72e-03, grad_scale: 16.0 2023-04-17 03:10:43,393 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59848.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:10:52,187 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-17 03:10:52,357 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=59858.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:11:01,367 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=59868.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:11:14,011 INFO [train.py:893] (1/4) Epoch 23, batch 500, loss[loss=0.1641, simple_loss=0.2266, pruned_loss=0.05077, over 13542.00 frames. ], tot_loss[loss=0.1798, simple_loss=0.2396, pruned_loss=0.06003, over 2434582.99 frames. ], batch size: 83, lr: 5.72e-03, grad_scale: 16.0 2023-04-17 03:11:21,069 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=59889.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:11:40,917 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59909.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:11:52,062 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.757e+02 2.557e+02 2.976e+02 3.530e+02 5.641e+02, threshold=5.953e+02, percent-clipped=0.0 2023-04-17 03:12:00,579 INFO [train.py:893] (1/4) Epoch 23, batch 550, loss[loss=0.1771, simple_loss=0.2437, pruned_loss=0.05527, over 13224.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.2387, pruned_loss=0.05912, over 2487505.98 frames. ], batch size: 117, lr: 5.71e-03, grad_scale: 16.0 2023-04-17 03:12:18,415 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=59950.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:12:47,120 INFO [train.py:893] (1/4) Epoch 23, batch 600, loss[loss=0.1774, simple_loss=0.2407, pruned_loss=0.05702, over 13384.00 frames. ], tot_loss[loss=0.1773, simple_loss=0.2373, pruned_loss=0.05866, over 2518546.38 frames. ], batch size: 109, lr: 5.71e-03, grad_scale: 16.0 2023-04-17 03:12:48,173 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7484, 3.9367, 3.8426, 3.7910, 3.9163, 3.7176, 4.0052, 4.0446], device='cuda:1'), covar=tensor([0.0214, 0.0280, 0.0247, 0.0305, 0.0264, 0.0302, 0.0271, 0.0277], device='cuda:1'), in_proj_covar=tensor([0.0138, 0.0199, 0.0166, 0.0182, 0.0154, 0.0201, 0.0137, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:13:29,647 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.782e+02 2.485e+02 2.931e+02 3.430e+02 4.559e+02, threshold=5.862e+02, percent-clipped=0.0 2023-04-17 03:13:38,790 INFO [train.py:893] (1/4) Epoch 23, batch 650, loss[loss=0.1659, simple_loss=0.2276, pruned_loss=0.05213, over 13465.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2367, pruned_loss=0.0586, over 2551199.13 frames. ], batch size: 65, lr: 5.71e-03, grad_scale: 16.0 2023-04-17 03:13:39,087 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1659, 3.9743, 4.1412, 2.4798, 4.4185, 4.1883, 4.1944, 4.4585], device='cuda:1'), covar=tensor([0.0224, 0.0146, 0.0139, 0.1146, 0.0158, 0.0282, 0.0162, 0.0099], device='cuda:1'), in_proj_covar=tensor([0.0093, 0.0057, 0.0084, 0.0104, 0.0100, 0.0110, 0.0082, 0.0077], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:13:46,467 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60040.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:13:54,160 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.0171, 4.8276, 5.0842, 4.8783, 5.3396, 4.8119, 5.3550, 5.3294], device='cuda:1'), covar=tensor([0.0462, 0.0595, 0.0577, 0.0635, 0.0510, 0.0870, 0.0491, 0.0470], device='cuda:1'), in_proj_covar=tensor([0.0293, 0.0301, 0.0305, 0.0229, 0.0439, 0.0346, 0.0284, 0.0301], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:13:58,249 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60052.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:14:10,422 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60065.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:14:21,075 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60078.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:14:24,062 INFO [train.py:893] (1/4) Epoch 23, batch 700, loss[loss=0.1471, simple_loss=0.2052, pruned_loss=0.0445, over 13161.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2362, pruned_loss=0.05803, over 2580029.57 frames. ], batch size: 58, lr: 5.71e-03, grad_scale: 16.0 2023-04-17 03:14:43,034 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60101.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 03:14:53,758 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60113.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 03:15:01,791 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.826e+02 2.678e+02 3.015e+02 3.584e+02 9.072e+02, threshold=6.030e+02, percent-clipped=5.0 2023-04-17 03:15:06,968 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60126.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:15:10,906 INFO [train.py:893] (1/4) Epoch 23, batch 750, loss[loss=0.1616, simple_loss=0.23, pruned_loss=0.04659, over 13249.00 frames. ], tot_loss[loss=0.177, simple_loss=0.2365, pruned_loss=0.0587, over 2598304.90 frames. ], batch size: 124, lr: 5.71e-03, grad_scale: 16.0 2023-04-17 03:15:17,942 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60139.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:15:36,552 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60158.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:15:57,190 INFO [train.py:893] (1/4) Epoch 23, batch 800, loss[loss=0.1736, simple_loss=0.2388, pruned_loss=0.05418, over 13202.00 frames. ], tot_loss[loss=0.1787, simple_loss=0.2384, pruned_loss=0.05951, over 2616214.46 frames. ], batch size: 132, lr: 5.70e-03, grad_scale: 16.0 2023-04-17 03:16:13,234 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60198.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:16:19,191 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60204.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:16:20,889 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60206.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:16:36,057 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.678e+02 2.576e+02 3.130e+02 3.550e+02 5.821e+02, threshold=6.259e+02, percent-clipped=0.0 2023-04-17 03:16:43,541 INFO [train.py:893] (1/4) Epoch 23, batch 850, loss[loss=0.1736, simple_loss=0.2367, pruned_loss=0.05526, over 13526.00 frames. ], tot_loss[loss=0.1795, simple_loss=0.2395, pruned_loss=0.05981, over 2628227.00 frames. ], batch size: 76, lr: 5.70e-03, grad_scale: 16.0 2023-04-17 03:16:57,108 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60245.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:17:10,310 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60259.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:17:29,946 INFO [train.py:893] (1/4) Epoch 23, batch 900, loss[loss=0.1659, simple_loss=0.2286, pruned_loss=0.05158, over 13456.00 frames. ], tot_loss[loss=0.1785, simple_loss=0.2381, pruned_loss=0.05943, over 2636151.22 frames. ], batch size: 79, lr: 5.70e-03, grad_scale: 16.0 2023-04-17 03:17:43,745 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1179, 2.4738, 1.9503, 4.0249, 4.5319, 3.2822, 4.3934, 4.2555], device='cuda:1'), covar=tensor([0.0099, 0.0926, 0.1043, 0.0090, 0.0054, 0.0460, 0.0070, 0.0077], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0091, 0.0098, 0.0082, 0.0069, 0.0082, 0.0058, 0.0079], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 03:18:02,182 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-17 03:18:07,923 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.882e+02 2.736e+02 3.142e+02 3.652e+02 7.263e+02, threshold=6.284e+02, percent-clipped=3.0 2023-04-17 03:18:15,683 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5610, 3.9539, 3.6909, 4.3232, 2.4322, 3.2773, 4.0639, 2.3430], device='cuda:1'), covar=tensor([0.0174, 0.0431, 0.0866, 0.0623, 0.1647, 0.0905, 0.0443, 0.1720], device='cuda:1'), in_proj_covar=tensor([0.0133, 0.0187, 0.0209, 0.0250, 0.0184, 0.0200, 0.0182, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:18:16,148 INFO [train.py:893] (1/4) Epoch 23, batch 950, loss[loss=0.1485, simple_loss=0.2144, pruned_loss=0.04129, over 13514.00 frames. ], tot_loss[loss=0.1788, simple_loss=0.2379, pruned_loss=0.05984, over 2642569.62 frames. ], batch size: 76, lr: 5.70e-03, grad_scale: 16.0 2023-04-17 03:18:43,142 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-04-17 03:19:01,356 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.4032, 4.8814, 4.7276, 4.9558, 4.6558, 4.6998, 5.3520, 4.9253], device='cuda:1'), covar=tensor([0.0672, 0.1238, 0.2109, 0.2183, 0.1022, 0.1720, 0.0880, 0.1206], device='cuda:1'), in_proj_covar=tensor([0.0339, 0.0388, 0.0477, 0.0482, 0.0298, 0.0359, 0.0441, 0.0356], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 03:19:02,849 INFO [train.py:893] (1/4) Epoch 23, batch 1000, loss[loss=0.1594, simple_loss=0.2197, pruned_loss=0.04956, over 13529.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2358, pruned_loss=0.05882, over 2649067.56 frames. ], batch size: 72, lr: 5.69e-03, grad_scale: 16.0 2023-04-17 03:19:16,277 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60396.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 03:19:27,727 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60408.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 03:19:39,638 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60421.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:19:40,245 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.841e+02 2.528e+02 2.847e+02 3.554e+02 6.904e+02, threshold=5.694e+02, percent-clipped=4.0 2023-04-17 03:19:48,609 INFO [train.py:893] (1/4) Epoch 23, batch 1050, loss[loss=0.1687, simple_loss=0.2338, pruned_loss=0.05179, over 13528.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2342, pruned_loss=0.05761, over 2651313.72 frames. ], batch size: 98, lr: 5.69e-03, grad_scale: 16.0 2023-04-17 03:19:51,910 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60434.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:20:35,610 INFO [train.py:893] (1/4) Epoch 23, batch 1100, loss[loss=0.1701, simple_loss=0.2363, pruned_loss=0.052, over 13540.00 frames. ], tot_loss[loss=0.1741, simple_loss=0.2343, pruned_loss=0.05698, over 2652135.04 frames. ], batch size: 91, lr: 5.69e-03, grad_scale: 16.0 2023-04-17 03:20:47,527 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8580, 4.0833, 3.1505, 2.8446, 2.9473, 2.5304, 4.2090, 2.4130], device='cuda:1'), covar=tensor([0.1840, 0.0320, 0.1307, 0.2096, 0.0895, 0.3306, 0.0265, 0.3988], device='cuda:1'), in_proj_covar=tensor([0.0318, 0.0293, 0.0321, 0.0339, 0.0264, 0.0333, 0.0218, 0.0370], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 03:20:57,195 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60504.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:21:12,731 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.950e+02 2.578e+02 3.034e+02 3.780e+02 7.669e+02, threshold=6.069e+02, percent-clipped=4.0 2023-04-17 03:21:21,257 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.5615, 2.6964, 2.3916, 1.5184, 1.5883, 2.2693, 2.1899, 2.9247], device='cuda:1'), covar=tensor([0.1209, 0.0322, 0.0583, 0.1693, 0.0209, 0.0431, 0.0810, 0.0183], device='cuda:1'), in_proj_covar=tensor([0.0172, 0.0147, 0.0125, 0.0215, 0.0117, 0.0169, 0.0177, 0.0133], device='cuda:1'), out_proj_covar=tensor([1.2752e-04, 1.1060e-04, 9.7529e-05, 1.5991e-04, 8.5111e-05, 1.2776e-04, 1.3406e-04, 9.8179e-05], device='cuda:1') 2023-04-17 03:21:21,692 INFO [train.py:893] (1/4) Epoch 23, batch 1150, loss[loss=0.2042, simple_loss=0.2644, pruned_loss=0.07206, over 13456.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2339, pruned_loss=0.05629, over 2652984.12 frames. ], batch size: 106, lr: 5.69e-03, grad_scale: 16.0 2023-04-17 03:21:35,655 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60545.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:21:41,488 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60552.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:21:43,922 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=60554.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:22:08,295 INFO [train.py:893] (1/4) Epoch 23, batch 1200, loss[loss=0.1791, simple_loss=0.237, pruned_loss=0.06057, over 13561.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2345, pruned_loss=0.05639, over 2654146.56 frames. ], batch size: 89, lr: 5.68e-03, grad_scale: 16.0 2023-04-17 03:22:20,124 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60593.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:22:30,339 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0767, 4.3665, 4.0836, 4.1913, 4.2053, 4.5207, 4.3294, 4.1553], device='cuda:1'), covar=tensor([0.0335, 0.0301, 0.0372, 0.0825, 0.0317, 0.0238, 0.0340, 0.0370], device='cuda:1'), in_proj_covar=tensor([0.0163, 0.0158, 0.0180, 0.0266, 0.0177, 0.0192, 0.0173, 0.0112], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 03:22:37,671 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-17 03:22:47,543 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.088e+02 2.864e+02 3.217e+02 3.727e+02 5.611e+02, threshold=6.435e+02, percent-clipped=0.0 2023-04-17 03:22:50,062 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-17 03:22:55,872 INFO [train.py:893] (1/4) Epoch 23, batch 1250, loss[loss=0.1988, simple_loss=0.2506, pruned_loss=0.07349, over 13446.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.2361, pruned_loss=0.05736, over 2653672.95 frames. ], batch size: 103, lr: 5.68e-03, grad_scale: 16.0 2023-04-17 03:23:43,287 INFO [train.py:893] (1/4) Epoch 23, batch 1300, loss[loss=0.1978, simple_loss=0.2523, pruned_loss=0.07168, over 11738.00 frames. ], tot_loss[loss=0.1764, simple_loss=0.2371, pruned_loss=0.05788, over 2656693.21 frames. ], batch size: 157, lr: 5.68e-03, grad_scale: 16.0 2023-04-17 03:23:57,533 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60696.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:24:08,201 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60708.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:24:15,777 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9494, 3.7475, 3.0216, 3.3647, 3.1066, 2.2431, 3.7798, 2.1144], device='cuda:1'), covar=tensor([0.0602, 0.0517, 0.0489, 0.0374, 0.0681, 0.1928, 0.0966, 0.1273], device='cuda:1'), in_proj_covar=tensor([0.0144, 0.0144, 0.0136, 0.0118, 0.0150, 0.0190, 0.0179, 0.0144], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 03:24:18,467 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=60718.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:24:20,795 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60721.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:24:21,331 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.809e+02 2.793e+02 3.175e+02 3.621e+02 6.568e+02, threshold=6.350e+02, percent-clipped=2.0 2023-04-17 03:24:29,570 INFO [train.py:893] (1/4) Epoch 23, batch 1350, loss[loss=0.1535, simple_loss=0.2189, pruned_loss=0.04411, over 13476.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2378, pruned_loss=0.05795, over 2660050.58 frames. ], batch size: 100, lr: 5.68e-03, grad_scale: 16.0 2023-04-17 03:24:32,390 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60734.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:24:42,230 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60744.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:24:53,074 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60756.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:25:04,564 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60769.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:25:14,291 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=60779.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:25:16,391 INFO [train.py:893] (1/4) Epoch 23, batch 1400, loss[loss=0.191, simple_loss=0.2499, pruned_loss=0.06603, over 13472.00 frames. ], tot_loss[loss=0.1767, simple_loss=0.2376, pruned_loss=0.05795, over 2659791.94 frames. ], batch size: 103, lr: 5.67e-03, grad_scale: 16.0 2023-04-17 03:25:17,322 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60782.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:25:52,986 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.093e+02 2.602e+02 2.908e+02 3.577e+02 9.144e+02, threshold=5.816e+02, percent-clipped=1.0 2023-04-17 03:26:01,603 INFO [train.py:893] (1/4) Epoch 23, batch 1450, loss[loss=0.1979, simple_loss=0.25, pruned_loss=0.07288, over 13565.00 frames. ], tot_loss[loss=0.1777, simple_loss=0.2381, pruned_loss=0.05859, over 2659551.29 frames. ], batch size: 89, lr: 5.67e-03, grad_scale: 16.0 2023-04-17 03:26:03,622 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8998, 3.5836, 4.5042, 3.3677, 3.0281, 3.0741, 4.7785, 4.8516], device='cuda:1'), covar=tensor([0.1144, 0.2049, 0.0329, 0.1600, 0.1472, 0.1429, 0.0247, 0.0303], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0271, 0.0197, 0.0226, 0.0221, 0.0182, 0.0214, 0.0185], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 03:26:24,276 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=60854.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:26:49,313 INFO [train.py:893] (1/4) Epoch 23, batch 1500, loss[loss=0.1509, simple_loss=0.2125, pruned_loss=0.04468, over 13501.00 frames. ], tot_loss[loss=0.1762, simple_loss=0.2367, pruned_loss=0.05788, over 2658273.38 frames. ], batch size: 70, lr: 5.67e-03, grad_scale: 16.0 2023-04-17 03:27:09,256 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=60902.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:27:27,729 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.986e+02 2.779e+02 3.136e+02 3.771e+02 6.785e+02, threshold=6.272e+02, percent-clipped=2.0 2023-04-17 03:27:36,802 INFO [train.py:893] (1/4) Epoch 23, batch 1550, loss[loss=0.1681, simple_loss=0.2321, pruned_loss=0.05202, over 13540.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.236, pruned_loss=0.05742, over 2653878.07 frames. ], batch size: 85, lr: 5.67e-03, grad_scale: 16.0 2023-04-17 03:27:52,135 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1665, 2.4358, 1.9497, 3.9939, 4.5424, 3.2090, 4.4704, 4.2306], device='cuda:1'), covar=tensor([0.0129, 0.1060, 0.1250, 0.0124, 0.0096, 0.0584, 0.0096, 0.0097], device='cuda:1'), in_proj_covar=tensor([0.0077, 0.0091, 0.0098, 0.0082, 0.0068, 0.0082, 0.0057, 0.0078], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 03:28:23,235 INFO [train.py:893] (1/4) Epoch 23, batch 1600, loss[loss=0.1605, simple_loss=0.2241, pruned_loss=0.04848, over 13498.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.236, pruned_loss=0.05721, over 2658989.93 frames. ], batch size: 70, lr: 5.67e-03, grad_scale: 16.0 2023-04-17 03:29:01,735 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.731e+02 2.549e+02 3.025e+02 3.497e+02 7.515e+02, threshold=6.050e+02, percent-clipped=2.0 2023-04-17 03:29:10,279 INFO [train.py:893] (1/4) Epoch 23, batch 1650, loss[loss=0.1734, simple_loss=0.233, pruned_loss=0.05687, over 13529.00 frames. ], tot_loss[loss=0.1753, simple_loss=0.2365, pruned_loss=0.05709, over 2656216.04 frames. ], batch size: 76, lr: 5.66e-03, grad_scale: 16.0 2023-04-17 03:29:39,761 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61061.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:29:40,705 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8831, 2.5220, 2.5459, 3.0103, 2.1865, 3.0551, 2.9468, 2.4113], device='cuda:1'), covar=tensor([0.0080, 0.0184, 0.0148, 0.0152, 0.0233, 0.0121, 0.0154, 0.0194], device='cuda:1'), in_proj_covar=tensor([0.0107, 0.0120, 0.0125, 0.0126, 0.0135, 0.0112, 0.0110, 0.0110], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 03:29:51,334 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61074.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:29:56,939 INFO [train.py:893] (1/4) Epoch 23, batch 1700, loss[loss=0.1694, simple_loss=0.2401, pruned_loss=0.04941, over 13358.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2372, pruned_loss=0.05705, over 2659172.82 frames. ], batch size: 118, lr: 5.66e-03, grad_scale: 16.0 2023-04-17 03:30:35,190 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0686, 2.8678, 2.4996, 1.8668, 1.8960, 2.4160, 2.6566, 3.1694], device='cuda:1'), covar=tensor([0.0975, 0.0459, 0.0796, 0.1721, 0.0403, 0.0742, 0.0741, 0.0249], device='cuda:1'), in_proj_covar=tensor([0.0170, 0.0146, 0.0124, 0.0213, 0.0116, 0.0167, 0.0175, 0.0131], device='cuda:1'), out_proj_covar=tensor([1.2617e-04, 1.0934e-04, 9.6752e-05, 1.5823e-04, 8.4272e-05, 1.2639e-04, 1.3204e-04, 9.6954e-05], device='cuda:1') 2023-04-17 03:30:35,663 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.757e+02 2.631e+02 2.974e+02 3.481e+02 8.450e+02, threshold=5.948e+02, percent-clipped=1.0 2023-04-17 03:30:36,760 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61122.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:30:41,840 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.8887, 4.2442, 3.9413, 4.6708, 2.6604, 3.5479, 4.4328, 2.7146], device='cuda:1'), covar=tensor([0.0123, 0.0406, 0.0715, 0.0517, 0.1478, 0.0848, 0.0303, 0.1663], device='cuda:1'), in_proj_covar=tensor([0.0131, 0.0185, 0.0207, 0.0246, 0.0183, 0.0197, 0.0179, 0.0176], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:30:43,998 INFO [train.py:893] (1/4) Epoch 23, batch 1750, loss[loss=0.1637, simple_loss=0.2252, pruned_loss=0.05107, over 13417.00 frames. ], tot_loss[loss=0.1745, simple_loss=0.2361, pruned_loss=0.0565, over 2663111.02 frames. ], batch size: 95, lr: 5.66e-03, grad_scale: 16.0 2023-04-17 03:31:30,112 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-04-17 03:31:31,095 INFO [train.py:893] (1/4) Epoch 23, batch 1800, loss[loss=0.144, simple_loss=0.2086, pruned_loss=0.03968, over 13361.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2353, pruned_loss=0.05603, over 2666973.75 frames. ], batch size: 67, lr: 5.66e-03, grad_scale: 16.0 2023-04-17 03:31:56,935 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.07 vs. limit=5.0 2023-04-17 03:32:09,372 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.629e+02 2.492e+02 2.908e+02 3.443e+02 5.707e+02, threshold=5.815e+02, percent-clipped=0.0 2023-04-17 03:32:17,474 INFO [train.py:893] (1/4) Epoch 23, batch 1850, loss[loss=0.1738, simple_loss=0.2291, pruned_loss=0.05924, over 13539.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2349, pruned_loss=0.05627, over 2663325.80 frames. ], batch size: 83, lr: 5.65e-03, grad_scale: 32.0 2023-04-17 03:32:19,212 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-17 03:32:34,158 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5462, 3.3895, 4.0675, 2.9069, 2.7829, 2.8307, 4.4143, 4.5586], device='cuda:1'), covar=tensor([0.1129, 0.1746, 0.0435, 0.1791, 0.1692, 0.1551, 0.0302, 0.0250], device='cuda:1'), in_proj_covar=tensor([0.0248, 0.0271, 0.0196, 0.0225, 0.0219, 0.0183, 0.0213, 0.0184], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 03:32:56,088 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2881, 4.5145, 4.3186, 4.3816, 4.4178, 4.7434, 4.4644, 4.4639], device='cuda:1'), covar=tensor([0.0293, 0.0253, 0.0298, 0.0838, 0.0266, 0.0220, 0.0297, 0.0235], device='cuda:1'), in_proj_covar=tensor([0.0160, 0.0154, 0.0176, 0.0261, 0.0174, 0.0188, 0.0172, 0.0109], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0004, 0.0004, 0.0003], device='cuda:1') 2023-04-17 03:33:03,917 INFO [train.py:893] (1/4) Epoch 23, batch 1900, loss[loss=0.1414, simple_loss=0.2115, pruned_loss=0.0356, over 13403.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2351, pruned_loss=0.05677, over 2662002.00 frames. ], batch size: 88, lr: 5.65e-03, grad_scale: 32.0 2023-04-17 03:33:41,136 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.059e+02 2.633e+02 3.066e+02 3.485e+02 7.604e+02, threshold=6.132e+02, percent-clipped=3.0 2023-04-17 03:33:41,545 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2617, 3.5582, 3.5757, 3.9159, 2.1943, 3.0810, 3.7897, 2.2414], device='cuda:1'), covar=tensor([0.0134, 0.0523, 0.0740, 0.0469, 0.1702, 0.0911, 0.0510, 0.1728], device='cuda:1'), in_proj_covar=tensor([0.0132, 0.0186, 0.0207, 0.0247, 0.0183, 0.0198, 0.0180, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:33:49,772 INFO [train.py:893] (1/4) Epoch 23, batch 1950, loss[loss=0.1489, simple_loss=0.2156, pruned_loss=0.04116, over 13538.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2349, pruned_loss=0.05676, over 2664378.98 frames. ], batch size: 78, lr: 5.65e-03, grad_scale: 32.0 2023-04-17 03:34:12,152 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8613, 3.9919, 3.1548, 2.7953, 2.7759, 2.5301, 4.1536, 2.3512], device='cuda:1'), covar=tensor([0.1709, 0.0341, 0.1168, 0.2071, 0.0906, 0.3098, 0.0273, 0.4034], device='cuda:1'), in_proj_covar=tensor([0.0312, 0.0289, 0.0317, 0.0334, 0.0260, 0.0328, 0.0215, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 03:34:29,806 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61374.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:34:34,931 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2830, 2.6233, 2.4291, 4.2303, 4.7668, 3.4925, 4.6489, 4.3549], device='cuda:1'), covar=tensor([0.0108, 0.0941, 0.0982, 0.0106, 0.0077, 0.0485, 0.0079, 0.0094], device='cuda:1'), in_proj_covar=tensor([0.0078, 0.0092, 0.0099, 0.0082, 0.0069, 0.0082, 0.0057, 0.0079], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 03:34:35,481 INFO [train.py:893] (1/4) Epoch 23, batch 2000, loss[loss=0.1607, simple_loss=0.2269, pruned_loss=0.04724, over 13541.00 frames. ], tot_loss[loss=0.1761, simple_loss=0.2368, pruned_loss=0.05771, over 2659903.02 frames. ], batch size: 78, lr: 5.65e-03, grad_scale: 32.0 2023-04-17 03:34:40,483 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-17 03:34:40,802 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9046, 2.6717, 2.2509, 1.5022, 1.5955, 2.2380, 2.3235, 2.9367], device='cuda:1'), covar=tensor([0.0881, 0.0368, 0.0579, 0.1639, 0.0259, 0.0567, 0.0819, 0.0209], device='cuda:1'), in_proj_covar=tensor([0.0170, 0.0145, 0.0124, 0.0211, 0.0115, 0.0165, 0.0175, 0.0131], device='cuda:1'), out_proj_covar=tensor([1.2571e-04, 1.0848e-04, 9.6806e-05, 1.5708e-04, 8.3841e-05, 1.2465e-04, 1.3222e-04, 9.6615e-05], device='cuda:1') 2023-04-17 03:34:45,728 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3395, 2.9966, 2.9400, 3.3556, 2.8395, 3.4368, 3.2070, 2.8964], device='cuda:1'), covar=tensor([0.0084, 0.0194, 0.0128, 0.0148, 0.0167, 0.0131, 0.0166, 0.0142], device='cuda:1'), in_proj_covar=tensor([0.0106, 0.0119, 0.0125, 0.0125, 0.0134, 0.0112, 0.0109, 0.0109], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 03:35:08,621 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=61417.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:35:13,265 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.858e+02 2.773e+02 3.121e+02 3.784e+02 7.293e+02, threshold=6.243e+02, percent-clipped=1.0 2023-04-17 03:35:13,506 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=61422.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:35:22,150 INFO [train.py:893] (1/4) Epoch 23, batch 2050, loss[loss=0.1699, simple_loss=0.2345, pruned_loss=0.05264, over 13558.00 frames. ], tot_loss[loss=0.1771, simple_loss=0.238, pruned_loss=0.05808, over 2660482.66 frames. ], batch size: 89, lr: 5.64e-03, grad_scale: 32.0 2023-04-17 03:35:28,239 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5847, 3.3084, 4.0907, 2.9311, 2.7204, 2.9297, 4.3782, 4.4821], device='cuda:1'), covar=tensor([0.1140, 0.1764, 0.0396, 0.1763, 0.1677, 0.1508, 0.0293, 0.0200], device='cuda:1'), in_proj_covar=tensor([0.0249, 0.0272, 0.0197, 0.0226, 0.0220, 0.0183, 0.0213, 0.0185], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 03:36:06,620 INFO [train.py:893] (1/4) Epoch 23, batch 2100, loss[loss=0.1773, simple_loss=0.2337, pruned_loss=0.06048, over 13431.00 frames. ], tot_loss[loss=0.1772, simple_loss=0.2382, pruned_loss=0.05805, over 2663012.62 frames. ], batch size: 65, lr: 5.64e-03, grad_scale: 32.0 2023-04-17 03:36:45,708 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.854e+02 2.633e+02 3.023e+02 3.535e+02 5.995e+02, threshold=6.045e+02, percent-clipped=0.0 2023-04-17 03:36:53,714 INFO [train.py:893] (1/4) Epoch 23, batch 2150, loss[loss=0.1805, simple_loss=0.2462, pruned_loss=0.05737, over 13480.00 frames. ], tot_loss[loss=0.1769, simple_loss=0.2382, pruned_loss=0.05777, over 2662743.73 frames. ], batch size: 100, lr: 5.64e-03, grad_scale: 32.0 2023-04-17 03:36:59,139 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5873, 3.9659, 3.8007, 4.3014, 2.5042, 3.4046, 4.1295, 2.5291], device='cuda:1'), covar=tensor([0.0179, 0.0455, 0.0733, 0.0475, 0.1541, 0.0947, 0.0476, 0.1707], device='cuda:1'), in_proj_covar=tensor([0.0132, 0.0187, 0.0207, 0.0248, 0.0183, 0.0198, 0.0182, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:37:19,791 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-04-17 03:37:39,562 INFO [train.py:893] (1/4) Epoch 23, batch 2200, loss[loss=0.1615, simple_loss=0.2232, pruned_loss=0.04991, over 13339.00 frames. ], tot_loss[loss=0.1756, simple_loss=0.2372, pruned_loss=0.057, over 2663131.89 frames. ], batch size: 67, lr: 5.64e-03, grad_scale: 32.0 2023-04-17 03:37:40,265 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.95 vs. limit=5.0 2023-04-17 03:37:44,400 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3839, 3.8350, 3.6794, 4.1583, 2.2900, 3.3019, 3.9910, 2.3315], device='cuda:1'), covar=tensor([0.0202, 0.0480, 0.0763, 0.0650, 0.1679, 0.0891, 0.0455, 0.1738], device='cuda:1'), in_proj_covar=tensor([0.0133, 0.0187, 0.0208, 0.0249, 0.0183, 0.0199, 0.0182, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:38:19,059 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.670e+02 2.488e+02 2.952e+02 3.468e+02 6.307e+02, threshold=5.905e+02, percent-clipped=2.0 2023-04-17 03:38:26,950 INFO [train.py:893] (1/4) Epoch 23, batch 2250, loss[loss=0.1721, simple_loss=0.2359, pruned_loss=0.05415, over 13525.00 frames. ], tot_loss[loss=0.1749, simple_loss=0.2364, pruned_loss=0.05677, over 2662410.27 frames. ], batch size: 98, lr: 5.64e-03, grad_scale: 32.0 2023-04-17 03:39:11,154 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=2.66 vs. limit=5.0 2023-04-17 03:39:13,933 INFO [train.py:893] (1/4) Epoch 23, batch 2300, loss[loss=0.1735, simple_loss=0.2347, pruned_loss=0.05615, over 13527.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2352, pruned_loss=0.05623, over 2660966.51 frames. ], batch size: 83, lr: 5.63e-03, grad_scale: 32.0 2023-04-17 03:39:47,471 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=61717.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:39:51,395 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.815e+02 2.435e+02 2.861e+02 3.476e+02 7.164e+02, threshold=5.722e+02, percent-clipped=2.0 2023-04-17 03:39:58,987 INFO [train.py:893] (1/4) Epoch 23, batch 2350, loss[loss=0.192, simple_loss=0.2523, pruned_loss=0.06586, over 13384.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2343, pruned_loss=0.05574, over 2665119.93 frames. ], batch size: 109, lr: 5.63e-03, grad_scale: 32.0 2023-04-17 03:40:00,160 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61732.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:40:11,648 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61743.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:40:22,152 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-17 03:40:30,530 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=61765.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:40:46,469 INFO [train.py:893] (1/4) Epoch 23, batch 2400, loss[loss=0.1668, simple_loss=0.2344, pruned_loss=0.04958, over 13433.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2335, pruned_loss=0.05569, over 2666147.02 frames. ], batch size: 95, lr: 5.63e-03, grad_scale: 32.0 2023-04-17 03:40:50,816 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4565, 3.8062, 3.5092, 4.3720, 2.0495, 2.7527, 3.8712, 2.3202], device='cuda:1'), covar=tensor([0.0109, 0.0557, 0.0813, 0.0404, 0.1905, 0.1244, 0.0571, 0.1841], device='cuda:1'), in_proj_covar=tensor([0.0133, 0.0188, 0.0210, 0.0250, 0.0185, 0.0200, 0.0183, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:40:56,444 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61793.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:41:06,552 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61804.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:41:14,134 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7592, 3.8856, 2.9878, 2.6888, 2.7629, 2.4408, 3.9969, 2.2428], device='cuda:1'), covar=tensor([0.1802, 0.0373, 0.1304, 0.2337, 0.0930, 0.3418, 0.0308, 0.4288], device='cuda:1'), in_proj_covar=tensor([0.0314, 0.0291, 0.0318, 0.0338, 0.0263, 0.0332, 0.0215, 0.0368], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 03:41:22,685 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.841e+02 2.567e+02 2.922e+02 3.454e+02 5.739e+02, threshold=5.845e+02, percent-clipped=1.0 2023-04-17 03:41:30,293 INFO [train.py:893] (1/4) Epoch 23, batch 2450, loss[loss=0.1872, simple_loss=0.244, pruned_loss=0.06522, over 13548.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2337, pruned_loss=0.05583, over 2669502.70 frames. ], batch size: 83, lr: 5.63e-03, grad_scale: 32.0 2023-04-17 03:42:08,918 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-04-17 03:42:15,702 INFO [train.py:893] (1/4) Epoch 23, batch 2500, loss[loss=0.1853, simple_loss=0.2424, pruned_loss=0.0641, over 13225.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2332, pruned_loss=0.05557, over 2670451.72 frames. ], batch size: 117, lr: 5.62e-03, grad_scale: 32.0 2023-04-17 03:42:45,102 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=61914.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:42:51,233 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.719e+02 2.535e+02 2.894e+02 3.374e+02 4.864e+02, threshold=5.788e+02, percent-clipped=0.0 2023-04-17 03:42:59,186 INFO [train.py:893] (1/4) Epoch 23, batch 2550, loss[loss=0.1705, simple_loss=0.2324, pruned_loss=0.05424, over 13036.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2332, pruned_loss=0.05557, over 2668454.55 frames. ], batch size: 142, lr: 5.62e-03, grad_scale: 32.0 2023-04-17 03:43:23,784 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-17 03:43:40,263 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=61975.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:43:44,876 INFO [train.py:893] (1/4) Epoch 23, batch 2600, loss[loss=0.1782, simple_loss=0.2306, pruned_loss=0.06289, over 13517.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.233, pruned_loss=0.05565, over 2667890.16 frames. ], batch size: 83, lr: 5.62e-03, grad_scale: 32.0 2023-04-17 03:44:23,302 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.883e+02 2.551e+02 3.036e+02 3.753e+02 6.121e+02, threshold=6.072e+02, percent-clipped=2.0 2023-04-17 03:44:27,511 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.00 vs. limit=2.0 2023-04-17 03:44:29,951 INFO [train.py:893] (1/4) Epoch 23, batch 2650, loss[loss=0.1901, simple_loss=0.2575, pruned_loss=0.06133, over 13449.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2345, pruned_loss=0.05678, over 2664605.01 frames. ], batch size: 103, lr: 5.62e-03, grad_scale: 32.0 2023-04-17 03:45:25,833 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-17 03:45:35,579 INFO [train.py:893] (1/4) Epoch 24, batch 0, loss[loss=0.1515, simple_loss=0.2068, pruned_loss=0.04808, over 13529.00 frames. ], tot_loss[loss=0.1515, simple_loss=0.2068, pruned_loss=0.04808, over 13529.00 frames. ], batch size: 70, lr: 5.49e-03, grad_scale: 32.0 2023-04-17 03:45:35,580 INFO [train.py:918] (1/4) Computing validation loss 2023-04-17 03:45:42,523 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9177, 3.9894, 2.9394, 2.6396, 2.7792, 2.3599, 4.0844, 2.1852], device='cuda:1'), covar=tensor([0.1837, 0.0445, 0.1446, 0.2633, 0.1070, 0.3974, 0.0326, 0.5222], device='cuda:1'), in_proj_covar=tensor([0.0314, 0.0289, 0.0316, 0.0337, 0.0262, 0.0331, 0.0214, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 03:45:58,005 INFO [train.py:927] (1/4) Epoch 24, validation: loss=0.1345, simple_loss=0.1942, pruned_loss=0.03737, over 2446609.00 frames. 2023-04-17 03:45:58,006 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12831MB 2023-04-17 03:46:03,988 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6131, 3.5781, 3.6171, 2.3160, 3.6846, 3.7205, 3.6703, 3.8506], device='cuda:1'), covar=tensor([0.0294, 0.0154, 0.0164, 0.1379, 0.0169, 0.0265, 0.0133, 0.0139], device='cuda:1'), in_proj_covar=tensor([0.0094, 0.0057, 0.0085, 0.0103, 0.0100, 0.0111, 0.0083, 0.0078], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:46:05,640 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62088.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:46:14,735 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62099.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:46:19,637 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8463, 3.7638, 2.7380, 3.4124, 3.7923, 2.4259, 3.4450, 2.5212], device='cuda:1'), covar=tensor([0.0298, 0.0267, 0.1123, 0.0491, 0.0253, 0.1355, 0.0545, 0.1574], device='cuda:1'), in_proj_covar=tensor([0.0163, 0.0181, 0.0180, 0.0223, 0.0140, 0.0162, 0.0163, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:46:25,699 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.24 vs. limit=5.0 2023-04-17 03:46:36,057 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62122.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:46:36,523 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.935e+02 2.622e+02 3.024e+02 3.638e+02 5.971e+02, threshold=6.048e+02, percent-clipped=0.0 2023-04-17 03:46:42,604 INFO [train.py:893] (1/4) Epoch 24, batch 50, loss[loss=0.1588, simple_loss=0.2166, pruned_loss=0.05052, over 13519.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.228, pruned_loss=0.05558, over 601871.29 frames. ], batch size: 98, lr: 5.49e-03, grad_scale: 16.0 2023-04-17 03:47:06,722 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-17 03:47:06,723 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-17 03:47:06,723 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-17 03:47:06,729 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-17 03:47:06,737 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-17 03:47:06,758 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-17 03:47:07,501 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-17 03:47:29,857 INFO [train.py:893] (1/4) Epoch 24, batch 100, loss[loss=0.1905, simple_loss=0.2466, pruned_loss=0.06722, over 13531.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2276, pruned_loss=0.05541, over 1063602.14 frames. ], batch size: 85, lr: 5.49e-03, grad_scale: 16.0 2023-04-17 03:47:30,894 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8823, 4.3193, 4.1657, 4.1429, 4.1475, 4.0122, 4.3952, 4.4381], device='cuda:1'), covar=tensor([0.0267, 0.0293, 0.0246, 0.0405, 0.0327, 0.0319, 0.0276, 0.0268], device='cuda:1'), in_proj_covar=tensor([0.0139, 0.0202, 0.0166, 0.0185, 0.0155, 0.0203, 0.0137, 0.0181], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:47:32,613 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62183.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:47:33,380 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3126, 4.5262, 4.3625, 4.1522, 4.3741, 4.7230, 4.5188, 4.4588], device='cuda:1'), covar=tensor([0.0318, 0.0293, 0.0327, 0.1119, 0.0316, 0.0299, 0.0327, 0.0247], device='cuda:1'), in_proj_covar=tensor([0.0163, 0.0158, 0.0178, 0.0265, 0.0178, 0.0194, 0.0174, 0.0112], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0005, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 03:48:08,095 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.677e+02 3.136e+02 3.979e+02 7.437e+02, threshold=6.271e+02, percent-clipped=3.0 2023-04-17 03:48:14,808 INFO [train.py:893] (1/4) Epoch 24, batch 150, loss[loss=0.1692, simple_loss=0.2355, pruned_loss=0.0515, over 13380.00 frames. ], tot_loss[loss=0.174, simple_loss=0.2317, pruned_loss=0.05809, over 1402094.12 frames. ], batch size: 109, lr: 5.49e-03, grad_scale: 16.0 2023-04-17 03:48:51,385 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62270.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:48:55,590 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62273.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:49:02,021 INFO [train.py:893] (1/4) Epoch 24, batch 200, loss[loss=0.1815, simple_loss=0.2443, pruned_loss=0.05935, over 13382.00 frames. ], tot_loss[loss=0.1763, simple_loss=0.2348, pruned_loss=0.05885, over 1680215.44 frames. ], batch size: 109, lr: 5.49e-03, grad_scale: 16.0 2023-04-17 03:49:17,450 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6966, 2.3255, 2.4210, 2.7986, 2.1100, 2.7856, 2.6717, 2.2659], device='cuda:1'), covar=tensor([0.0085, 0.0213, 0.0152, 0.0153, 0.0217, 0.0130, 0.0168, 0.0203], device='cuda:1'), in_proj_covar=tensor([0.0106, 0.0120, 0.0125, 0.0126, 0.0135, 0.0112, 0.0109, 0.0111], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 03:49:37,746 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4959, 2.3322, 2.6876, 3.9545, 3.6175, 4.0814, 3.2029, 2.5562], device='cuda:1'), covar=tensor([0.0274, 0.0958, 0.0876, 0.0075, 0.0219, 0.0060, 0.0645, 0.0908], device='cuda:1'), in_proj_covar=tensor([0.0123, 0.0151, 0.0171, 0.0102, 0.0122, 0.0099, 0.0172, 0.0152], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:49:40,835 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.804e+02 2.766e+02 3.047e+02 3.449e+02 5.084e+02, threshold=6.094e+02, percent-clipped=0.0 2023-04-17 03:49:47,380 INFO [train.py:893] (1/4) Epoch 24, batch 250, loss[loss=0.1693, simple_loss=0.2318, pruned_loss=0.05346, over 13329.00 frames. ], tot_loss[loss=0.1766, simple_loss=0.2356, pruned_loss=0.05876, over 1899505.96 frames. ], batch size: 118, lr: 5.48e-03, grad_scale: 16.0 2023-04-17 03:49:50,904 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62334.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:50:33,907 INFO [train.py:893] (1/4) Epoch 24, batch 300, loss[loss=0.174, simple_loss=0.2457, pruned_loss=0.05119, over 13492.00 frames. ], tot_loss[loss=0.1759, simple_loss=0.2352, pruned_loss=0.05827, over 2070906.29 frames. ], batch size: 93, lr: 5.48e-03, grad_scale: 16.0 2023-04-17 03:50:41,561 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62388.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:50:44,053 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62391.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:50:52,431 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62399.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:51:10,514 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-04-17 03:51:14,251 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.671e+02 2.637e+02 2.996e+02 3.547e+02 6.141e+02, threshold=5.992e+02, percent-clipped=2.0 2023-04-17 03:51:21,379 INFO [train.py:893] (1/4) Epoch 24, batch 350, loss[loss=0.1451, simple_loss=0.2023, pruned_loss=0.04392, over 13418.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2355, pruned_loss=0.05822, over 2203759.95 frames. ], batch size: 65, lr: 5.48e-03, grad_scale: 16.0 2023-04-17 03:51:27,227 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62436.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:51:36,337 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62447.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:51:41,254 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62452.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:52:05,103 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62478.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:52:07,180 INFO [train.py:893] (1/4) Epoch 24, batch 400, loss[loss=0.1737, simple_loss=0.2363, pruned_loss=0.05552, over 13526.00 frames. ], tot_loss[loss=0.176, simple_loss=0.2362, pruned_loss=0.05788, over 2305614.46 frames. ], batch size: 72, lr: 5.48e-03, grad_scale: 16.0 2023-04-17 03:52:13,527 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=2.59 vs. limit=5.0 2023-04-17 03:52:39,663 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.8350, 4.3191, 4.3217, 4.3484, 4.1489, 4.1198, 4.7995, 4.3322], device='cuda:1'), covar=tensor([0.0906, 0.1470, 0.2379, 0.2917, 0.1198, 0.1873, 0.1049, 0.1506], device='cuda:1'), in_proj_covar=tensor([0.0341, 0.0393, 0.0481, 0.0484, 0.0307, 0.0362, 0.0448, 0.0360], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 03:52:39,908 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-04-17 03:52:46,748 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.724e+02 2.666e+02 3.083e+02 3.648e+02 7.210e+02, threshold=6.167e+02, percent-clipped=2.0 2023-04-17 03:52:54,081 INFO [train.py:893] (1/4) Epoch 24, batch 450, loss[loss=0.165, simple_loss=0.2287, pruned_loss=0.05062, over 13459.00 frames. ], tot_loss[loss=0.1768, simple_loss=0.237, pruned_loss=0.05828, over 2383807.78 frames. ], batch size: 79, lr: 5.48e-03, grad_scale: 16.0 2023-04-17 03:52:56,932 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-04-17 03:53:07,479 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62545.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:53:18,454 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-17 03:53:30,053 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62570.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:53:38,743 INFO [train.py:893] (1/4) Epoch 24, batch 500, loss[loss=0.1799, simple_loss=0.2426, pruned_loss=0.05864, over 13544.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2362, pruned_loss=0.05757, over 2448993.83 frames. ], batch size: 85, lr: 5.47e-03, grad_scale: 16.0 2023-04-17 03:53:43,738 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62585.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:54:03,836 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62606.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:54:05,839 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 2023-04-17 03:54:09,996 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-04-17 03:54:14,536 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62618.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:54:19,281 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.673e+02 2.688e+02 3.153e+02 3.603e+02 8.252e+02, threshold=6.305e+02, percent-clipped=3.0 2023-04-17 03:54:25,246 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62629.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:54:25,852 INFO [train.py:893] (1/4) Epoch 24, batch 550, loss[loss=0.1805, simple_loss=0.2353, pruned_loss=0.06284, over 13266.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2362, pruned_loss=0.05745, over 2495059.00 frames. ], batch size: 124, lr: 5.47e-03, grad_scale: 16.0 2023-04-17 03:54:40,302 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62646.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:54:59,644 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62666.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:55:11,613 INFO [train.py:893] (1/4) Epoch 24, batch 600, loss[loss=0.1316, simple_loss=0.1981, pruned_loss=0.03253, over 13484.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2353, pruned_loss=0.05706, over 2532982.91 frames. ], batch size: 81, lr: 5.47e-03, grad_scale: 16.0 2023-04-17 03:55:13,897 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.90 vs. limit=5.0 2023-04-17 03:55:14,627 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-04-17 03:55:23,010 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-04-17 03:55:51,671 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.567e+02 2.553e+02 3.005e+02 3.508e+02 5.989e+02, threshold=6.009e+02, percent-clipped=0.0 2023-04-17 03:55:55,544 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62727.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:55:58,430 INFO [train.py:893] (1/4) Epoch 24, batch 650, loss[loss=0.1588, simple_loss=0.2309, pruned_loss=0.04336, over 13453.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2345, pruned_loss=0.05662, over 2552208.37 frames. ], batch size: 95, lr: 5.47e-03, grad_scale: 16.0 2023-04-17 03:56:13,672 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62747.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:56:37,421 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=62773.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:56:42,873 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62778.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:56:44,228 INFO [train.py:893] (1/4) Epoch 24, batch 700, loss[loss=0.1738, simple_loss=0.2356, pruned_loss=0.05599, over 13455.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2336, pruned_loss=0.05588, over 2577362.37 frames. ], batch size: 79, lr: 5.46e-03, grad_scale: 16.0 2023-04-17 03:57:24,529 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.679e+02 2.589e+02 3.020e+02 3.370e+02 5.327e+02, threshold=6.040e+02, percent-clipped=0.0 2023-04-17 03:57:26,595 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4285, 4.2088, 4.3118, 2.8478, 4.7487, 4.5117, 4.4951, 4.6824], device='cuda:1'), covar=tensor([0.0255, 0.0161, 0.0178, 0.1041, 0.0146, 0.0231, 0.0133, 0.0104], device='cuda:1'), in_proj_covar=tensor([0.0093, 0.0057, 0.0084, 0.0101, 0.0099, 0.0110, 0.0081, 0.0077], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 03:57:27,336 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62826.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:57:30,677 INFO [train.py:893] (1/4) Epoch 24, batch 750, loss[loss=0.1802, simple_loss=0.2379, pruned_loss=0.06127, over 13084.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2335, pruned_loss=0.05617, over 2597510.25 frames. ], batch size: 142, lr: 5.46e-03, grad_scale: 16.0 2023-04-17 03:57:34,420 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=62834.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:57:37,535 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0871, 2.5914, 2.7520, 3.1483, 2.4625, 3.2340, 3.0678, 2.5707], device='cuda:1'), covar=tensor([0.0099, 0.0219, 0.0161, 0.0167, 0.0215, 0.0137, 0.0180, 0.0225], device='cuda:1'), in_proj_covar=tensor([0.0108, 0.0123, 0.0128, 0.0128, 0.0138, 0.0115, 0.0112, 0.0114], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 03:58:11,920 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0696, 1.8748, 3.5494, 3.4869, 3.3477, 2.8431, 3.2037, 2.6736], device='cuda:1'), covar=tensor([0.2122, 0.1650, 0.0214, 0.0218, 0.0341, 0.0789, 0.0319, 0.0890], device='cuda:1'), in_proj_covar=tensor([0.0216, 0.0185, 0.0128, 0.0131, 0.0136, 0.0177, 0.0147, 0.0185], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-17 03:58:12,850 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1644, 2.9225, 3.4688, 2.6953, 2.3862, 2.4898, 3.8591, 3.9221], device='cuda:1'), covar=tensor([0.1317, 0.2072, 0.0406, 0.1667, 0.1685, 0.1679, 0.0327, 0.0208], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0274, 0.0200, 0.0226, 0.0221, 0.0183, 0.0216, 0.0188], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 03:58:17,435 INFO [train.py:893] (1/4) Epoch 24, batch 800, loss[loss=0.2026, simple_loss=0.2496, pruned_loss=0.07781, over 13445.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2343, pruned_loss=0.05674, over 2605599.49 frames. ], batch size: 65, lr: 5.46e-03, grad_scale: 16.0 2023-04-17 03:58:36,571 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62901.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:58:56,775 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.925e+02 2.701e+02 3.020e+02 3.657e+02 6.500e+02, threshold=6.040e+02, percent-clipped=1.0 2023-04-17 03:59:02,104 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=62929.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:59:02,713 INFO [train.py:893] (1/4) Epoch 24, batch 850, loss[loss=0.2006, simple_loss=0.2611, pruned_loss=0.07007, over 13545.00 frames. ], tot_loss[loss=0.1755, simple_loss=0.2362, pruned_loss=0.05747, over 2620264.47 frames. ], batch size: 85, lr: 5.46e-03, grad_scale: 16.0 2023-04-17 03:59:14,281 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=62941.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:59:26,949 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-04-17 03:59:48,076 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=62977.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 03:59:50,403 INFO [train.py:893] (1/4) Epoch 24, batch 900, loss[loss=0.1684, simple_loss=0.2318, pruned_loss=0.05246, over 13519.00 frames. ], tot_loss[loss=0.1757, simple_loss=0.2361, pruned_loss=0.05767, over 2631897.67 frames. ], batch size: 85, lr: 5.46e-03, grad_scale: 16.0 2023-04-17 04:00:21,163 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-17 04:00:28,790 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63022.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:00:29,470 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.868e+02 2.672e+02 3.167e+02 4.016e+02 1.001e+03, threshold=6.335e+02, percent-clipped=2.0 2023-04-17 04:00:36,751 INFO [train.py:893] (1/4) Epoch 24, batch 950, loss[loss=0.1938, simple_loss=0.2374, pruned_loss=0.07506, over 11945.00 frames. ], tot_loss[loss=0.175, simple_loss=0.2351, pruned_loss=0.05739, over 2639206.74 frames. ], batch size: 157, lr: 5.45e-03, grad_scale: 16.0 2023-04-17 04:00:52,699 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63047.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:01:13,337 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0649, 1.9013, 3.8028, 3.6791, 3.5829, 2.9342, 3.4265, 2.7777], device='cuda:1'), covar=tensor([0.2039, 0.1692, 0.0163, 0.0196, 0.0286, 0.0775, 0.0279, 0.0843], device='cuda:1'), in_proj_covar=tensor([0.0215, 0.0184, 0.0127, 0.0132, 0.0136, 0.0175, 0.0147, 0.0184], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-17 04:01:22,940 INFO [train.py:893] (1/4) Epoch 24, batch 1000, loss[loss=0.1739, simple_loss=0.2357, pruned_loss=0.05609, over 13473.00 frames. ], tot_loss[loss=0.173, simple_loss=0.2329, pruned_loss=0.05651, over 2645568.73 frames. ], batch size: 79, lr: 5.45e-03, grad_scale: 16.0 2023-04-17 04:01:36,835 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63095.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:01:49,137 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0655, 2.3885, 1.9222, 3.9063, 4.3882, 3.3184, 4.2366, 4.1200], device='cuda:1'), covar=tensor([0.0106, 0.1027, 0.1162, 0.0108, 0.0078, 0.0472, 0.0109, 0.0098], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0094, 0.0102, 0.0084, 0.0071, 0.0084, 0.0059, 0.0081], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 04:02:02,204 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.910e+02 2.653e+02 3.155e+02 3.689e+02 7.228e+02, threshold=6.310e+02, percent-clipped=3.0 2023-04-17 04:02:08,129 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63129.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:02:08,727 INFO [train.py:893] (1/4) Epoch 24, batch 1050, loss[loss=0.1814, simple_loss=0.2373, pruned_loss=0.06273, over 13253.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2318, pruned_loss=0.05575, over 2651301.96 frames. ], batch size: 124, lr: 5.45e-03, grad_scale: 16.0 2023-04-17 04:02:55,242 INFO [train.py:893] (1/4) Epoch 24, batch 1100, loss[loss=0.1956, simple_loss=0.2553, pruned_loss=0.06791, over 13526.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2327, pruned_loss=0.05536, over 2654240.87 frames. ], batch size: 83, lr: 5.45e-03, grad_scale: 16.0 2023-04-17 04:03:03,663 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7047, 3.8440, 2.7411, 3.5848, 3.7853, 2.5224, 3.1846, 2.8459], device='cuda:1'), covar=tensor([0.0324, 0.0373, 0.0962, 0.0362, 0.0327, 0.1067, 0.0700, 0.0900], device='cuda:1'), in_proj_covar=tensor([0.0163, 0.0181, 0.0180, 0.0226, 0.0141, 0.0162, 0.0163, 0.0159], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:03:16,472 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63201.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:03:33,924 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63221.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:03:35,169 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.822e+02 2.654e+02 2.983e+02 3.494e+02 4.938e+02, threshold=5.966e+02, percent-clipped=0.0 2023-04-17 04:03:42,640 INFO [train.py:893] (1/4) Epoch 24, batch 1150, loss[loss=0.1813, simple_loss=0.2437, pruned_loss=0.05946, over 13069.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2324, pruned_loss=0.05494, over 2654986.05 frames. ], batch size: 142, lr: 5.45e-03, grad_scale: 16.0 2023-04-17 04:03:52,086 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63241.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:03:59,620 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63249.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:04:11,702 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-04-17 04:04:27,369 INFO [train.py:893] (1/4) Epoch 24, batch 1200, loss[loss=0.191, simple_loss=0.2442, pruned_loss=0.06889, over 13508.00 frames. ], tot_loss[loss=0.1717, simple_loss=0.2333, pruned_loss=0.05509, over 2658303.51 frames. ], batch size: 91, lr: 5.44e-03, grad_scale: 16.0 2023-04-17 04:04:29,357 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63282.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:04:35,118 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63289.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:04:54,559 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-17 04:05:06,040 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-17 04:05:06,350 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63322.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:05:06,508 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7918, 3.9350, 3.0765, 2.7324, 2.7422, 2.4954, 3.9877, 2.3065], device='cuda:1'), covar=tensor([0.1726, 0.0332, 0.1235, 0.2173, 0.0949, 0.3317, 0.0300, 0.4303], device='cuda:1'), in_proj_covar=tensor([0.0315, 0.0293, 0.0322, 0.0342, 0.0264, 0.0333, 0.0218, 0.0371], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 04:05:06,918 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.834e+02 2.724e+02 3.082e+02 3.852e+02 6.773e+02, threshold=6.164e+02, percent-clipped=2.0 2023-04-17 04:05:13,734 INFO [train.py:893] (1/4) Epoch 24, batch 1250, loss[loss=0.1621, simple_loss=0.2263, pruned_loss=0.04894, over 13340.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2336, pruned_loss=0.05541, over 2657059.23 frames. ], batch size: 67, lr: 5.44e-03, grad_scale: 16.0 2023-04-17 04:05:41,466 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6926, 4.5348, 4.7347, 4.6923, 4.9787, 4.4864, 5.0364, 4.9765], device='cuda:1'), covar=tensor([0.0432, 0.0542, 0.0639, 0.0527, 0.0563, 0.0919, 0.0407, 0.0459], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0302, 0.0311, 0.0232, 0.0449, 0.0351, 0.0284, 0.0307], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:05:50,731 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63370.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:05:59,085 INFO [train.py:893] (1/4) Epoch 24, batch 1300, loss[loss=0.1876, simple_loss=0.2499, pruned_loss=0.06266, over 13360.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2349, pruned_loss=0.05564, over 2661098.10 frames. ], batch size: 118, lr: 5.44e-03, grad_scale: 16.0 2023-04-17 04:06:08,469 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9512, 4.2716, 3.2990, 2.8996, 3.0208, 2.6558, 4.3980, 2.5313], device='cuda:1'), covar=tensor([0.1838, 0.0301, 0.1222, 0.2181, 0.0874, 0.3411, 0.0237, 0.4083], device='cuda:1'), in_proj_covar=tensor([0.0317, 0.0295, 0.0323, 0.0343, 0.0265, 0.0335, 0.0219, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 04:06:27,731 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1096, 4.5763, 4.3604, 4.3675, 4.4289, 4.1366, 4.6321, 4.6812], device='cuda:1'), covar=tensor([0.0271, 0.0244, 0.0282, 0.0383, 0.0286, 0.0334, 0.0319, 0.0235], device='cuda:1'), in_proj_covar=tensor([0.0144, 0.0208, 0.0170, 0.0189, 0.0159, 0.0207, 0.0138, 0.0186], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:06:37,570 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63420.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:06:39,204 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63422.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:06:39,715 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.944e+02 2.696e+02 3.109e+02 3.925e+02 6.868e+02, threshold=6.218e+02, percent-clipped=1.0 2023-04-17 04:06:44,972 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63429.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:06:45,533 INFO [train.py:893] (1/4) Epoch 24, batch 1350, loss[loss=0.1792, simple_loss=0.2437, pruned_loss=0.05732, over 13531.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2353, pruned_loss=0.05596, over 2661236.78 frames. ], batch size: 87, lr: 5.44e-03, grad_scale: 16.0 2023-04-17 04:07:30,470 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63477.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:07:32,652 INFO [train.py:893] (1/4) Epoch 24, batch 1400, loss[loss=0.1909, simple_loss=0.2496, pruned_loss=0.06613, over 13527.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2349, pruned_loss=0.05574, over 2662321.97 frames. ], batch size: 70, lr: 5.43e-03, grad_scale: 16.0 2023-04-17 04:07:34,606 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63481.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:07:36,245 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63483.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:07:48,657 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2461, 4.7066, 4.4937, 4.4767, 4.4616, 4.3079, 4.7851, 4.8019], device='cuda:1'), covar=tensor([0.0213, 0.0209, 0.0179, 0.0323, 0.0302, 0.0248, 0.0224, 0.0199], device='cuda:1'), in_proj_covar=tensor([0.0142, 0.0206, 0.0168, 0.0187, 0.0158, 0.0205, 0.0136, 0.0184], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:07:53,157 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-04-17 04:08:12,179 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.671e+02 2.694e+02 3.164e+02 3.726e+02 7.551e+02, threshold=6.327e+02, percent-clipped=1.0 2023-04-17 04:08:18,627 INFO [train.py:893] (1/4) Epoch 24, batch 1450, loss[loss=0.1759, simple_loss=0.2476, pruned_loss=0.05208, over 13361.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2336, pruned_loss=0.05544, over 2663717.39 frames. ], batch size: 109, lr: 5.43e-03, grad_scale: 16.0 2023-04-17 04:08:28,957 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.28 vs. limit=5.0 2023-04-17 04:08:55,236 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63569.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:09:02,492 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63577.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:09:04,241 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1779, 4.2179, 3.0065, 3.8539, 4.1271, 2.6962, 3.7797, 2.8451], device='cuda:1'), covar=tensor([0.0243, 0.0199, 0.0979, 0.0434, 0.0198, 0.1221, 0.0453, 0.1191], device='cuda:1'), in_proj_covar=tensor([0.0162, 0.0180, 0.0179, 0.0226, 0.0141, 0.0161, 0.0163, 0.0159], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:09:05,523 INFO [train.py:893] (1/4) Epoch 24, batch 1500, loss[loss=0.1624, simple_loss=0.2266, pruned_loss=0.04909, over 13474.00 frames. ], tot_loss[loss=0.172, simple_loss=0.2335, pruned_loss=0.05525, over 2666522.64 frames. ], batch size: 79, lr: 5.43e-03, grad_scale: 16.0 2023-04-17 04:09:10,737 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63586.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:09:39,257 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5465, 4.5735, 3.3369, 4.1745, 4.4658, 3.0171, 4.0887, 3.0864], device='cuda:1'), covar=tensor([0.0238, 0.0199, 0.0948, 0.0438, 0.0173, 0.1076, 0.0356, 0.1242], device='cuda:1'), in_proj_covar=tensor([0.0163, 0.0181, 0.0179, 0.0226, 0.0141, 0.0162, 0.0164, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:09:42,163 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.79 vs. limit=5.0 2023-04-17 04:09:44,963 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.034e+02 2.667e+02 3.072e+02 3.554e+02 8.313e+02, threshold=6.144e+02, percent-clipped=1.0 2023-04-17 04:09:52,388 INFO [train.py:893] (1/4) Epoch 24, batch 1550, loss[loss=0.1639, simple_loss=0.2294, pruned_loss=0.04917, over 13454.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2333, pruned_loss=0.05511, over 2661558.01 frames. ], batch size: 79, lr: 5.43e-03, grad_scale: 16.0 2023-04-17 04:09:52,676 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63630.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:09:56,744 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4253, 3.1585, 3.7934, 2.6957, 2.5290, 2.6345, 4.1234, 4.2548], device='cuda:1'), covar=tensor([0.1124, 0.2000, 0.0418, 0.1769, 0.1679, 0.1520, 0.0297, 0.0182], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0275, 0.0201, 0.0225, 0.0221, 0.0184, 0.0215, 0.0187], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 04:10:08,276 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63647.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:10:36,638 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6449, 4.0457, 3.8615, 4.4713, 2.4976, 3.3927, 4.2495, 2.5381], device='cuda:1'), covar=tensor([0.0123, 0.0420, 0.0716, 0.0443, 0.1576, 0.0848, 0.0403, 0.1630], device='cuda:1'), in_proj_covar=tensor([0.0134, 0.0187, 0.0207, 0.0248, 0.0185, 0.0199, 0.0180, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:10:38,781 INFO [train.py:893] (1/4) Epoch 24, batch 1600, loss[loss=0.1616, simple_loss=0.2334, pruned_loss=0.04491, over 13460.00 frames. ], tot_loss[loss=0.1721, simple_loss=0.2341, pruned_loss=0.05503, over 2663814.86 frames. ], batch size: 106, lr: 5.43e-03, grad_scale: 16.0 2023-04-17 04:11:19,070 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.884e+02 2.516e+02 2.981e+02 3.582e+02 5.645e+02, threshold=5.962e+02, percent-clipped=0.0 2023-04-17 04:11:25,481 INFO [train.py:893] (1/4) Epoch 24, batch 1650, loss[loss=0.1825, simple_loss=0.2475, pruned_loss=0.05878, over 13069.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2349, pruned_loss=0.05489, over 2662884.03 frames. ], batch size: 142, lr: 5.42e-03, grad_scale: 16.0 2023-04-17 04:11:32,101 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8504, 4.0421, 3.0598, 2.7550, 2.9267, 2.5027, 4.1316, 2.3662], device='cuda:1'), covar=tensor([0.1728, 0.0359, 0.1230, 0.2230, 0.0843, 0.3274, 0.0269, 0.3808], device='cuda:1'), in_proj_covar=tensor([0.0316, 0.0294, 0.0322, 0.0342, 0.0264, 0.0335, 0.0219, 0.0370], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 04:12:08,794 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63776.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:12:11,256 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63778.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:12:12,623 INFO [train.py:893] (1/4) Epoch 24, batch 1700, loss[loss=0.1744, simple_loss=0.2328, pruned_loss=0.05805, over 13534.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2352, pruned_loss=0.05477, over 2665096.87 frames. ], batch size: 83, lr: 5.42e-03, grad_scale: 16.0 2023-04-17 04:12:41,180 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63810.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:12:53,069 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.875e+02 2.594e+02 3.016e+02 3.621e+02 1.016e+03, threshold=6.032e+02, percent-clipped=3.0 2023-04-17 04:12:59,064 INFO [train.py:893] (1/4) Epoch 24, batch 1750, loss[loss=0.1571, simple_loss=0.2246, pruned_loss=0.04479, over 13415.00 frames. ], tot_loss[loss=0.1712, simple_loss=0.2342, pruned_loss=0.05417, over 2665098.90 frames. ], batch size: 65, lr: 5.42e-03, grad_scale: 16.0 2023-04-17 04:13:09,381 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-04-17 04:13:13,889 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63845.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:13:17,311 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.3688, 2.2775, 4.3121, 3.9887, 4.1354, 3.4329, 3.7409, 3.1792], device='cuda:1'), covar=tensor([0.1764, 0.1487, 0.0083, 0.0202, 0.0156, 0.0526, 0.0273, 0.0671], device='cuda:1'), in_proj_covar=tensor([0.0216, 0.0182, 0.0127, 0.0130, 0.0133, 0.0173, 0.0146, 0.0183], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-17 04:13:30,767 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.99 vs. limit=2.0 2023-04-17 04:13:38,052 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63871.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:13:43,696 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=63877.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:13:46,034 INFO [train.py:893] (1/4) Epoch 24, batch 1800, loss[loss=0.173, simple_loss=0.2264, pruned_loss=0.05983, over 9456.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2331, pruned_loss=0.05377, over 2661581.22 frames. ], batch size: 38, lr: 5.42e-03, grad_scale: 16.0 2023-04-17 04:13:54,453 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.1325, 4.6123, 4.5310, 4.6255, 4.3987, 4.4447, 5.0946, 4.6697], device='cuda:1'), covar=tensor([0.0785, 0.1413, 0.2173, 0.2737, 0.1064, 0.1715, 0.0963, 0.1204], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0397, 0.0489, 0.0493, 0.0313, 0.0366, 0.0458, 0.0364], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 04:14:10,141 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63906.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:14:19,821 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=63916.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:14:25,502 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.929e+02 2.462e+02 2.949e+02 3.735e+02 7.345e+02, threshold=5.897e+02, percent-clipped=3.0 2023-04-17 04:14:28,144 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=63925.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:14:28,194 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63925.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:14:32,094 INFO [train.py:893] (1/4) Epoch 24, batch 1850, loss[loss=0.2026, simple_loss=0.2557, pruned_loss=0.07476, over 13422.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.233, pruned_loss=0.05377, over 2659480.58 frames. ], batch size: 95, lr: 5.42e-03, grad_scale: 16.0 2023-04-17 04:14:37,107 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-17 04:14:40,678 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7879, 2.6806, 3.1899, 4.4090, 3.8989, 4.3936, 3.4532, 2.8132], device='cuda:1'), covar=tensor([0.0350, 0.0995, 0.0740, 0.0054, 0.0263, 0.0073, 0.0612, 0.0915], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0147, 0.0166, 0.0100, 0.0121, 0.0098, 0.0167, 0.0149], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:14:43,838 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=63942.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:15:17,192 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=63977.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:15:19,404 INFO [train.py:893] (1/4) Epoch 24, batch 1900, loss[loss=0.1497, simple_loss=0.218, pruned_loss=0.04065, over 13474.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2325, pruned_loss=0.05396, over 2660580.08 frames. ], batch size: 79, lr: 5.41e-03, grad_scale: 16.0 2023-04-17 04:15:42,345 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.15 vs. limit=5.0 2023-04-17 04:16:03,220 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.832e+02 2.373e+02 2.755e+02 3.277e+02 5.989e+02, threshold=5.509e+02, percent-clipped=1.0 2023-04-17 04:16:03,520 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64023.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:16:09,826 INFO [train.py:893] (1/4) Epoch 24, batch 1950, loss[loss=0.1519, simple_loss=0.1986, pruned_loss=0.05256, over 12260.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2319, pruned_loss=0.054, over 2657405.63 frames. ], batch size: 50, lr: 5.41e-03, grad_scale: 16.0 2023-04-17 04:16:30,445 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7667, 3.4293, 4.2989, 2.9894, 2.8891, 2.9673, 4.5919, 4.6702], device='cuda:1'), covar=tensor([0.1126, 0.1809, 0.0326, 0.1796, 0.1573, 0.1664, 0.0239, 0.0284], device='cuda:1'), in_proj_covar=tensor([0.0253, 0.0276, 0.0201, 0.0226, 0.0223, 0.0185, 0.0216, 0.0188], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 04:16:44,373 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7753, 3.5585, 3.7584, 2.3812, 3.8077, 3.8222, 3.7178, 3.9181], device='cuda:1'), covar=tensor([0.0203, 0.0151, 0.0129, 0.1069, 0.0115, 0.0195, 0.0116, 0.0100], device='cuda:1'), in_proj_covar=tensor([0.0097, 0.0059, 0.0087, 0.0105, 0.0102, 0.0114, 0.0085, 0.0080], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:16:52,604 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64076.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:16:54,202 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64078.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:16:55,632 INFO [train.py:893] (1/4) Epoch 24, batch 2000, loss[loss=0.1393, simple_loss=0.1949, pruned_loss=0.04187, over 12705.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2343, pruned_loss=0.05517, over 2655730.58 frames. ], batch size: 52, lr: 5.41e-03, grad_scale: 16.0 2023-04-17 04:16:59,944 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64084.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:17:02,988 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-17 04:17:31,641 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=2.05 vs. limit=2.0 2023-04-17 04:17:35,975 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.856e+02 2.644e+02 3.150e+02 3.695e+02 7.820e+02, threshold=6.300e+02, percent-clipped=3.0 2023-04-17 04:17:36,245 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.1834, 4.6780, 4.6086, 4.7068, 4.4643, 4.5207, 5.1330, 4.6826], device='cuda:1'), covar=tensor([0.0719, 0.1255, 0.1974, 0.2380, 0.0934, 0.1588, 0.0895, 0.1034], device='cuda:1'), in_proj_covar=tensor([0.0350, 0.0399, 0.0492, 0.0495, 0.0311, 0.0366, 0.0458, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 04:17:37,644 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64124.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:17:37,857 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8504, 4.1492, 3.1938, 2.8859, 2.8650, 2.6425, 4.2172, 2.4474], device='cuda:1'), covar=tensor([0.1961, 0.0387, 0.1284, 0.2238, 0.1003, 0.3400, 0.0303, 0.4356], device='cuda:1'), in_proj_covar=tensor([0.0314, 0.0292, 0.0322, 0.0341, 0.0263, 0.0333, 0.0218, 0.0370], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 04:17:39,272 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64126.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:17:43,181 INFO [train.py:893] (1/4) Epoch 24, batch 2050, loss[loss=0.1643, simple_loss=0.2246, pruned_loss=0.05199, over 13434.00 frames. ], tot_loss[loss=0.1739, simple_loss=0.2359, pruned_loss=0.056, over 2657898.75 frames. ], batch size: 65, lr: 5.41e-03, grad_scale: 32.0 2023-04-17 04:18:00,241 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0026, 4.2816, 3.2340, 2.9331, 3.0878, 2.6285, 4.3761, 2.4562], device='cuda:1'), covar=tensor([0.1795, 0.0340, 0.1312, 0.2237, 0.0870, 0.3339, 0.0282, 0.3976], device='cuda:1'), in_proj_covar=tensor([0.0314, 0.0292, 0.0322, 0.0341, 0.0263, 0.0333, 0.0218, 0.0369], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 04:18:07,981 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-04-17 04:18:16,685 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64166.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:18:29,318 INFO [train.py:893] (1/4) Epoch 24, batch 2100, loss[loss=0.1725, simple_loss=0.2428, pruned_loss=0.05112, over 13446.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2349, pruned_loss=0.05527, over 2662122.03 frames. ], batch size: 106, lr: 5.40e-03, grad_scale: 32.0 2023-04-17 04:18:49,019 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64201.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:19:09,884 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.961e+02 2.501e+02 2.846e+02 3.361e+02 6.447e+02, threshold=5.692e+02, percent-clipped=1.0 2023-04-17 04:19:12,567 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64225.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:19:16,489 INFO [train.py:893] (1/4) Epoch 24, batch 2150, loss[loss=0.184, simple_loss=0.2432, pruned_loss=0.06245, over 13380.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2352, pruned_loss=0.05482, over 2664773.21 frames. ], batch size: 109, lr: 5.40e-03, grad_scale: 32.0 2023-04-17 04:19:18,407 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64232.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:19:28,294 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64242.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:19:52,689 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.6132, 2.5044, 2.1953, 1.4639, 1.4817, 2.0367, 2.0902, 2.6702], device='cuda:1'), covar=tensor([0.1173, 0.0383, 0.0793, 0.1835, 0.0231, 0.0587, 0.0903, 0.0217], device='cuda:1'), in_proj_covar=tensor([0.0174, 0.0150, 0.0128, 0.0214, 0.0117, 0.0169, 0.0182, 0.0139], device='cuda:1'), out_proj_covar=tensor([1.2884e-04, 1.1242e-04, 9.9155e-05, 1.5885e-04, 8.4750e-05, 1.2816e-04, 1.3687e-04, 1.0293e-04], device='cuda:1') 2023-04-17 04:19:55,003 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64272.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:19:56,588 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64273.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:19:57,071 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-04-17 04:20:02,256 INFO [train.py:893] (1/4) Epoch 24, batch 2200, loss[loss=0.1774, simple_loss=0.2376, pruned_loss=0.05866, over 13392.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2346, pruned_loss=0.05457, over 2664526.44 frames. ], batch size: 84, lr: 5.40e-03, grad_scale: 32.0 2023-04-17 04:20:12,109 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64290.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:20:14,754 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64293.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:20:42,710 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.722e+02 2.378e+02 2.714e+02 3.263e+02 6.451e+02, threshold=5.428e+02, percent-clipped=1.0 2023-04-17 04:20:48,358 INFO [train.py:893] (1/4) Epoch 24, batch 2250, loss[loss=0.1469, simple_loss=0.2084, pruned_loss=0.04274, over 13412.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2332, pruned_loss=0.05412, over 2659954.00 frames. ], batch size: 65, lr: 5.40e-03, grad_scale: 32.0 2023-04-17 04:21:02,867 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64345.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:21:34,732 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64379.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:21:35,366 INFO [train.py:893] (1/4) Epoch 24, batch 2300, loss[loss=0.1942, simple_loss=0.2542, pruned_loss=0.06709, over 13511.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2328, pruned_loss=0.05387, over 2661315.85 frames. ], batch size: 91, lr: 5.40e-03, grad_scale: 32.0 2023-04-17 04:21:51,485 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7671, 3.7199, 4.2408, 3.0878, 2.8098, 2.8555, 4.7238, 4.8511], device='cuda:1'), covar=tensor([0.1011, 0.1399, 0.0369, 0.1542, 0.1521, 0.1451, 0.0225, 0.0148], device='cuda:1'), in_proj_covar=tensor([0.0254, 0.0276, 0.0201, 0.0226, 0.0222, 0.0184, 0.0215, 0.0189], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 04:22:00,086 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64406.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:22:15,076 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.764e+02 2.581e+02 2.900e+02 3.410e+02 5.772e+02, threshold=5.800e+02, percent-clipped=1.0 2023-04-17 04:22:20,671 INFO [train.py:893] (1/4) Epoch 24, batch 2350, loss[loss=0.1738, simple_loss=0.2337, pruned_loss=0.05688, over 13268.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2323, pruned_loss=0.05378, over 2662114.95 frames. ], batch size: 124, lr: 5.39e-03, grad_scale: 32.0 2023-04-17 04:22:46,535 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-17 04:22:55,342 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64465.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:22:56,179 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64466.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:22:58,038 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.73 vs. limit=5.0 2023-04-17 04:23:08,870 INFO [train.py:893] (1/4) Epoch 24, batch 2400, loss[loss=0.1767, simple_loss=0.233, pruned_loss=0.06018, over 13364.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2316, pruned_loss=0.05405, over 2658630.83 frames. ], batch size: 118, lr: 5.39e-03, grad_scale: 32.0 2023-04-17 04:23:28,376 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64501.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:23:28,458 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3579, 2.2726, 2.6189, 3.7879, 3.4222, 3.8721, 2.9141, 2.3573], device='cuda:1'), covar=tensor([0.0322, 0.0860, 0.0800, 0.0078, 0.0267, 0.0061, 0.0678, 0.0897], device='cuda:1'), in_proj_covar=tensor([0.0123, 0.0151, 0.0168, 0.0102, 0.0123, 0.0101, 0.0171, 0.0152], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:23:40,521 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64514.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:23:47,922 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.856e+02 2.611e+02 3.047e+02 3.549e+02 5.859e+02, threshold=6.095e+02, percent-clipped=1.0 2023-04-17 04:23:51,659 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64526.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:23:55,417 INFO [train.py:893] (1/4) Epoch 24, batch 2450, loss[loss=0.1591, simple_loss=0.2091, pruned_loss=0.05455, over 11782.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2319, pruned_loss=0.05433, over 2654769.93 frames. ], batch size: 48, lr: 5.39e-03, grad_scale: 32.0 2023-04-17 04:24:12,590 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64549.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:24:34,773 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64572.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:24:41,297 INFO [train.py:893] (1/4) Epoch 24, batch 2500, loss[loss=0.1701, simple_loss=0.2357, pruned_loss=0.05222, over 13530.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2321, pruned_loss=0.05428, over 2654568.15 frames. ], batch size: 91, lr: 5.39e-03, grad_scale: 32.0 2023-04-17 04:24:48,986 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64588.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:24:49,925 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64589.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:25:17,787 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3078, 3.6503, 3.5427, 4.0484, 2.3365, 3.0984, 3.8414, 2.3005], device='cuda:1'), covar=tensor([0.0152, 0.0443, 0.0735, 0.0418, 0.1560, 0.0957, 0.0496, 0.1621], device='cuda:1'), in_proj_covar=tensor([0.0133, 0.0185, 0.0206, 0.0249, 0.0183, 0.0198, 0.0179, 0.0175], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:25:18,424 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64620.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:25:21,447 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.754e+02 2.530e+02 2.853e+02 3.517e+02 5.646e+02, threshold=5.707e+02, percent-clipped=0.0 2023-04-17 04:25:27,347 INFO [train.py:893] (1/4) Epoch 24, batch 2550, loss[loss=0.1729, simple_loss=0.2344, pruned_loss=0.05568, over 13475.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2325, pruned_loss=0.05441, over 2651812.35 frames. ], batch size: 93, lr: 5.39e-03, grad_scale: 32.0 2023-04-17 04:25:43,911 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64648.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:25:46,414 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64650.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:25:51,171 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-17 04:26:12,637 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64679.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:26:13,197 INFO [train.py:893] (1/4) Epoch 24, batch 2600, loss[loss=0.1805, simple_loss=0.2414, pruned_loss=0.05975, over 13481.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2323, pruned_loss=0.05431, over 2652553.09 frames. ], batch size: 100, lr: 5.38e-03, grad_scale: 32.0 2023-04-17 04:26:20,811 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.9547, 4.7659, 4.9895, 4.8765, 5.2767, 4.7870, 5.2794, 5.2266], device='cuda:1'), covar=tensor([0.0364, 0.0575, 0.0651, 0.0523, 0.0522, 0.0786, 0.0434, 0.0492], device='cuda:1'), in_proj_covar=tensor([0.0299, 0.0302, 0.0312, 0.0232, 0.0449, 0.0352, 0.0286, 0.0308], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:26:33,504 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64701.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:26:35,094 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64703.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:26:37,672 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-04-17 04:26:39,768 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64709.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:26:50,188 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.964e+02 2.522e+02 3.008e+02 3.561e+02 6.412e+02, threshold=6.016e+02, percent-clipped=1.0 2023-04-17 04:26:53,414 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64727.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:26:55,605 INFO [train.py:893] (1/4) Epoch 24, batch 2650, loss[loss=0.1638, simple_loss=0.2279, pruned_loss=0.04984, over 13546.00 frames. ], tot_loss[loss=0.1711, simple_loss=0.2326, pruned_loss=0.05482, over 2655222.48 frames. ], batch size: 72, lr: 5.38e-03, grad_scale: 32.0 2023-04-17 04:27:07,242 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64745.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:27:12,574 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64752.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:27:22,185 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64764.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:27:53,303 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-17 04:28:00,592 INFO [train.py:893] (1/4) Epoch 25, batch 0, loss[loss=0.1701, simple_loss=0.2303, pruned_loss=0.05495, over 13498.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2303, pruned_loss=0.05495, over 13498.00 frames. ], batch size: 93, lr: 5.27e-03, grad_scale: 32.0 2023-04-17 04:28:00,593 INFO [train.py:918] (1/4) Computing validation loss 2023-04-17 04:28:09,054 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.7072, 4.2295, 4.2900, 4.4464, 4.2046, 4.1640, 4.7195, 4.2464], device='cuda:1'), covar=tensor([0.0790, 0.1330, 0.1846, 0.1872, 0.1094, 0.1496, 0.0949, 0.1272], device='cuda:1'), in_proj_covar=tensor([0.0347, 0.0394, 0.0488, 0.0488, 0.0308, 0.0362, 0.0454, 0.0361], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 04:28:14,121 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7536, 3.4884, 4.3883, 3.2387, 2.9807, 2.9871, 4.7165, 4.7724], device='cuda:1'), covar=tensor([0.1274, 0.2169, 0.0393, 0.1666, 0.1604, 0.1687, 0.0245, 0.0259], device='cuda:1'), in_proj_covar=tensor([0.0251, 0.0275, 0.0200, 0.0225, 0.0221, 0.0184, 0.0215, 0.0188], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 04:28:22,769 INFO [train.py:927] (1/4) Epoch 25, validation: loss=0.134, simple_loss=0.1935, pruned_loss=0.03718, over 2446609.00 frames. 2023-04-17 04:28:22,770 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12831MB 2023-04-17 04:28:46,859 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64806.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:28:54,062 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=64813.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:28:54,902 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8330, 3.0430, 3.2276, 4.3932, 3.9378, 4.4146, 3.5072, 2.9609], device='cuda:1'), covar=tensor([0.0282, 0.0764, 0.0731, 0.0064, 0.0203, 0.0062, 0.0606, 0.0818], device='cuda:1'), in_proj_covar=tensor([0.0120, 0.0147, 0.0164, 0.0100, 0.0120, 0.0099, 0.0168, 0.0148], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:29:00,511 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64821.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:29:02,045 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.683e+02 2.688e+02 3.039e+02 3.680e+02 5.833e+02, threshold=6.079e+02, percent-clipped=0.0 2023-04-17 04:29:07,894 INFO [train.py:893] (1/4) Epoch 25, batch 50, loss[loss=0.1669, simple_loss=0.2266, pruned_loss=0.05359, over 13459.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2268, pruned_loss=0.05446, over 601436.67 frames. ], batch size: 79, lr: 5.27e-03, grad_scale: 32.0 2023-04-17 04:29:26,412 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.68 vs. limit=5.0 2023-04-17 04:29:31,574 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-17 04:29:31,574 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-17 04:29:31,575 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-17 04:29:31,581 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-17 04:29:31,588 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-17 04:29:31,609 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-17 04:29:31,990 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 2023-04-17 04:29:32,378 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-17 04:29:45,902 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-04-17 04:29:52,471 INFO [train.py:893] (1/4) Epoch 25, batch 100, loss[loss=0.1877, simple_loss=0.2416, pruned_loss=0.0669, over 13544.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.231, pruned_loss=0.0563, over 1053171.77 frames. ], batch size: 87, lr: 5.27e-03, grad_scale: 32.0 2023-04-17 04:30:00,684 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=64888.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:30:32,600 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.977e+02 2.699e+02 3.087e+02 3.746e+02 6.680e+02, threshold=6.173e+02, percent-clipped=1.0 2023-04-17 04:30:37,711 INFO [train.py:893] (1/4) Epoch 25, batch 150, loss[loss=0.1539, simple_loss=0.2192, pruned_loss=0.04427, over 13398.00 frames. ], tot_loss[loss=0.1734, simple_loss=0.2324, pruned_loss=0.05724, over 1402844.47 frames. ], batch size: 113, lr: 5.26e-03, grad_scale: 32.0 2023-04-17 04:30:43,049 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4147, 3.7754, 3.6718, 4.1728, 2.3199, 3.2003, 3.9106, 2.2436], device='cuda:1'), covar=tensor([0.0142, 0.0478, 0.0711, 0.0593, 0.1654, 0.0950, 0.0490, 0.1820], device='cuda:1'), in_proj_covar=tensor([0.0136, 0.0189, 0.0209, 0.0253, 0.0185, 0.0201, 0.0183, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:30:43,781 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=64936.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:30:52,651 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=64945.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:31:21,082 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=64976.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:31:23,224 INFO [train.py:893] (1/4) Epoch 25, batch 200, loss[loss=0.1924, simple_loss=0.2539, pruned_loss=0.06548, over 13513.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2338, pruned_loss=0.05744, over 1671472.53 frames. ], batch size: 91, lr: 5.26e-03, grad_scale: 32.0 2023-04-17 04:31:43,553 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65001.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:31:45,952 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65004.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:32:02,059 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.794e+02 2.476e+02 2.997e+02 3.532e+02 5.574e+02, threshold=5.994e+02, percent-clipped=0.0 2023-04-17 04:32:06,834 INFO [train.py:893] (1/4) Epoch 25, batch 250, loss[loss=0.166, simple_loss=0.2254, pruned_loss=0.05326, over 13362.00 frames. ], tot_loss[loss=0.1754, simple_loss=0.235, pruned_loss=0.05794, over 1886288.88 frames. ], batch size: 73, lr: 5.26e-03, grad_scale: 32.0 2023-04-17 04:32:15,134 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65037.0, num_to_drop=1, layers_to_drop={2} 2023-04-17 04:32:26,095 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65049.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:32:31,184 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65055.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:32:34,319 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65059.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:32:52,652 INFO [train.py:893] (1/4) Epoch 25, batch 300, loss[loss=0.1913, simple_loss=0.2501, pruned_loss=0.06629, over 13231.00 frames. ], tot_loss[loss=0.1752, simple_loss=0.2351, pruned_loss=0.05765, over 2050636.72 frames. ], batch size: 132, lr: 5.26e-03, grad_scale: 32.0 2023-04-17 04:33:12,414 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65101.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:33:13,348 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7932, 2.6931, 2.3363, 1.8033, 1.7251, 2.2069, 2.3460, 2.9006], device='cuda:1'), covar=tensor([0.1015, 0.0394, 0.0729, 0.1534, 0.0260, 0.0638, 0.0749, 0.0276], device='cuda:1'), in_proj_covar=tensor([0.0173, 0.0147, 0.0125, 0.0210, 0.0115, 0.0167, 0.0178, 0.0137], device='cuda:1'), out_proj_covar=tensor([1.2819e-04, 1.1007e-04, 9.7623e-05, 1.5606e-04, 8.3373e-05, 1.2668e-04, 1.3414e-04, 1.0099e-04], device='cuda:1') 2023-04-17 04:33:18,021 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65108.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:33:25,594 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65116.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:33:29,569 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65121.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:33:30,966 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.916e+02 2.607e+02 3.049e+02 3.512e+02 7.816e+02, threshold=6.098e+02, percent-clipped=2.0 2023-04-17 04:33:37,305 INFO [train.py:893] (1/4) Epoch 25, batch 350, loss[loss=0.154, simple_loss=0.2179, pruned_loss=0.04503, over 13525.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2341, pruned_loss=0.05731, over 2182646.10 frames. ], batch size: 76, lr: 5.26e-03, grad_scale: 32.0 2023-04-17 04:34:13,294 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65169.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:34:20,722 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9935, 2.6871, 2.7465, 3.0510, 2.5006, 3.1878, 3.2231, 2.6120], device='cuda:1'), covar=tensor([0.0082, 0.0201, 0.0134, 0.0206, 0.0227, 0.0126, 0.0138, 0.0173], device='cuda:1'), in_proj_covar=tensor([0.0111, 0.0124, 0.0130, 0.0131, 0.0140, 0.0116, 0.0114, 0.0115], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 04:34:22,074 INFO [train.py:893] (1/4) Epoch 25, batch 400, loss[loss=0.2, simple_loss=0.264, pruned_loss=0.068, over 13428.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.2345, pruned_loss=0.05711, over 2284961.56 frames. ], batch size: 95, lr: 5.25e-03, grad_scale: 32.0 2023-04-17 04:34:39,116 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2693, 3.9966, 4.1562, 2.5644, 4.5747, 4.3004, 4.1984, 4.4908], device='cuda:1'), covar=tensor([0.0264, 0.0144, 0.0167, 0.1141, 0.0164, 0.0275, 0.0178, 0.0120], device='cuda:1'), in_proj_covar=tensor([0.0097, 0.0059, 0.0087, 0.0106, 0.0101, 0.0113, 0.0085, 0.0080], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:35:01,573 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.808e+02 2.550e+02 3.090e+02 3.761e+02 8.454e+02, threshold=6.180e+02, percent-clipped=2.0 2023-04-17 04:35:05,471 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-04-17 04:35:06,638 INFO [train.py:893] (1/4) Epoch 25, batch 450, loss[loss=0.1428, simple_loss=0.21, pruned_loss=0.03783, over 13351.00 frames. ], tot_loss[loss=0.1744, simple_loss=0.235, pruned_loss=0.0569, over 2369013.14 frames. ], batch size: 67, lr: 5.25e-03, grad_scale: 32.0 2023-04-17 04:35:12,671 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2705, 4.2092, 3.0389, 3.9372, 4.1955, 2.8207, 3.7950, 2.9655], device='cuda:1'), covar=tensor([0.0254, 0.0211, 0.0957, 0.0461, 0.0222, 0.1187, 0.0420, 0.1170], device='cuda:1'), in_proj_covar=tensor([0.0164, 0.0181, 0.0179, 0.0229, 0.0142, 0.0163, 0.0163, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:35:16,929 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.97 vs. limit=2.0 2023-04-17 04:35:21,477 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65245.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:35:32,536 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-17 04:35:50,781 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3823, 3.8447, 3.7558, 4.2055, 2.4185, 3.2383, 4.0326, 2.3818], device='cuda:1'), covar=tensor([0.0249, 0.0452, 0.0709, 0.0624, 0.1643, 0.0880, 0.0419, 0.1694], device='cuda:1'), in_proj_covar=tensor([0.0137, 0.0191, 0.0210, 0.0255, 0.0186, 0.0201, 0.0183, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:35:52,074 INFO [train.py:893] (1/4) Epoch 25, batch 500, loss[loss=0.1521, simple_loss=0.2099, pruned_loss=0.04716, over 13204.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2342, pruned_loss=0.05603, over 2434065.19 frames. ], batch size: 58, lr: 5.25e-03, grad_scale: 32.0 2023-04-17 04:35:58,596 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65285.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:36:05,053 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65293.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:36:10,733 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5643, 4.7783, 4.5077, 4.6019, 4.5960, 4.9664, 4.6700, 4.6239], device='cuda:1'), covar=tensor([0.0275, 0.0287, 0.0294, 0.0866, 0.0270, 0.0238, 0.0331, 0.0215], device='cuda:1'), in_proj_covar=tensor([0.0167, 0.0162, 0.0183, 0.0271, 0.0182, 0.0200, 0.0181, 0.0113], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 04:36:15,611 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65304.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:36:32,420 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.759e+02 2.424e+02 2.748e+02 3.322e+02 7.565e+02, threshold=5.496e+02, percent-clipped=1.0 2023-04-17 04:36:37,397 INFO [train.py:893] (1/4) Epoch 25, batch 550, loss[loss=0.1607, simple_loss=0.2322, pruned_loss=0.04464, over 13529.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2346, pruned_loss=0.05591, over 2482931.00 frames. ], batch size: 76, lr: 5.25e-03, grad_scale: 32.0 2023-04-17 04:36:38,557 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1778, 4.0113, 4.1915, 2.5789, 4.4731, 4.2236, 4.1479, 4.3917], device='cuda:1'), covar=tensor([0.0250, 0.0142, 0.0155, 0.1133, 0.0157, 0.0287, 0.0182, 0.0135], device='cuda:1'), in_proj_covar=tensor([0.0096, 0.0058, 0.0086, 0.0105, 0.0101, 0.0113, 0.0084, 0.0079], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:36:41,527 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65332.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 04:36:42,519 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7714, 4.0248, 3.0329, 2.7846, 2.8108, 2.4504, 4.0940, 2.3054], device='cuda:1'), covar=tensor([0.1983, 0.0371, 0.1328, 0.2326, 0.0977, 0.3522, 0.0278, 0.4314], device='cuda:1'), in_proj_covar=tensor([0.0316, 0.0293, 0.0323, 0.0344, 0.0264, 0.0334, 0.0219, 0.0371], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 04:36:53,797 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65346.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:36:59,110 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65352.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:37:04,871 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65359.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:37:22,605 INFO [train.py:893] (1/4) Epoch 25, batch 600, loss[loss=0.169, simple_loss=0.2323, pruned_loss=0.05288, over 13539.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.234, pruned_loss=0.05586, over 2518222.44 frames. ], batch size: 78, lr: 5.25e-03, grad_scale: 32.0 2023-04-17 04:37:42,701 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65401.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:37:47,776 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65407.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:37:48,689 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65408.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:37:51,920 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65411.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:38:02,390 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.737e+02 2.398e+02 2.864e+02 3.237e+02 5.045e+02, threshold=5.727e+02, percent-clipped=0.0 2023-04-17 04:38:08,304 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7105, 3.8263, 2.8439, 2.5730, 2.7007, 2.3029, 3.8678, 2.1812], device='cuda:1'), covar=tensor([0.1950, 0.0409, 0.1461, 0.2525, 0.0980, 0.3702, 0.0339, 0.4242], device='cuda:1'), in_proj_covar=tensor([0.0318, 0.0295, 0.0325, 0.0346, 0.0266, 0.0337, 0.0221, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 04:38:08,683 INFO [train.py:893] (1/4) Epoch 25, batch 650, loss[loss=0.1778, simple_loss=0.2374, pruned_loss=0.05908, over 13234.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2335, pruned_loss=0.05561, over 2549011.94 frames. ], batch size: 124, lr: 5.24e-03, grad_scale: 32.0 2023-04-17 04:38:13,043 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5484, 4.9300, 4.7112, 4.7217, 4.7146, 4.5356, 4.9708, 4.9855], device='cuda:1'), covar=tensor([0.0201, 0.0202, 0.0204, 0.0314, 0.0286, 0.0234, 0.0259, 0.0195], device='cuda:1'), in_proj_covar=tensor([0.0144, 0.0208, 0.0170, 0.0188, 0.0159, 0.0208, 0.0138, 0.0188], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:38:20,667 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9087, 4.1173, 3.1691, 2.8920, 3.0038, 2.5633, 4.2056, 2.4713], device='cuda:1'), covar=tensor([0.1795, 0.0346, 0.1253, 0.2189, 0.0879, 0.3432, 0.0278, 0.4089], device='cuda:1'), in_proj_covar=tensor([0.0318, 0.0295, 0.0325, 0.0346, 0.0265, 0.0337, 0.0221, 0.0374], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 04:38:26,805 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65449.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:38:32,742 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65456.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:38:54,396 INFO [train.py:893] (1/4) Epoch 25, batch 700, loss[loss=0.1774, simple_loss=0.233, pruned_loss=0.06086, over 13577.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2333, pruned_loss=0.0555, over 2572479.48 frames. ], batch size: 89, lr: 5.24e-03, grad_scale: 32.0 2023-04-17 04:39:33,904 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.923e+02 2.683e+02 3.062e+02 3.577e+02 6.124e+02, threshold=6.124e+02, percent-clipped=1.0 2023-04-17 04:39:38,797 INFO [train.py:893] (1/4) Epoch 25, batch 750, loss[loss=0.1734, simple_loss=0.2399, pruned_loss=0.0535, over 13501.00 frames. ], tot_loss[loss=0.1727, simple_loss=0.2336, pruned_loss=0.0559, over 2593240.70 frames. ], batch size: 91, lr: 5.24e-03, grad_scale: 32.0 2023-04-17 04:39:41,544 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65532.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 04:40:09,041 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5512, 3.4137, 4.0843, 2.8890, 2.6161, 2.7477, 4.3646, 4.4712], device='cuda:1'), covar=tensor([0.1184, 0.1769, 0.0373, 0.1794, 0.1706, 0.1625, 0.0289, 0.0227], device='cuda:1'), in_proj_covar=tensor([0.0251, 0.0277, 0.0202, 0.0228, 0.0222, 0.0185, 0.0216, 0.0190], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 04:40:17,915 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.0773, 4.9286, 4.9078, 4.8185, 5.4119, 4.6467, 5.2819, 5.3427], device='cuda:1'), covar=tensor([0.0792, 0.1153, 0.1490, 0.1062, 0.1133, 0.2151, 0.0904, 0.0958], device='cuda:1'), in_proj_covar=tensor([0.0298, 0.0300, 0.0307, 0.0229, 0.0443, 0.0350, 0.0285, 0.0303], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:40:23,421 INFO [train.py:893] (1/4) Epoch 25, batch 800, loss[loss=0.1663, simple_loss=0.2363, pruned_loss=0.04811, over 13523.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2346, pruned_loss=0.05618, over 2609891.07 frames. ], batch size: 83, lr: 5.24e-03, grad_scale: 32.0 2023-04-17 04:40:36,133 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65593.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 04:40:37,870 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=2.63 vs. limit=5.0 2023-04-17 04:41:03,598 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.878e+02 2.542e+02 2.840e+02 3.512e+02 5.061e+02, threshold=5.680e+02, percent-clipped=0.0 2023-04-17 04:41:09,865 INFO [train.py:893] (1/4) Epoch 25, batch 850, loss[loss=0.1668, simple_loss=0.2297, pruned_loss=0.05194, over 13210.00 frames. ], tot_loss[loss=0.1743, simple_loss=0.2356, pruned_loss=0.0565, over 2621847.86 frames. ], batch size: 132, lr: 5.24e-03, grad_scale: 32.0 2023-04-17 04:41:12,671 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65632.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:41:20,071 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65641.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:41:27,313 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9381, 3.8830, 3.9323, 2.5960, 4.3643, 4.0770, 4.1111, 4.3585], device='cuda:1'), covar=tensor([0.0358, 0.0221, 0.0227, 0.1341, 0.0231, 0.0362, 0.0193, 0.0182], device='cuda:1'), in_proj_covar=tensor([0.0098, 0.0059, 0.0088, 0.0107, 0.0103, 0.0115, 0.0085, 0.0080], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:41:55,389 INFO [train.py:893] (1/4) Epoch 25, batch 900, loss[loss=0.1627, simple_loss=0.2293, pruned_loss=0.048, over 13375.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2352, pruned_loss=0.05662, over 2629469.12 frames. ], batch size: 109, lr: 5.23e-03, grad_scale: 32.0 2023-04-17 04:41:56,378 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65680.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:42:08,662 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.1204, 1.8094, 3.7952, 3.7345, 3.6961, 2.9759, 3.4264, 2.9231], device='cuda:1'), covar=tensor([0.2166, 0.1743, 0.0179, 0.0177, 0.0226, 0.0744, 0.0290, 0.0835], device='cuda:1'), in_proj_covar=tensor([0.0218, 0.0186, 0.0129, 0.0135, 0.0137, 0.0177, 0.0148, 0.0186], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-17 04:42:09,466 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3245, 4.6362, 4.2899, 4.3992, 4.4102, 4.7500, 4.5283, 4.4754], device='cuda:1'), covar=tensor([0.0307, 0.0232, 0.0309, 0.0798, 0.0262, 0.0242, 0.0290, 0.0254], device='cuda:1'), in_proj_covar=tensor([0.0169, 0.0163, 0.0184, 0.0272, 0.0183, 0.0201, 0.0181, 0.0114], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 04:42:20,636 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6648, 4.1795, 4.1046, 4.2300, 4.0627, 4.0267, 4.6114, 4.1404], device='cuda:1'), covar=tensor([0.0784, 0.1289, 0.2266, 0.2548, 0.1100, 0.1698, 0.0981, 0.1575], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0400, 0.0495, 0.0497, 0.0312, 0.0367, 0.0459, 0.0365], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 04:42:24,665 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65711.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:42:26,052 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-17 04:42:35,419 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.039e+02 2.809e+02 3.107e+02 3.712e+02 7.214e+02, threshold=6.214e+02, percent-clipped=4.0 2023-04-17 04:42:40,563 INFO [train.py:893] (1/4) Epoch 25, batch 950, loss[loss=0.1686, simple_loss=0.2333, pruned_loss=0.05195, over 13541.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2346, pruned_loss=0.05689, over 2635035.96 frames. ], batch size: 98, lr: 5.23e-03, grad_scale: 32.0 2023-04-17 04:43:08,409 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65759.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:43:26,045 INFO [train.py:893] (1/4) Epoch 25, batch 1000, loss[loss=0.1787, simple_loss=0.2273, pruned_loss=0.06504, over 12147.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2328, pruned_loss=0.05641, over 2638626.71 frames. ], batch size: 49, lr: 5.23e-03, grad_scale: 32.0 2023-04-17 04:44:03,856 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.2460, 2.9201, 3.6627, 2.6787, 2.5015, 2.5711, 3.9772, 4.0742], device='cuda:1'), covar=tensor([0.1234, 0.2051, 0.0365, 0.1753, 0.1675, 0.1602, 0.0304, 0.0206], device='cuda:1'), in_proj_covar=tensor([0.0250, 0.0276, 0.0201, 0.0226, 0.0221, 0.0184, 0.0216, 0.0189], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 04:44:06,781 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.851e+02 2.700e+02 3.165e+02 3.648e+02 7.276e+02, threshold=6.330e+02, percent-clipped=2.0 2023-04-17 04:44:11,778 INFO [train.py:893] (1/4) Epoch 25, batch 1050, loss[loss=0.1481, simple_loss=0.2109, pruned_loss=0.04263, over 13464.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2306, pruned_loss=0.05489, over 2646119.01 frames. ], batch size: 79, lr: 5.23e-03, grad_scale: 32.0 2023-04-17 04:44:20,873 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=65838.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 04:44:25,933 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-04-17 04:44:53,846 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0219, 4.2806, 2.8684, 3.6371, 4.1074, 2.7083, 3.6444, 2.8196], device='cuda:1'), covar=tensor([0.0305, 0.0283, 0.1137, 0.0523, 0.0273, 0.1294, 0.0516, 0.1286], device='cuda:1'), in_proj_covar=tensor([0.0166, 0.0184, 0.0181, 0.0232, 0.0142, 0.0165, 0.0166, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:44:54,723 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-04-17 04:44:57,411 INFO [train.py:893] (1/4) Epoch 25, batch 1100, loss[loss=0.1622, simple_loss=0.2266, pruned_loss=0.04888, over 13570.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2314, pruned_loss=0.05448, over 2647868.24 frames. ], batch size: 89, lr: 5.23e-03, grad_scale: 32.0 2023-04-17 04:45:05,535 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=65888.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 04:45:15,533 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=65899.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 04:45:37,545 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.961e+02 2.539e+02 2.910e+02 3.534e+02 6.326e+02, threshold=5.820e+02, percent-clipped=0.0 2023-04-17 04:45:43,068 INFO [train.py:893] (1/4) Epoch 25, batch 1150, loss[loss=0.179, simple_loss=0.2301, pruned_loss=0.06397, over 13366.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2309, pruned_loss=0.05363, over 2650040.95 frames. ], batch size: 62, lr: 5.22e-03, grad_scale: 32.0 2023-04-17 04:45:54,899 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=65941.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:46:25,158 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.61 vs. limit=5.0 2023-04-17 04:46:28,757 INFO [train.py:893] (1/4) Epoch 25, batch 1200, loss[loss=0.1612, simple_loss=0.213, pruned_loss=0.05469, over 12799.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2302, pruned_loss=0.05329, over 2653029.29 frames. ], batch size: 52, lr: 5.22e-03, grad_scale: 32.0 2023-04-17 04:46:37,316 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=65989.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:46:57,339 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1310, 4.1997, 2.9349, 3.8695, 4.1352, 2.6850, 3.7360, 2.7629], device='cuda:1'), covar=tensor([0.0267, 0.0212, 0.1010, 0.0378, 0.0230, 0.1209, 0.0493, 0.1189], device='cuda:1'), in_proj_covar=tensor([0.0165, 0.0183, 0.0179, 0.0230, 0.0141, 0.0164, 0.0165, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:46:58,599 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-17 04:47:00,804 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.45 vs. limit=5.0 2023-04-17 04:47:09,387 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-17 04:47:11,541 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.650e+02 2.513e+02 2.947e+02 3.399e+02 7.667e+02, threshold=5.893e+02, percent-clipped=1.0 2023-04-17 04:47:17,274 INFO [train.py:893] (1/4) Epoch 25, batch 1250, loss[loss=0.1872, simple_loss=0.2401, pruned_loss=0.06716, over 13530.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2318, pruned_loss=0.05449, over 2654647.30 frames. ], batch size: 76, lr: 5.22e-03, grad_scale: 32.0 2023-04-17 04:47:35,479 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9479, 3.8032, 2.9208, 3.6034, 3.8644, 2.5567, 3.6268, 2.5408], device='cuda:1'), covar=tensor([0.0242, 0.0185, 0.0892, 0.0279, 0.0217, 0.1123, 0.0396, 0.1211], device='cuda:1'), in_proj_covar=tensor([0.0166, 0.0183, 0.0180, 0.0230, 0.0141, 0.0164, 0.0165, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:48:02,016 INFO [train.py:893] (1/4) Epoch 25, batch 1300, loss[loss=0.1815, simple_loss=0.2429, pruned_loss=0.06009, over 13313.00 frames. ], tot_loss[loss=0.1736, simple_loss=0.2346, pruned_loss=0.0563, over 2651283.17 frames. ], batch size: 118, lr: 5.22e-03, grad_scale: 32.0 2023-04-17 04:48:41,327 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.956e+02 2.611e+02 3.159e+02 3.839e+02 8.053e+02, threshold=6.319e+02, percent-clipped=2.0 2023-04-17 04:48:47,607 INFO [train.py:893] (1/4) Epoch 25, batch 1350, loss[loss=0.1869, simple_loss=0.2466, pruned_loss=0.06362, over 13538.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2343, pruned_loss=0.05602, over 2658585.84 frames. ], batch size: 98, lr: 5.22e-03, grad_scale: 64.0 2023-04-17 04:49:00,343 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66144.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:49:02,881 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6086, 3.9892, 3.7700, 4.3852, 2.4773, 3.4112, 4.1620, 2.3935], device='cuda:1'), covar=tensor([0.0163, 0.0394, 0.0749, 0.0448, 0.1610, 0.0857, 0.0417, 0.1695], device='cuda:1'), in_proj_covar=tensor([0.0138, 0.0192, 0.0210, 0.0254, 0.0187, 0.0203, 0.0183, 0.0180], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:49:16,151 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66160.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 04:49:31,716 INFO [train.py:893] (1/4) Epoch 25, batch 1400, loss[loss=0.1809, simple_loss=0.2244, pruned_loss=0.06875, over 9386.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2337, pruned_loss=0.05547, over 2657850.58 frames. ], batch size: 38, lr: 5.21e-03, grad_scale: 64.0 2023-04-17 04:49:40,654 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66188.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 04:49:40,794 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3243, 2.9719, 3.5772, 2.6733, 2.3806, 2.4654, 3.9977, 4.0701], device='cuda:1'), covar=tensor([0.1264, 0.2001, 0.0412, 0.1874, 0.1828, 0.1816, 0.0328, 0.0187], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0277, 0.0202, 0.0227, 0.0222, 0.0185, 0.0218, 0.0190], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 04:49:46,781 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66194.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 04:49:56,200 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66205.0, num_to_drop=1, layers_to_drop={2} 2023-04-17 04:50:11,207 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66221.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 04:50:13,202 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.900e+02 2.492e+02 2.974e+02 3.412e+02 5.608e+02, threshold=5.949e+02, percent-clipped=0.0 2023-04-17 04:50:17,977 INFO [train.py:893] (1/4) Epoch 25, batch 1450, loss[loss=0.1755, simple_loss=0.2375, pruned_loss=0.05675, over 13539.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2348, pruned_loss=0.05631, over 2663277.62 frames. ], batch size: 85, lr: 5.21e-03, grad_scale: 64.0 2023-04-17 04:50:23,929 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66236.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 04:50:26,531 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0992, 2.7564, 3.4066, 2.6809, 2.4663, 2.4674, 3.7222, 3.7534], device='cuda:1'), covar=tensor([0.1234, 0.1969, 0.0371, 0.1557, 0.1565, 0.1565, 0.0269, 0.0191], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0277, 0.0202, 0.0228, 0.0222, 0.0185, 0.0219, 0.0190], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 04:50:44,632 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66258.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:51:02,132 INFO [train.py:893] (1/4) Epoch 25, batch 1500, loss[loss=0.1977, simple_loss=0.2609, pruned_loss=0.06722, over 13386.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2345, pruned_loss=0.05592, over 2664853.99 frames. ], batch size: 118, lr: 5.21e-03, grad_scale: 64.0 2023-04-17 04:51:23,392 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6980, 3.8022, 2.8934, 2.6048, 2.7556, 2.3627, 3.8568, 2.2096], device='cuda:1'), covar=tensor([0.1825, 0.0451, 0.1408, 0.2319, 0.0973, 0.3623, 0.0316, 0.4561], device='cuda:1'), in_proj_covar=tensor([0.0318, 0.0297, 0.0327, 0.0347, 0.0267, 0.0338, 0.0222, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 04:51:40,320 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66319.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:51:43,267 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.769e+02 2.622e+02 3.018e+02 3.858e+02 6.130e+02, threshold=6.036e+02, percent-clipped=2.0 2023-04-17 04:51:48,104 INFO [train.py:893] (1/4) Epoch 25, batch 1550, loss[loss=0.176, simple_loss=0.2366, pruned_loss=0.0577, over 13372.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2346, pruned_loss=0.05599, over 2657437.06 frames. ], batch size: 113, lr: 5.21e-03, grad_scale: 64.0 2023-04-17 04:51:49,200 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0360, 4.4933, 4.2349, 4.2649, 4.3055, 4.0804, 4.5414, 4.5439], device='cuda:1'), covar=tensor([0.0221, 0.0197, 0.0208, 0.0313, 0.0259, 0.0261, 0.0258, 0.0241], device='cuda:1'), in_proj_covar=tensor([0.0143, 0.0207, 0.0171, 0.0186, 0.0157, 0.0207, 0.0137, 0.0186], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:52:33,396 INFO [train.py:893] (1/4) Epoch 25, batch 1600, loss[loss=0.1785, simple_loss=0.2527, pruned_loss=0.05211, over 13358.00 frames. ], tot_loss[loss=0.1737, simple_loss=0.2351, pruned_loss=0.05613, over 2657165.01 frames. ], batch size: 118, lr: 5.21e-03, grad_scale: 64.0 2023-04-17 04:52:40,340 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-04-17 04:52:47,714 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7772, 3.6946, 2.5333, 3.2791, 2.8782, 1.9618, 3.7182, 2.0198], device='cuda:1'), covar=tensor([0.0627, 0.0442, 0.0655, 0.0341, 0.0607, 0.1897, 0.0770, 0.1227], device='cuda:1'), in_proj_covar=tensor([0.0145, 0.0147, 0.0137, 0.0121, 0.0150, 0.0193, 0.0184, 0.0147], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 04:53:13,883 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.786e+02 2.679e+02 3.209e+02 3.685e+02 7.595e+02, threshold=6.418e+02, percent-clipped=1.0 2023-04-17 04:53:18,068 INFO [train.py:893] (1/4) Epoch 25, batch 1650, loss[loss=0.1494, simple_loss=0.2204, pruned_loss=0.03919, over 13041.00 frames. ], tot_loss[loss=0.1729, simple_loss=0.2353, pruned_loss=0.05532, over 2656273.47 frames. ], batch size: 142, lr: 5.20e-03, grad_scale: 32.0 2023-04-17 04:54:04,122 INFO [train.py:893] (1/4) Epoch 25, batch 1700, loss[loss=0.1677, simple_loss=0.2273, pruned_loss=0.05402, over 13517.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.235, pruned_loss=0.05474, over 2659856.71 frames. ], batch size: 70, lr: 5.20e-03, grad_scale: 32.0 2023-04-17 04:54:17,280 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66494.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 04:54:22,926 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66500.0, num_to_drop=1, layers_to_drop={2} 2023-04-17 04:54:37,338 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66516.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 04:54:41,428 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66521.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 04:54:44,279 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.637e+02 2.534e+02 2.889e+02 3.612e+02 5.828e+02, threshold=5.778e+02, percent-clipped=0.0 2023-04-17 04:54:48,321 INFO [train.py:893] (1/4) Epoch 25, batch 1750, loss[loss=0.1856, simple_loss=0.2458, pruned_loss=0.06269, over 13224.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2338, pruned_loss=0.05404, over 2661209.95 frames. ], batch size: 132, lr: 5.20e-03, grad_scale: 32.0 2023-04-17 04:54:54,179 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1146, 2.4210, 1.8738, 4.0054, 4.4025, 3.3145, 4.3347, 4.1324], device='cuda:1'), covar=tensor([0.0090, 0.0977, 0.1137, 0.0090, 0.0072, 0.0469, 0.0072, 0.0081], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0094, 0.0102, 0.0085, 0.0071, 0.0084, 0.0059, 0.0081], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 04:55:01,242 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66542.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 04:55:06,155 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8828, 4.2462, 4.0630, 4.0265, 4.1160, 3.9292, 4.3139, 4.3096], device='cuda:1'), covar=tensor([0.0235, 0.0263, 0.0227, 0.0385, 0.0285, 0.0289, 0.0274, 0.0300], device='cuda:1'), in_proj_covar=tensor([0.0144, 0.0209, 0.0172, 0.0188, 0.0158, 0.0208, 0.0139, 0.0188], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:55:09,344 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0018, 4.4792, 4.2395, 4.2108, 4.2795, 4.0756, 4.5346, 4.5270], device='cuda:1'), covar=tensor([0.0254, 0.0271, 0.0250, 0.0445, 0.0375, 0.0328, 0.0288, 0.0305], device='cuda:1'), in_proj_covar=tensor([0.0144, 0.0209, 0.0172, 0.0188, 0.0159, 0.0208, 0.0139, 0.0188], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:55:33,742 INFO [train.py:893] (1/4) Epoch 25, batch 1800, loss[loss=0.1541, simple_loss=0.2176, pruned_loss=0.04534, over 13551.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2335, pruned_loss=0.05387, over 2662020.40 frames. ], batch size: 78, lr: 5.20e-03, grad_scale: 32.0 2023-04-17 04:55:35,746 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4798, 2.4932, 2.7138, 4.0451, 3.6206, 4.1082, 3.2730, 2.5462], device='cuda:1'), covar=tensor([0.0291, 0.0872, 0.0854, 0.0069, 0.0225, 0.0063, 0.0635, 0.0906], device='cuda:1'), in_proj_covar=tensor([0.0123, 0.0150, 0.0166, 0.0102, 0.0120, 0.0099, 0.0170, 0.0149], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 04:55:36,563 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66582.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 04:56:06,334 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66614.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:56:15,765 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.962e+02 2.657e+02 3.005e+02 3.637e+02 5.123e+02, threshold=6.010e+02, percent-clipped=0.0 2023-04-17 04:56:21,313 INFO [train.py:893] (1/4) Epoch 25, batch 1850, loss[loss=0.1811, simple_loss=0.2423, pruned_loss=0.05998, over 11805.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2333, pruned_loss=0.05376, over 2663143.23 frames. ], batch size: 157, lr: 5.20e-03, grad_scale: 32.0 2023-04-17 04:56:23,015 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-17 04:57:05,931 INFO [train.py:893] (1/4) Epoch 25, batch 1900, loss[loss=0.1503, simple_loss=0.2153, pruned_loss=0.04261, over 13455.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2326, pruned_loss=0.05402, over 2662035.57 frames. ], batch size: 79, lr: 5.19e-03, grad_scale: 32.0 2023-04-17 04:57:47,646 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.644e+02 2.632e+02 3.055e+02 3.564e+02 5.579e+02, threshold=6.111e+02, percent-clipped=0.0 2023-04-17 04:57:51,037 INFO [train.py:893] (1/4) Epoch 25, batch 1950, loss[loss=0.1753, simple_loss=0.2406, pruned_loss=0.05499, over 13190.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2325, pruned_loss=0.05401, over 2664574.39 frames. ], batch size: 132, lr: 5.19e-03, grad_scale: 16.0 2023-04-17 04:57:52,948 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66731.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 04:58:34,710 INFO [train.py:893] (1/4) Epoch 25, batch 2000, loss[loss=0.206, simple_loss=0.2609, pruned_loss=0.07552, over 13243.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2348, pruned_loss=0.05535, over 2663089.52 frames. ], batch size: 124, lr: 5.19e-03, grad_scale: 16.0 2023-04-17 04:58:39,699 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-17 04:58:47,227 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=66792.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 04:58:54,027 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66800.0, num_to_drop=1, layers_to_drop={2} 2023-04-17 04:59:09,363 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66816.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 04:59:17,059 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.867e+02 2.596e+02 3.110e+02 3.946e+02 7.233e+02, threshold=6.221e+02, percent-clipped=2.0 2023-04-17 04:59:20,397 INFO [train.py:893] (1/4) Epoch 25, batch 2050, loss[loss=0.1342, simple_loss=0.1939, pruned_loss=0.03727, over 12721.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2355, pruned_loss=0.05581, over 2657132.01 frames. ], batch size: 52, lr: 5.19e-03, grad_scale: 16.0 2023-04-17 04:59:37,915 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66848.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 04:59:52,366 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66864.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 05:00:03,541 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=66877.0, num_to_drop=1, layers_to_drop={2} 2023-04-17 05:00:04,895 INFO [train.py:893] (1/4) Epoch 25, batch 2100, loss[loss=0.1623, simple_loss=0.2264, pruned_loss=0.04911, over 13527.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2355, pruned_loss=0.05554, over 2656619.91 frames. ], batch size: 72, lr: 5.19e-03, grad_scale: 16.0 2023-04-17 05:00:12,506 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5380, 2.4303, 2.9627, 4.0969, 3.6540, 4.1461, 3.3177, 2.6189], device='cuda:1'), covar=tensor([0.0269, 0.0984, 0.0761, 0.0067, 0.0212, 0.0061, 0.0651, 0.0915], device='cuda:1'), in_proj_covar=tensor([0.0124, 0.0151, 0.0167, 0.0103, 0.0121, 0.0100, 0.0171, 0.0150], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:00:20,671 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0239, 3.7082, 3.1338, 3.4634, 3.0088, 2.4207, 3.8336, 2.3651], device='cuda:1'), covar=tensor([0.0618, 0.0603, 0.0424, 0.0341, 0.0692, 0.1618, 0.0754, 0.1114], device='cuda:1'), in_proj_covar=tensor([0.0146, 0.0148, 0.0136, 0.0120, 0.0152, 0.0192, 0.0184, 0.0146], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 05:00:37,518 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=66914.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:00:46,284 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.756e+02 2.554e+02 2.938e+02 3.443e+02 6.083e+02, threshold=5.876e+02, percent-clipped=0.0 2023-04-17 05:00:50,275 INFO [train.py:893] (1/4) Epoch 25, batch 2150, loss[loss=0.1604, simple_loss=0.2175, pruned_loss=0.05169, over 13408.00 frames. ], tot_loss[loss=0.1728, simple_loss=0.2353, pruned_loss=0.05519, over 2656656.72 frames. ], batch size: 62, lr: 5.18e-03, grad_scale: 16.0 2023-04-17 05:01:16,009 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=66957.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:01:20,011 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=66962.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:01:35,979 INFO [train.py:893] (1/4) Epoch 25, batch 2200, loss[loss=0.1516, simple_loss=0.201, pruned_loss=0.0511, over 12768.00 frames. ], tot_loss[loss=0.1715, simple_loss=0.2342, pruned_loss=0.05447, over 2653828.81 frames. ], batch size: 52, lr: 5.18e-03, grad_scale: 16.0 2023-04-17 05:02:02,807 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.1530, 4.9793, 5.2450, 5.0459, 5.4956, 4.9904, 5.4888, 5.4515], device='cuda:1'), covar=tensor([0.0396, 0.0572, 0.0581, 0.0503, 0.0515, 0.0795, 0.0436, 0.0470], device='cuda:1'), in_proj_covar=tensor([0.0303, 0.0309, 0.0316, 0.0234, 0.0454, 0.0357, 0.0292, 0.0312], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:02:11,524 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67018.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 05:02:16,886 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.719e+02 2.330e+02 2.750e+02 3.208e+02 5.338e+02, threshold=5.499e+02, percent-clipped=0.0 2023-04-17 05:02:20,289 INFO [train.py:893] (1/4) Epoch 25, batch 2250, loss[loss=0.1811, simple_loss=0.2446, pruned_loss=0.0588, over 13437.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2334, pruned_loss=0.05409, over 2658104.85 frames. ], batch size: 95, lr: 5.18e-03, grad_scale: 16.0 2023-04-17 05:02:40,784 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-04-17 05:02:45,355 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7766, 3.4686, 2.8175, 3.0814, 2.9137, 2.2215, 3.5848, 2.2415], device='cuda:1'), covar=tensor([0.0728, 0.0609, 0.0543, 0.0493, 0.0722, 0.1970, 0.0895, 0.1167], device='cuda:1'), in_proj_covar=tensor([0.0148, 0.0148, 0.0138, 0.0121, 0.0153, 0.0194, 0.0185, 0.0147], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 05:03:05,912 INFO [train.py:893] (1/4) Epoch 25, batch 2300, loss[loss=0.1557, simple_loss=0.2156, pruned_loss=0.04792, over 13164.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2328, pruned_loss=0.05354, over 2657874.56 frames. ], batch size: 58, lr: 5.18e-03, grad_scale: 16.0 2023-04-17 05:03:10,364 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67083.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:03:13,590 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67087.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 05:03:24,796 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2247, 3.5578, 3.4149, 3.9806, 2.3030, 3.0118, 3.6876, 2.1748], device='cuda:1'), covar=tensor([0.0194, 0.0484, 0.0823, 0.0530, 0.1604, 0.1010, 0.0578, 0.1784], device='cuda:1'), in_proj_covar=tensor([0.0138, 0.0191, 0.0209, 0.0254, 0.0186, 0.0202, 0.0183, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:03:47,496 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.495e+02 2.297e+02 2.747e+02 3.196e+02 6.009e+02, threshold=5.494e+02, percent-clipped=2.0 2023-04-17 05:03:52,279 INFO [train.py:893] (1/4) Epoch 25, batch 2350, loss[loss=0.166, simple_loss=0.2298, pruned_loss=0.05106, over 13531.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2324, pruned_loss=0.05354, over 2660301.56 frames. ], batch size: 98, lr: 5.18e-03, grad_scale: 16.0 2023-04-17 05:04:05,409 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67144.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:04:10,770 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-17 05:04:34,590 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67177.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 05:04:35,980 INFO [train.py:893] (1/4) Epoch 25, batch 2400, loss[loss=0.1515, simple_loss=0.2135, pruned_loss=0.04471, over 13366.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2311, pruned_loss=0.05304, over 2664970.56 frames. ], batch size: 73, lr: 5.17e-03, grad_scale: 16.0 2023-04-17 05:04:37,958 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.4831, 2.2879, 2.1846, 2.8146, 1.8954, 2.6849, 2.5229, 2.0828], device='cuda:1'), covar=tensor([0.0243, 0.0353, 0.0276, 0.0174, 0.0381, 0.0204, 0.0359, 0.0340], device='cuda:1'), in_proj_covar=tensor([0.0113, 0.0124, 0.0130, 0.0130, 0.0140, 0.0117, 0.0115, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 05:05:18,057 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.660e+02 2.350e+02 2.674e+02 3.216e+02 6.257e+02, threshold=5.348e+02, percent-clipped=1.0 2023-04-17 05:05:18,988 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67225.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 05:05:22,087 INFO [train.py:893] (1/4) Epoch 25, batch 2450, loss[loss=0.1634, simple_loss=0.2205, pruned_loss=0.05318, over 13536.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2299, pruned_loss=0.0524, over 2664855.78 frames. ], batch size: 83, lr: 5.17e-03, grad_scale: 16.0 2023-04-17 05:05:51,252 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.3993, 1.9895, 3.9325, 3.7186, 3.8074, 3.1565, 3.5649, 3.0935], device='cuda:1'), covar=tensor([0.2033, 0.1658, 0.0159, 0.0344, 0.0229, 0.0686, 0.0286, 0.0800], device='cuda:1'), in_proj_covar=tensor([0.0220, 0.0184, 0.0130, 0.0136, 0.0137, 0.0177, 0.0149, 0.0187], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-17 05:06:06,344 INFO [train.py:893] (1/4) Epoch 25, batch 2500, loss[loss=0.1811, simple_loss=0.247, pruned_loss=0.05761, over 13431.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2301, pruned_loss=0.05244, over 2665002.54 frames. ], batch size: 95, lr: 5.17e-03, grad_scale: 16.0 2023-04-17 05:06:15,512 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1480, 2.5296, 2.0490, 4.0412, 4.5176, 3.4335, 4.4347, 4.2805], device='cuda:1'), covar=tensor([0.0111, 0.0959, 0.1096, 0.0116, 0.0065, 0.0480, 0.0082, 0.0101], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0093, 0.0101, 0.0085, 0.0071, 0.0083, 0.0059, 0.0081], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 05:06:37,117 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67313.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 05:06:48,953 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.592e+02 2.493e+02 2.922e+02 3.566e+02 6.370e+02, threshold=5.844e+02, percent-clipped=1.0 2023-04-17 05:06:52,188 INFO [train.py:893] (1/4) Epoch 25, batch 2550, loss[loss=0.1565, simple_loss=0.223, pruned_loss=0.04499, over 13537.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.231, pruned_loss=0.05286, over 2669020.48 frames. ], batch size: 83, lr: 5.17e-03, grad_scale: 16.0 2023-04-17 05:07:13,762 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-17 05:07:25,806 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9306, 2.8768, 2.5752, 1.8167, 1.8955, 2.4133, 2.4883, 3.0970], device='cuda:1'), covar=tensor([0.0939, 0.0357, 0.0620, 0.1501, 0.0376, 0.0613, 0.0751, 0.0205], device='cuda:1'), in_proj_covar=tensor([0.0174, 0.0151, 0.0125, 0.0212, 0.0117, 0.0172, 0.0181, 0.0139], device='cuda:1'), out_proj_covar=tensor([1.2868e-04, 1.1306e-04, 9.7588e-05, 1.5699e-04, 8.4844e-05, 1.2993e-04, 1.3614e-04, 1.0200e-04], device='cuda:1') 2023-04-17 05:07:37,513 INFO [train.py:893] (1/4) Epoch 25, batch 2600, loss[loss=0.1486, simple_loss=0.2104, pruned_loss=0.0434, over 13569.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2312, pruned_loss=0.05319, over 2667006.34 frames. ], batch size: 72, lr: 5.17e-03, grad_scale: 16.0 2023-04-17 05:07:45,731 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67387.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 05:08:15,380 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.933e+02 2.504e+02 2.907e+02 3.489e+02 6.172e+02, threshold=5.813e+02, percent-clipped=2.0 2023-04-17 05:08:18,484 INFO [train.py:893] (1/4) Epoch 25, batch 2650, loss[loss=0.1618, simple_loss=0.227, pruned_loss=0.0483, over 13535.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2326, pruned_loss=0.05438, over 2665668.43 frames. ], batch size: 76, lr: 5.17e-03, grad_scale: 16.0 2023-04-17 05:08:23,106 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67435.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 05:08:26,009 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67439.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:08:28,588 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-04-17 05:08:35,936 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2723, 3.6173, 3.4690, 4.0152, 2.2364, 3.1097, 3.8213, 2.2829], device='cuda:1'), covar=tensor([0.0171, 0.0489, 0.0811, 0.0528, 0.1681, 0.0985, 0.0554, 0.1686], device='cuda:1'), in_proj_covar=tensor([0.0138, 0.0191, 0.0209, 0.0254, 0.0186, 0.0202, 0.0183, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:09:15,126 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-17 05:09:24,736 INFO [train.py:893] (1/4) Epoch 26, batch 0, loss[loss=0.1731, simple_loss=0.2324, pruned_loss=0.05694, over 13546.00 frames. ], tot_loss[loss=0.1731, simple_loss=0.2324, pruned_loss=0.05694, over 13546.00 frames. ], batch size: 70, lr: 5.06e-03, grad_scale: 16.0 2023-04-17 05:09:24,736 INFO [train.py:918] (1/4) Computing validation loss 2023-04-17 05:09:40,736 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8674, 2.4572, 2.5606, 2.9903, 2.1549, 2.9387, 2.9588, 2.5210], device='cuda:1'), covar=tensor([0.0090, 0.0287, 0.0165, 0.0161, 0.0284, 0.0161, 0.0173, 0.0207], device='cuda:1'), in_proj_covar=tensor([0.0112, 0.0125, 0.0130, 0.0130, 0.0141, 0.0117, 0.0114, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 05:09:46,367 INFO [train.py:927] (1/4) Epoch 26, validation: loss=0.1333, simple_loss=0.1928, pruned_loss=0.03689, over 2446609.00 frames. 2023-04-17 05:09:46,368 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12831MB 2023-04-17 05:10:19,070 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67513.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:10:30,053 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.826e+02 2.528e+02 2.949e+02 3.569e+02 5.747e+02, threshold=5.898e+02, percent-clipped=0.0 2023-04-17 05:10:33,220 INFO [train.py:893] (1/4) Epoch 26, batch 50, loss[loss=0.1366, simple_loss=0.1954, pruned_loss=0.03888, over 13181.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2276, pruned_loss=0.05345, over 602501.28 frames. ], batch size: 58, lr: 5.06e-03, grad_scale: 16.0 2023-04-17 05:10:41,767 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.1693, 1.8482, 3.4164, 3.3239, 3.2745, 2.6880, 3.1343, 2.6753], device='cuda:1'), covar=tensor([0.1696, 0.1406, 0.0189, 0.0240, 0.0263, 0.0741, 0.0281, 0.0744], device='cuda:1'), in_proj_covar=tensor([0.0218, 0.0184, 0.0130, 0.0135, 0.0137, 0.0177, 0.0149, 0.0187], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-17 05:10:55,632 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-17 05:10:55,633 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-17 05:10:55,633 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-17 05:10:55,648 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-17 05:10:55,656 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-17 05:10:55,676 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-17 05:10:55,685 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-17 05:11:03,743 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1584, 2.4054, 2.0561, 4.0661, 4.5115, 3.3595, 4.4365, 4.2601], device='cuda:1'), covar=tensor([0.0098, 0.1040, 0.1120, 0.0102, 0.0076, 0.0487, 0.0083, 0.0088], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0092, 0.0100, 0.0084, 0.0071, 0.0083, 0.0058, 0.0081], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 05:11:15,236 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67574.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 05:11:17,748 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1880, 2.0594, 2.4322, 3.5182, 3.1663, 3.5325, 2.8873, 2.2679], device='cuda:1'), covar=tensor([0.0254, 0.0856, 0.0772, 0.0073, 0.0273, 0.0077, 0.0625, 0.0881], device='cuda:1'), in_proj_covar=tensor([0.0121, 0.0148, 0.0164, 0.0101, 0.0119, 0.0098, 0.0169, 0.0147], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:11:18,302 INFO [train.py:893] (1/4) Epoch 26, batch 100, loss[loss=0.1828, simple_loss=0.2396, pruned_loss=0.06295, over 13139.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.229, pruned_loss=0.05433, over 1059407.35 frames. ], batch size: 142, lr: 5.06e-03, grad_scale: 16.0 2023-04-17 05:11:50,604 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67613.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:12:02,531 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.938e+02 2.463e+02 2.907e+02 3.549e+02 6.282e+02, threshold=5.815e+02, percent-clipped=1.0 2023-04-17 05:12:05,239 INFO [train.py:893] (1/4) Epoch 26, batch 150, loss[loss=0.1676, simple_loss=0.2148, pruned_loss=0.06017, over 13209.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2295, pruned_loss=0.05501, over 1408337.92 frames. ], batch size: 58, lr: 5.06e-03, grad_scale: 16.0 2023-04-17 05:12:12,943 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.3231, 4.8682, 4.6917, 4.7810, 4.5773, 4.6803, 5.2252, 4.8280], device='cuda:1'), covar=tensor([0.0685, 0.1234, 0.2228, 0.2572, 0.0903, 0.1649, 0.0888, 0.1141], device='cuda:1'), in_proj_covar=tensor([0.0344, 0.0398, 0.0484, 0.0488, 0.0309, 0.0364, 0.0451, 0.0361], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 05:12:35,580 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67661.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:12:50,112 INFO [train.py:893] (1/4) Epoch 26, batch 200, loss[loss=0.1762, simple_loss=0.2366, pruned_loss=0.05786, over 13527.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2319, pruned_loss=0.05586, over 1670938.60 frames. ], batch size: 87, lr: 5.05e-03, grad_scale: 16.0 2023-04-17 05:12:52,099 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67680.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:13:03,900 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67692.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:13:33,556 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.787e+02 2.465e+02 2.845e+02 3.399e+02 5.580e+02, threshold=5.689e+02, percent-clipped=0.0 2023-04-17 05:13:36,053 INFO [train.py:893] (1/4) Epoch 26, batch 250, loss[loss=0.1491, simple_loss=0.214, pruned_loss=0.04212, over 13490.00 frames. ], tot_loss[loss=0.1718, simple_loss=0.2321, pruned_loss=0.05574, over 1892515.98 frames. ], batch size: 81, lr: 5.05e-03, grad_scale: 16.0 2023-04-17 05:13:44,431 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.7059, 5.1049, 4.8705, 4.8636, 4.9761, 4.7144, 5.1412, 5.1341], device='cuda:1'), covar=tensor([0.0206, 0.0195, 0.0195, 0.0331, 0.0224, 0.0234, 0.0244, 0.0209], device='cuda:1'), in_proj_covar=tensor([0.0144, 0.0210, 0.0174, 0.0188, 0.0159, 0.0209, 0.0139, 0.0188], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:13:46,187 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=67739.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:13:47,927 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67741.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:13:59,452 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=67753.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 05:14:20,962 INFO [train.py:893] (1/4) Epoch 26, batch 300, loss[loss=0.1955, simple_loss=0.2581, pruned_loss=0.06646, over 13331.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2333, pruned_loss=0.05655, over 2059767.06 frames. ], batch size: 118, lr: 5.05e-03, grad_scale: 16.0 2023-04-17 05:14:29,765 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=67787.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:14:31,084 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 2023-04-17 05:15:00,155 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0994, 2.9841, 2.7096, 1.9556, 2.1288, 2.5082, 2.6595, 3.2174], device='cuda:1'), covar=tensor([0.1029, 0.0349, 0.0595, 0.1644, 0.0528, 0.0620, 0.0800, 0.0288], device='cuda:1'), in_proj_covar=tensor([0.0176, 0.0153, 0.0127, 0.0215, 0.0118, 0.0174, 0.0184, 0.0139], device='cuda:1'), out_proj_covar=tensor([1.2990e-04, 1.1403e-04, 9.8706e-05, 1.5949e-04, 8.5319e-05, 1.3143e-04, 1.3827e-04, 1.0218e-04], device='cuda:1') 2023-04-17 05:15:03,942 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.916e+02 2.664e+02 3.111e+02 3.733e+02 5.868e+02, threshold=6.222e+02, percent-clipped=2.0 2023-04-17 05:15:04,433 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-04-17 05:15:06,380 INFO [train.py:893] (1/4) Epoch 26, batch 350, loss[loss=0.1486, simple_loss=0.204, pruned_loss=0.04663, over 13416.00 frames. ], tot_loss[loss=0.1742, simple_loss=0.2342, pruned_loss=0.05712, over 2195608.13 frames. ], batch size: 62, lr: 5.05e-03, grad_scale: 16.0 2023-04-17 05:15:45,643 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=67869.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 05:15:53,462 INFO [train.py:893] (1/4) Epoch 26, batch 400, loss[loss=0.1726, simple_loss=0.2371, pruned_loss=0.05405, over 13430.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2343, pruned_loss=0.05661, over 2294572.76 frames. ], batch size: 95, lr: 5.05e-03, grad_scale: 16.0 2023-04-17 05:16:36,538 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.742e+02 2.634e+02 2.859e+02 3.427e+02 7.730e+02, threshold=5.718e+02, percent-clipped=1.0 2023-04-17 05:16:38,922 INFO [train.py:893] (1/4) Epoch 26, batch 450, loss[loss=0.1753, simple_loss=0.2384, pruned_loss=0.05611, over 13532.00 frames. ], tot_loss[loss=0.1747, simple_loss=0.2357, pruned_loss=0.05688, over 2375248.40 frames. ], batch size: 98, lr: 5.04e-03, grad_scale: 16.0 2023-04-17 05:17:04,165 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-17 05:17:25,150 INFO [train.py:893] (1/4) Epoch 26, batch 500, loss[loss=0.1488, simple_loss=0.2115, pruned_loss=0.04304, over 13490.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2351, pruned_loss=0.05589, over 2441284.96 frames. ], batch size: 70, lr: 5.04e-03, grad_scale: 16.0 2023-04-17 05:17:27,953 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=67981.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:18:12,049 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.734e+02 2.466e+02 2.871e+02 3.533e+02 1.025e+03, threshold=5.741e+02, percent-clipped=2.0 2023-04-17 05:18:14,573 INFO [train.py:893] (1/4) Epoch 26, batch 550, loss[loss=0.1684, simple_loss=0.23, pruned_loss=0.05339, over 13530.00 frames. ], tot_loss[loss=0.1735, simple_loss=0.2352, pruned_loss=0.05586, over 2488467.79 frames. ], batch size: 85, lr: 5.04e-03, grad_scale: 16.0 2023-04-17 05:18:22,722 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68036.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:18:27,837 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68042.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:18:32,594 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68048.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 05:18:47,119 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1379, 2.8010, 3.3822, 2.7795, 2.3698, 2.4977, 3.6713, 3.7403], device='cuda:1'), covar=tensor([0.1140, 0.1957, 0.0355, 0.1483, 0.1685, 0.1544, 0.0304, 0.0270], device='cuda:1'), in_proj_covar=tensor([0.0252, 0.0278, 0.0202, 0.0228, 0.0225, 0.0186, 0.0218, 0.0192], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0002, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 05:18:59,725 INFO [train.py:893] (1/4) Epoch 26, batch 600, loss[loss=0.1461, simple_loss=0.2134, pruned_loss=0.03942, over 13489.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2337, pruned_loss=0.05548, over 2523279.18 frames. ], batch size: 81, lr: 5.04e-03, grad_scale: 16.0 2023-04-17 05:19:07,970 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68087.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 05:19:28,103 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2931, 4.7202, 4.4780, 4.5521, 4.5316, 4.3673, 4.7696, 4.7833], device='cuda:1'), covar=tensor([0.0216, 0.0194, 0.0185, 0.0280, 0.0236, 0.0224, 0.0236, 0.0200], device='cuda:1'), in_proj_covar=tensor([0.0143, 0.0208, 0.0172, 0.0185, 0.0158, 0.0208, 0.0139, 0.0188], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:19:43,076 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.786e+02 2.523e+02 2.959e+02 3.513e+02 6.901e+02, threshold=5.918e+02, percent-clipped=1.0 2023-04-17 05:19:45,591 INFO [train.py:893] (1/4) Epoch 26, batch 650, loss[loss=0.1631, simple_loss=0.2308, pruned_loss=0.04768, over 13241.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2329, pruned_loss=0.05491, over 2555732.89 frames. ], batch size: 124, lr: 5.04e-03, grad_scale: 16.0 2023-04-17 05:20:03,538 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68148.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 05:20:11,549 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-04-17 05:20:22,626 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68169.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:20:30,599 INFO [train.py:893] (1/4) Epoch 26, batch 700, loss[loss=0.1649, simple_loss=0.2279, pruned_loss=0.05091, over 13537.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.232, pruned_loss=0.0545, over 2581200.63 frames. ], batch size: 87, lr: 5.04e-03, grad_scale: 16.0 2023-04-17 05:20:48,630 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-04-17 05:21:06,373 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68217.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:21:13,387 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.625e+02 2.575e+02 3.034e+02 3.634e+02 6.033e+02, threshold=6.069e+02, percent-clipped=2.0 2023-04-17 05:21:15,945 INFO [train.py:893] (1/4) Epoch 26, batch 750, loss[loss=0.1726, simple_loss=0.2278, pruned_loss=0.0587, over 13525.00 frames. ], tot_loss[loss=0.1714, simple_loss=0.2324, pruned_loss=0.05524, over 2597795.08 frames. ], batch size: 70, lr: 5.03e-03, grad_scale: 16.0 2023-04-17 05:21:33,922 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1133, 4.4037, 3.2741, 2.9751, 3.1686, 2.6883, 4.5125, 2.5056], device='cuda:1'), covar=tensor([0.1772, 0.0314, 0.1401, 0.2390, 0.0900, 0.3446, 0.0241, 0.4025], device='cuda:1'), in_proj_covar=tensor([0.0318, 0.0295, 0.0326, 0.0347, 0.0267, 0.0335, 0.0221, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 05:21:46,311 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.33 vs. limit=2.0 2023-04-17 05:22:01,810 INFO [train.py:893] (1/4) Epoch 26, batch 800, loss[loss=0.1652, simple_loss=0.2261, pruned_loss=0.05209, over 13015.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2334, pruned_loss=0.05544, over 2611074.83 frames. ], batch size: 142, lr: 5.03e-03, grad_scale: 16.0 2023-04-17 05:22:45,506 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.980e+02 2.741e+02 3.112e+02 3.676e+02 5.585e+02, threshold=6.224e+02, percent-clipped=0.0 2023-04-17 05:22:47,973 INFO [train.py:893] (1/4) Epoch 26, batch 850, loss[loss=0.1844, simple_loss=0.2466, pruned_loss=0.06108, over 13450.00 frames. ], tot_loss[loss=0.1732, simple_loss=0.2351, pruned_loss=0.05569, over 2625172.83 frames. ], batch size: 103, lr: 5.03e-03, grad_scale: 16.0 2023-04-17 05:22:55,867 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68336.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:22:57,302 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68337.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:23:07,066 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68348.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:23:07,222 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7032, 3.5637, 2.8196, 2.4976, 2.5904, 2.2999, 3.6147, 2.1465], device='cuda:1'), covar=tensor([0.1598, 0.0390, 0.1263, 0.2274, 0.0904, 0.3397, 0.0368, 0.4069], device='cuda:1'), in_proj_covar=tensor([0.0318, 0.0296, 0.0326, 0.0347, 0.0267, 0.0336, 0.0221, 0.0373], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 05:23:34,142 INFO [train.py:893] (1/4) Epoch 26, batch 900, loss[loss=0.1747, simple_loss=0.2308, pruned_loss=0.05927, over 13066.00 frames. ], tot_loss[loss=0.1733, simple_loss=0.2347, pruned_loss=0.056, over 2628094.13 frames. ], batch size: 142, lr: 5.03e-03, grad_scale: 16.0 2023-04-17 05:23:39,855 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68384.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:23:50,505 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68396.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:23:53,647 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68399.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:23:57,435 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9111, 2.6499, 2.5990, 3.0188, 2.3498, 3.0845, 3.0561, 2.5366], device='cuda:1'), covar=tensor([0.0091, 0.0206, 0.0170, 0.0172, 0.0230, 0.0140, 0.0168, 0.0202], device='cuda:1'), in_proj_covar=tensor([0.0112, 0.0124, 0.0130, 0.0130, 0.0140, 0.0118, 0.0115, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 05:24:03,594 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-17 05:24:16,637 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.847e+02 2.678e+02 3.031e+02 3.663e+02 7.979e+02, threshold=6.063e+02, percent-clipped=2.0 2023-04-17 05:24:19,106 INFO [train.py:893] (1/4) Epoch 26, batch 950, loss[loss=0.1789, simple_loss=0.2334, pruned_loss=0.06217, over 13536.00 frames. ], tot_loss[loss=0.1738, simple_loss=0.2342, pruned_loss=0.05668, over 2624864.01 frames. ], batch size: 85, lr: 5.03e-03, grad_scale: 16.0 2023-04-17 05:24:24,958 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3864, 3.1787, 3.8773, 2.7968, 2.6089, 2.7333, 4.1665, 4.2565], device='cuda:1'), covar=tensor([0.1370, 0.1834, 0.0408, 0.1913, 0.1706, 0.1551, 0.0310, 0.0272], device='cuda:1'), in_proj_covar=tensor([0.0254, 0.0279, 0.0204, 0.0230, 0.0226, 0.0187, 0.0219, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:24:33,745 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68443.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 05:24:49,374 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68460.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:25:05,288 INFO [train.py:893] (1/4) Epoch 26, batch 1000, loss[loss=0.183, simple_loss=0.2387, pruned_loss=0.06361, over 13382.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2322, pruned_loss=0.05584, over 2634527.94 frames. ], batch size: 109, lr: 5.02e-03, grad_scale: 16.0 2023-04-17 05:25:09,475 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4816, 2.5119, 2.3785, 4.3935, 4.7846, 3.5421, 4.7301, 4.5938], device='cuda:1'), covar=tensor([0.0161, 0.1222, 0.1265, 0.0144, 0.0209, 0.0663, 0.0121, 0.0123], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0092, 0.0101, 0.0084, 0.0072, 0.0082, 0.0059, 0.0082], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 05:25:48,340 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.804e+02 2.639e+02 2.912e+02 3.425e+02 6.549e+02, threshold=5.823e+02, percent-clipped=1.0 2023-04-17 05:25:51,460 INFO [train.py:893] (1/4) Epoch 26, batch 1050, loss[loss=0.1448, simple_loss=0.2064, pruned_loss=0.04158, over 13441.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2308, pruned_loss=0.05484, over 2642429.55 frames. ], batch size: 65, lr: 5.02e-03, grad_scale: 16.0 2023-04-17 05:26:04,760 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2737, 4.1646, 4.2357, 2.6618, 4.5592, 4.3194, 4.2661, 4.5111], device='cuda:1'), covar=tensor([0.0235, 0.0134, 0.0128, 0.1058, 0.0132, 0.0243, 0.0141, 0.0099], device='cuda:1'), in_proj_covar=tensor([0.0098, 0.0060, 0.0088, 0.0106, 0.0103, 0.0116, 0.0086, 0.0081], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:26:36,937 INFO [train.py:893] (1/4) Epoch 26, batch 1100, loss[loss=0.178, simple_loss=0.2483, pruned_loss=0.05388, over 13411.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2305, pruned_loss=0.05406, over 2643929.90 frames. ], batch size: 113, lr: 5.02e-03, grad_scale: 16.0 2023-04-17 05:26:37,205 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68578.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:26:48,894 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-04-17 05:27:20,082 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.731e+02 2.635e+02 3.013e+02 3.644e+02 5.346e+02, threshold=6.026e+02, percent-clipped=0.0 2023-04-17 05:27:22,551 INFO [train.py:893] (1/4) Epoch 26, batch 1150, loss[loss=0.1543, simple_loss=0.215, pruned_loss=0.04683, over 13347.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2308, pruned_loss=0.05368, over 2646747.52 frames. ], batch size: 73, lr: 5.02e-03, grad_scale: 16.0 2023-04-17 05:27:30,193 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68637.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:27:32,557 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68639.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:28:07,331 INFO [train.py:893] (1/4) Epoch 26, batch 1200, loss[loss=0.1515, simple_loss=0.2194, pruned_loss=0.04176, over 13287.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2306, pruned_loss=0.05319, over 2648229.94 frames. ], batch size: 124, lr: 5.02e-03, grad_scale: 16.0 2023-04-17 05:28:13,923 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68685.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:28:14,038 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=68685.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:28:17,672 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-04-17 05:28:31,957 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-17 05:28:37,653 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.9564, 4.4946, 4.3486, 4.5120, 4.2262, 4.3558, 4.9095, 4.4072], device='cuda:1'), covar=tensor([0.0826, 0.1144, 0.2136, 0.2570, 0.1080, 0.1571, 0.0906, 0.1216], device='cuda:1'), in_proj_covar=tensor([0.0352, 0.0405, 0.0497, 0.0498, 0.0316, 0.0373, 0.0459, 0.0370], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:28:44,139 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-17 05:28:50,387 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.795e+02 2.463e+02 2.772e+02 3.545e+02 7.103e+02, threshold=5.544e+02, percent-clipped=1.0 2023-04-17 05:28:52,807 INFO [train.py:893] (1/4) Epoch 26, batch 1250, loss[loss=0.1465, simple_loss=0.2029, pruned_loss=0.04501, over 13227.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2313, pruned_loss=0.05363, over 2647953.29 frames. ], batch size: 58, lr: 5.02e-03, grad_scale: 32.0 2023-04-17 05:29:07,547 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=68743.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 05:29:08,499 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6925, 3.8229, 2.8856, 2.6275, 2.7034, 2.3954, 3.8526, 2.2462], device='cuda:1'), covar=tensor([0.1829, 0.0401, 0.1386, 0.2287, 0.0931, 0.3442, 0.0361, 0.4321], device='cuda:1'), in_proj_covar=tensor([0.0320, 0.0298, 0.0328, 0.0350, 0.0268, 0.0339, 0.0223, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 05:29:10,047 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=68746.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:29:18,085 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68755.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:29:39,306 INFO [train.py:893] (1/4) Epoch 26, batch 1300, loss[loss=0.166, simple_loss=0.2341, pruned_loss=0.04897, over 13440.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2322, pruned_loss=0.05384, over 2650831.51 frames. ], batch size: 103, lr: 5.01e-03, grad_scale: 32.0 2023-04-17 05:29:50,699 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=68791.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 05:30:02,586 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6434, 4.1420, 4.1482, 4.2320, 3.9591, 3.9779, 4.6179, 4.1231], device='cuda:1'), covar=tensor([0.0861, 0.1488, 0.2267, 0.2508, 0.1197, 0.1816, 0.1096, 0.1509], device='cuda:1'), in_proj_covar=tensor([0.0354, 0.0407, 0.0498, 0.0500, 0.0318, 0.0374, 0.0462, 0.0371], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:30:22,443 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.892e+02 2.771e+02 3.228e+02 3.857e+02 7.338e+02, threshold=6.456e+02, percent-clipped=3.0 2023-04-17 05:30:24,977 INFO [train.py:893] (1/4) Epoch 26, batch 1350, loss[loss=0.1764, simple_loss=0.2404, pruned_loss=0.05616, over 13540.00 frames. ], tot_loss[loss=0.1706, simple_loss=0.2328, pruned_loss=0.05424, over 2650257.11 frames. ], batch size: 87, lr: 5.01e-03, grad_scale: 32.0 2023-04-17 05:30:34,114 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9229, 4.0946, 3.1492, 2.8314, 2.8964, 2.5733, 4.1596, 2.3939], device='cuda:1'), covar=tensor([0.1725, 0.0377, 0.1298, 0.2290, 0.0896, 0.3270, 0.0281, 0.4198], device='cuda:1'), in_proj_covar=tensor([0.0321, 0.0299, 0.0329, 0.0351, 0.0269, 0.0340, 0.0224, 0.0376], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 05:31:11,703 INFO [train.py:893] (1/4) Epoch 26, batch 1400, loss[loss=0.1814, simple_loss=0.2421, pruned_loss=0.06033, over 13427.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2323, pruned_loss=0.05398, over 2656295.35 frames. ], batch size: 95, lr: 5.01e-03, grad_scale: 32.0 2023-04-17 05:31:54,580 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.787e+02 2.459e+02 2.883e+02 3.453e+02 7.124e+02, threshold=5.766e+02, percent-clipped=1.0 2023-04-17 05:31:57,795 INFO [train.py:893] (1/4) Epoch 26, batch 1450, loss[loss=0.1615, simple_loss=0.2298, pruned_loss=0.04658, over 13450.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2313, pruned_loss=0.05367, over 2656274.56 frames. ], batch size: 106, lr: 5.01e-03, grad_scale: 32.0 2023-04-17 05:32:04,115 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=68934.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:32:04,653 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-04-17 05:32:25,280 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6796, 3.9451, 3.8086, 3.8080, 3.8977, 3.6860, 3.9843, 4.0359], device='cuda:1'), covar=tensor([0.0246, 0.0286, 0.0258, 0.0342, 0.0273, 0.0303, 0.0271, 0.0286], device='cuda:1'), in_proj_covar=tensor([0.0146, 0.0212, 0.0176, 0.0190, 0.0160, 0.0211, 0.0140, 0.0191], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:32:43,396 INFO [train.py:893] (1/4) Epoch 26, batch 1500, loss[loss=0.159, simple_loss=0.2126, pruned_loss=0.05269, over 12802.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2322, pruned_loss=0.05383, over 2658826.50 frames. ], batch size: 52, lr: 5.01e-03, grad_scale: 32.0 2023-04-17 05:33:25,077 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.553e+02 2.658e+02 3.045e+02 3.572e+02 6.026e+02, threshold=6.090e+02, percent-clipped=3.0 2023-04-17 05:33:28,348 INFO [train.py:893] (1/4) Epoch 26, batch 1550, loss[loss=0.1836, simple_loss=0.2438, pruned_loss=0.06174, over 13356.00 frames. ], tot_loss[loss=0.1696, simple_loss=0.2318, pruned_loss=0.05367, over 2650990.09 frames. ], batch size: 118, lr: 5.00e-03, grad_scale: 32.0 2023-04-17 05:33:39,919 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69041.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:33:41,674 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69043.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:33:52,964 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69055.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:34:12,710 INFO [train.py:893] (1/4) Epoch 26, batch 1600, loss[loss=0.1488, simple_loss=0.2048, pruned_loss=0.04646, over 13197.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2332, pruned_loss=0.05445, over 2646778.37 frames. ], batch size: 58, lr: 5.00e-03, grad_scale: 32.0 2023-04-17 05:34:36,258 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69103.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:34:37,257 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69104.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:34:53,998 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9767, 4.3188, 3.9979, 4.0691, 4.0704, 4.4090, 4.2362, 4.0062], device='cuda:1'), covar=tensor([0.0318, 0.0255, 0.0352, 0.0829, 0.0308, 0.0263, 0.0325, 0.0400], device='cuda:1'), in_proj_covar=tensor([0.0169, 0.0165, 0.0186, 0.0272, 0.0185, 0.0202, 0.0183, 0.0114], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 05:34:55,975 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-04-17 05:34:56,247 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.041e+02 2.558e+02 2.916e+02 3.500e+02 7.035e+02, threshold=5.832e+02, percent-clipped=3.0 2023-04-17 05:34:58,809 INFO [train.py:893] (1/4) Epoch 26, batch 1650, loss[loss=0.1783, simple_loss=0.2466, pruned_loss=0.05506, over 13431.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.233, pruned_loss=0.05366, over 2650257.16 frames. ], batch size: 106, lr: 5.00e-03, grad_scale: 32.0 2023-04-17 05:35:30,461 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69162.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:35:35,463 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69168.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:35:43,446 INFO [train.py:893] (1/4) Epoch 26, batch 1700, loss[loss=0.1831, simple_loss=0.2511, pruned_loss=0.0576, over 13559.00 frames. ], tot_loss[loss=0.1705, simple_loss=0.2338, pruned_loss=0.05365, over 2650325.28 frames. ], batch size: 89, lr: 5.00e-03, grad_scale: 32.0 2023-04-17 05:36:18,828 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6696, 3.3505, 4.2183, 2.9964, 2.8853, 2.8020, 4.5233, 4.6133], device='cuda:1'), covar=tensor([0.1351, 0.2070, 0.0362, 0.1866, 0.1640, 0.1842, 0.0256, 0.0219], device='cuda:1'), in_proj_covar=tensor([0.0254, 0.0281, 0.0204, 0.0231, 0.0226, 0.0187, 0.0220, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:36:26,017 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69223.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:36:27,384 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.818e+02 2.684e+02 3.102e+02 3.690e+02 9.152e+02, threshold=6.204e+02, percent-clipped=1.0 2023-04-17 05:36:27,733 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69225.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:36:29,896 INFO [train.py:893] (1/4) Epoch 26, batch 1750, loss[loss=0.1645, simple_loss=0.2276, pruned_loss=0.05074, over 13559.00 frames. ], tot_loss[loss=0.1695, simple_loss=0.2322, pruned_loss=0.05337, over 2648920.02 frames. ], batch size: 89, lr: 5.00e-03, grad_scale: 32.0 2023-04-17 05:36:31,050 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69229.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:36:34,877 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69234.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:36:45,559 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69245.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 05:37:10,405 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69273.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:37:15,600 INFO [train.py:893] (1/4) Epoch 26, batch 1800, loss[loss=0.1695, simple_loss=0.2265, pruned_loss=0.05622, over 11777.00 frames. ], tot_loss[loss=0.1688, simple_loss=0.2313, pruned_loss=0.05317, over 2649156.31 frames. ], batch size: 157, lr: 5.00e-03, grad_scale: 32.0 2023-04-17 05:37:18,954 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69282.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:37:23,136 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69286.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:37:34,644 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6220, 5.0808, 4.8649, 4.8930, 4.9221, 4.6926, 5.1434, 5.1345], device='cuda:1'), covar=tensor([0.0203, 0.0188, 0.0207, 0.0287, 0.0215, 0.0238, 0.0260, 0.0190], device='cuda:1'), in_proj_covar=tensor([0.0148, 0.0213, 0.0178, 0.0190, 0.0162, 0.0213, 0.0142, 0.0193], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-04-17 05:37:41,107 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69306.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 05:37:43,063 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.60 vs. limit=2.0 2023-04-17 05:37:53,204 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69320.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:37:56,843 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.794e+02 2.500e+02 2.995e+02 3.542e+02 6.193e+02, threshold=5.991e+02, percent-clipped=0.0 2023-04-17 05:37:59,405 INFO [train.py:893] (1/4) Epoch 26, batch 1850, loss[loss=0.1385, simple_loss=0.1972, pruned_loss=0.03989, over 13197.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2309, pruned_loss=0.05287, over 2653141.28 frames. ], batch size: 58, lr: 4.99e-03, grad_scale: 32.0 2023-04-17 05:38:02,699 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-17 05:38:04,482 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5796, 2.5187, 2.9747, 4.1328, 3.6900, 4.1621, 3.2774, 2.5946], device='cuda:1'), covar=tensor([0.0313, 0.0940, 0.0771, 0.0067, 0.0228, 0.0060, 0.0671, 0.0941], device='cuda:1'), in_proj_covar=tensor([0.0125, 0.0150, 0.0168, 0.0104, 0.0122, 0.0100, 0.0172, 0.0152], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:38:04,521 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69334.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 05:38:10,009 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.8679, 4.5658, 4.6246, 4.6561, 5.1277, 4.3227, 5.1472, 5.1103], device='cuda:1'), covar=tensor([0.0826, 0.1285, 0.1461, 0.1108, 0.1196, 0.1984, 0.0828, 0.0899], device='cuda:1'), in_proj_covar=tensor([0.0312, 0.0315, 0.0322, 0.0241, 0.0463, 0.0364, 0.0303, 0.0319], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:38:10,771 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69341.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:38:35,940 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69368.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:38:44,612 INFO [train.py:893] (1/4) Epoch 26, batch 1900, loss[loss=0.1717, simple_loss=0.2365, pruned_loss=0.05343, over 13367.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2302, pruned_loss=0.0529, over 2652101.17 frames. ], batch size: 109, lr: 4.99e-03, grad_scale: 32.0 2023-04-17 05:38:47,389 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69381.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:38:53,901 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69389.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:38:58,121 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69394.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:39:03,435 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69399.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:39:15,621 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-04-17 05:39:26,382 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.879e+02 2.463e+02 2.828e+02 3.585e+02 5.409e+02, threshold=5.655e+02, percent-clipped=0.0 2023-04-17 05:39:29,045 INFO [train.py:893] (1/4) Epoch 26, batch 1950, loss[loss=0.1715, simple_loss=0.229, pruned_loss=0.05695, over 13511.00 frames. ], tot_loss[loss=0.168, simple_loss=0.23, pruned_loss=0.05296, over 2653712.17 frames. ], batch size: 70, lr: 4.99e-03, grad_scale: 32.0 2023-04-17 05:39:30,249 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69429.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:39:37,384 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6809, 2.5090, 2.0051, 3.6675, 3.9504, 3.1273, 3.9616, 3.7753], device='cuda:1'), covar=tensor([0.0081, 0.0885, 0.0996, 0.0084, 0.0072, 0.0411, 0.0072, 0.0089], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0093, 0.0102, 0.0085, 0.0072, 0.0083, 0.0059, 0.0082], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 05:39:53,792 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69455.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:40:15,394 INFO [train.py:893] (1/4) Epoch 26, batch 2000, loss[loss=0.1899, simple_loss=0.251, pruned_loss=0.06437, over 13409.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2326, pruned_loss=0.05436, over 2650390.82 frames. ], batch size: 95, lr: 4.99e-03, grad_scale: 32.0 2023-04-17 05:40:21,113 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-17 05:40:51,441 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69518.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:40:57,007 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69524.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:40:57,627 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.874e+02 2.941e+02 3.266e+02 3.787e+02 7.734e+02, threshold=6.532e+02, percent-clipped=5.0 2023-04-17 05:41:00,880 INFO [train.py:893] (1/4) Epoch 26, batch 2050, loss[loss=0.185, simple_loss=0.2493, pruned_loss=0.06035, over 13404.00 frames. ], tot_loss[loss=0.1723, simple_loss=0.2345, pruned_loss=0.05503, over 2655440.08 frames. ], batch size: 113, lr: 4.99e-03, grad_scale: 32.0 2023-04-17 05:41:46,302 INFO [train.py:893] (1/4) Epoch 26, batch 2100, loss[loss=0.1587, simple_loss=0.221, pruned_loss=0.04818, over 13514.00 frames. ], tot_loss[loss=0.1709, simple_loss=0.2331, pruned_loss=0.05436, over 2652497.81 frames. ], batch size: 70, lr: 4.98e-03, grad_scale: 32.0 2023-04-17 05:41:48,933 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69581.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:42:07,620 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69601.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 05:42:29,063 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.741e+02 2.433e+02 2.898e+02 3.474e+02 5.673e+02, threshold=5.796e+02, percent-clipped=0.0 2023-04-17 05:42:32,237 INFO [train.py:893] (1/4) Epoch 26, batch 2150, loss[loss=0.1482, simple_loss=0.2199, pruned_loss=0.0383, over 13540.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2326, pruned_loss=0.05335, over 2659891.48 frames. ], batch size: 85, lr: 4.98e-03, grad_scale: 32.0 2023-04-17 05:42:33,155 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69629.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 05:43:01,351 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7192, 3.4363, 3.6418, 2.2345, 3.7826, 3.7150, 3.6592, 3.8138], device='cuda:1'), covar=tensor([0.0192, 0.0172, 0.0138, 0.1190, 0.0131, 0.0181, 0.0108, 0.0104], device='cuda:1'), in_proj_covar=tensor([0.0098, 0.0061, 0.0088, 0.0107, 0.0104, 0.0116, 0.0087, 0.0082], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:43:04,179 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.77 vs. limit=5.0 2023-04-17 05:43:12,802 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=69674.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 05:43:15,013 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69676.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:43:17,085 INFO [train.py:893] (1/4) Epoch 26, batch 2200, loss[loss=0.1677, simple_loss=0.2306, pruned_loss=0.0524, over 13115.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.232, pruned_loss=0.05272, over 2661240.97 frames. ], batch size: 142, lr: 4.98e-03, grad_scale: 32.0 2023-04-17 05:43:35,839 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69699.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:43:58,445 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69724.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:43:59,053 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.856e+02 2.542e+02 2.911e+02 3.538e+02 6.573e+02, threshold=5.821e+02, percent-clipped=1.0 2023-04-17 05:44:01,565 INFO [train.py:893] (1/4) Epoch 26, batch 2250, loss[loss=0.1851, simple_loss=0.2454, pruned_loss=0.06245, over 13481.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2321, pruned_loss=0.05313, over 2660973.20 frames. ], batch size: 100, lr: 4.98e-03, grad_scale: 32.0 2023-04-17 05:44:07,636 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=69735.0, num_to_drop=1, layers_to_drop={2} 2023-04-17 05:44:19,090 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69747.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:44:21,570 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=69750.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:44:46,008 INFO [train.py:893] (1/4) Epoch 26, batch 2300, loss[loss=0.1777, simple_loss=0.2355, pruned_loss=0.05998, over 13038.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2307, pruned_loss=0.05226, over 2659429.73 frames. ], batch size: 142, lr: 4.98e-03, grad_scale: 32.0 2023-04-17 05:45:23,291 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69818.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:45:23,650 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-04-17 05:45:28,997 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69824.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:45:29,582 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.562e+02 2.373e+02 2.733e+02 3.115e+02 4.841e+02, threshold=5.467e+02, percent-clipped=0.0 2023-04-17 05:45:32,195 INFO [train.py:893] (1/4) Epoch 26, batch 2350, loss[loss=0.1389, simple_loss=0.2053, pruned_loss=0.03629, over 13422.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2298, pruned_loss=0.0518, over 2659683.59 frames. ], batch size: 65, lr: 4.98e-03, grad_scale: 32.0 2023-04-17 05:45:34,191 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3534, 2.5283, 2.1875, 4.3150, 4.8107, 3.5454, 4.7108, 4.4655], device='cuda:1'), covar=tensor([0.0087, 0.0974, 0.1032, 0.0096, 0.0064, 0.0456, 0.0071, 0.0086], device='cuda:1'), in_proj_covar=tensor([0.0081, 0.0094, 0.0103, 0.0086, 0.0073, 0.0084, 0.0060, 0.0083], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 05:45:49,330 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.9352, 4.7147, 5.0219, 4.8587, 5.2342, 4.7353, 5.2408, 5.1960], device='cuda:1'), covar=tensor([0.0417, 0.0632, 0.0577, 0.0580, 0.0517, 0.0866, 0.0439, 0.0447], device='cuda:1'), in_proj_covar=tensor([0.0309, 0.0312, 0.0321, 0.0238, 0.0457, 0.0358, 0.0300, 0.0315], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:45:55,778 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-17 05:46:07,480 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69866.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:46:12,975 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69872.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:46:18,620 INFO [train.py:893] (1/4) Epoch 26, batch 2400, loss[loss=0.1532, simple_loss=0.2167, pruned_loss=0.04479, over 13424.00 frames. ], tot_loss[loss=0.1654, simple_loss=0.2287, pruned_loss=0.05109, over 2663548.83 frames. ], batch size: 62, lr: 4.97e-03, grad_scale: 32.0 2023-04-17 05:46:19,895 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 2023-04-17 05:46:22,036 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69881.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:46:26,780 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.5333, 5.0718, 4.8522, 4.9904, 4.8043, 4.8630, 5.4952, 5.0877], device='cuda:1'), covar=tensor([0.0619, 0.1178, 0.2266, 0.2445, 0.0957, 0.1564, 0.0790, 0.1111], device='cuda:1'), in_proj_covar=tensor([0.0349, 0.0400, 0.0489, 0.0493, 0.0313, 0.0366, 0.0453, 0.0366], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 05:46:38,714 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69901.0, num_to_drop=1, layers_to_drop={2} 2023-04-17 05:46:45,297 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0115, 3.8850, 3.2234, 3.5841, 3.1676, 2.5057, 3.9871, 2.3610], device='cuda:1'), covar=tensor([0.0769, 0.0534, 0.0512, 0.0399, 0.0744, 0.1901, 0.0928, 0.1395], device='cuda:1'), in_proj_covar=tensor([0.0150, 0.0151, 0.0141, 0.0123, 0.0155, 0.0197, 0.0190, 0.0149], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 05:46:52,524 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6026, 2.5217, 3.0610, 4.0871, 3.6080, 4.0758, 3.3351, 2.5848], device='cuda:1'), covar=tensor([0.0257, 0.0797, 0.0565, 0.0055, 0.0239, 0.0053, 0.0481, 0.0800], device='cuda:1'), in_proj_covar=tensor([0.0125, 0.0150, 0.0168, 0.0104, 0.0122, 0.0101, 0.0171, 0.0151], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:46:59,676 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.919e+02 2.426e+02 2.838e+02 3.396e+02 4.576e+02, threshold=5.675e+02, percent-clipped=0.0 2023-04-17 05:47:02,274 INFO [train.py:893] (1/4) Epoch 26, batch 2450, loss[loss=0.1371, simple_loss=0.2082, pruned_loss=0.03304, over 13478.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2289, pruned_loss=0.05121, over 2667430.81 frames. ], batch size: 79, lr: 4.97e-03, grad_scale: 32.0 2023-04-17 05:47:03,900 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69929.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:47:03,951 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69929.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 05:47:21,944 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2365, 3.6097, 3.4248, 4.0398, 2.1428, 3.0737, 3.7666, 2.1723], device='cuda:1'), covar=tensor([0.0277, 0.0485, 0.0831, 0.0540, 0.1691, 0.0997, 0.0543, 0.1716], device='cuda:1'), in_proj_covar=tensor([0.0140, 0.0192, 0.0208, 0.0253, 0.0186, 0.0203, 0.0182, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:47:22,575 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69949.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 05:47:46,903 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=69976.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:47:48,340 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=69977.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:47:49,008 INFO [train.py:893] (1/4) Epoch 26, batch 2500, loss[loss=0.1782, simple_loss=0.24, pruned_loss=0.05818, over 13524.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2292, pruned_loss=0.05126, over 2666409.54 frames. ], batch size: 91, lr: 4.97e-03, grad_scale: 32.0 2023-04-17 05:48:35,298 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70024.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:48:35,351 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70024.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:48:35,949 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.846e+02 2.455e+02 2.831e+02 3.406e+02 8.435e+02, threshold=5.661e+02, percent-clipped=3.0 2023-04-17 05:48:38,171 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.87 vs. limit=2.0 2023-04-17 05:48:38,454 INFO [train.py:893] (1/4) Epoch 26, batch 2550, loss[loss=0.1856, simple_loss=0.2446, pruned_loss=0.0633, over 12037.00 frames. ], tot_loss[loss=0.166, simple_loss=0.229, pruned_loss=0.05145, over 2663211.38 frames. ], batch size: 158, lr: 4.97e-03, grad_scale: 32.0 2023-04-17 05:48:40,996 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70030.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 05:48:58,021 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70050.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:48:58,708 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-17 05:49:18,620 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70072.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:49:20,358 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3402, 2.5359, 2.0683, 4.1898, 4.6896, 3.5172, 4.6369, 4.4004], device='cuda:1'), covar=tensor([0.0086, 0.0899, 0.1094, 0.0093, 0.0065, 0.0456, 0.0070, 0.0076], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0094, 0.0102, 0.0086, 0.0073, 0.0083, 0.0060, 0.0082], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 05:49:23,310 INFO [train.py:893] (1/4) Epoch 26, batch 2600, loss[loss=0.1973, simple_loss=0.2532, pruned_loss=0.07067, over 13217.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2292, pruned_loss=0.05154, over 2666812.74 frames. ], batch size: 132, lr: 4.97e-03, grad_scale: 32.0 2023-04-17 05:49:41,772 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70098.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:49:51,383 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70110.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:50:02,223 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.895e+02 2.576e+02 3.040e+02 3.615e+02 6.720e+02, threshold=6.080e+02, percent-clipped=2.0 2023-04-17 05:50:04,354 INFO [train.py:893] (1/4) Epoch 26, batch 2650, loss[loss=0.1746, simple_loss=0.2408, pruned_loss=0.05416, over 13491.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.23, pruned_loss=0.05217, over 2667940.83 frames. ], batch size: 93, lr: 4.96e-03, grad_scale: 32.0 2023-04-17 05:50:37,276 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70171.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:51:00,945 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-17 05:51:10,718 INFO [train.py:893] (1/4) Epoch 27, batch 0, loss[loss=0.1943, simple_loss=0.2335, pruned_loss=0.07751, over 12550.00 frames. ], tot_loss[loss=0.1943, simple_loss=0.2335, pruned_loss=0.07751, over 12550.00 frames. ], batch size: 51, lr: 4.87e-03, grad_scale: 32.0 2023-04-17 05:51:10,719 INFO [train.py:918] (1/4) Computing validation loss 2023-04-17 05:51:17,046 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.3433, 2.1781, 3.7245, 3.6067, 3.5756, 3.0339, 3.3907, 2.9107], device='cuda:1'), covar=tensor([0.1705, 0.1200, 0.0167, 0.0231, 0.0229, 0.0684, 0.0306, 0.0731], device='cuda:1'), in_proj_covar=tensor([0.0220, 0.0182, 0.0131, 0.0136, 0.0138, 0.0177, 0.0151, 0.0187], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-17 05:51:17,995 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3320, 4.4267, 4.2295, 4.3457, 4.4261, 4.7267, 4.4962, 4.7216], device='cuda:1'), covar=tensor([0.0280, 0.0293, 0.0336, 0.0868, 0.0286, 0.0240, 0.0281, 0.0107], device='cuda:1'), in_proj_covar=tensor([0.0171, 0.0167, 0.0187, 0.0273, 0.0187, 0.0203, 0.0184, 0.0114], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 05:51:24,765 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4327, 4.8333, 4.6108, 4.6401, 4.7293, 4.5679, 4.8726, 4.9334], device='cuda:1'), covar=tensor([0.0230, 0.0202, 0.0195, 0.0306, 0.0278, 0.0250, 0.0254, 0.0193], device='cuda:1'), in_proj_covar=tensor([0.0152, 0.0219, 0.0182, 0.0196, 0.0167, 0.0219, 0.0145, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-04-17 05:51:29,353 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.6014, 3.8813, 3.8929, 4.3194, 2.6703, 3.6032, 4.1970, 2.5409], device='cuda:1'), covar=tensor([0.0200, 0.0481, 0.0649, 0.0419, 0.1345, 0.0831, 0.0405, 0.1591], device='cuda:1'), in_proj_covar=tensor([0.0142, 0.0194, 0.0211, 0.0257, 0.0188, 0.0205, 0.0184, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:51:33,435 INFO [train.py:927] (1/4) Epoch 27, validation: loss=0.1326, simple_loss=0.1923, pruned_loss=0.03639, over 2446609.00 frames. 2023-04-17 05:51:33,435 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12831MB 2023-04-17 05:52:16,509 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.830e+02 2.471e+02 2.835e+02 3.361e+02 5.660e+02, threshold=5.669e+02, percent-clipped=0.0 2023-04-17 05:52:18,130 INFO [train.py:893] (1/4) Epoch 27, batch 50, loss[loss=0.158, simple_loss=0.2163, pruned_loss=0.04985, over 13448.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2238, pruned_loss=0.05291, over 597904.13 frames. ], batch size: 65, lr: 4.87e-03, grad_scale: 32.0 2023-04-17 05:52:24,276 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.81 vs. limit=5.0 2023-04-17 05:52:41,785 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-17 05:52:41,786 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-17 05:52:41,786 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-17 05:52:41,801 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-17 05:52:41,809 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-17 05:52:42,549 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-17 05:52:42,568 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-17 05:52:49,333 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70260.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:52:54,237 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.8166, 2.6693, 2.5023, 1.6609, 1.6640, 2.2326, 2.2711, 2.9451], device='cuda:1'), covar=tensor([0.0959, 0.0368, 0.0510, 0.1651, 0.0251, 0.0603, 0.0851, 0.0220], device='cuda:1'), in_proj_covar=tensor([0.0176, 0.0153, 0.0128, 0.0213, 0.0116, 0.0171, 0.0183, 0.0140], device='cuda:1'), out_proj_covar=tensor([1.3005e-04, 1.1437e-04, 9.9110e-05, 1.5755e-04, 8.3575e-05, 1.2933e-04, 1.3790e-04, 1.0296e-04], device='cuda:1') 2023-04-17 05:53:05,178 INFO [train.py:893] (1/4) Epoch 27, batch 100, loss[loss=0.1656, simple_loss=0.2253, pruned_loss=0.05297, over 13109.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2264, pruned_loss=0.05335, over 1053757.76 frames. ], batch size: 142, lr: 4.87e-03, grad_scale: 32.0 2023-04-17 05:53:44,634 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=70321.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:53:48,117 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.920e+02 2.593e+02 3.180e+02 3.805e+02 8.227e+02, threshold=6.361e+02, percent-clipped=5.0 2023-04-17 05:53:50,471 INFO [train.py:893] (1/4) Epoch 27, batch 150, loss[loss=0.1787, simple_loss=0.2403, pruned_loss=0.05852, over 13556.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.2311, pruned_loss=0.0552, over 1411580.33 frames. ], batch size: 89, lr: 4.86e-03, grad_scale: 32.0 2023-04-17 05:53:53,203 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70330.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 05:54:07,340 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.48 vs. limit=5.0 2023-04-17 05:54:36,133 INFO [train.py:893] (1/4) Epoch 27, batch 200, loss[loss=0.1892, simple_loss=0.2354, pruned_loss=0.07149, over 13378.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.233, pruned_loss=0.05604, over 1683523.02 frames. ], batch size: 62, lr: 4.86e-03, grad_scale: 32.0 2023-04-17 05:54:37,147 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70378.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 05:54:39,625 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0939, 1.8884, 3.6911, 3.5456, 3.5040, 2.9587, 3.3755, 2.8593], device='cuda:1'), covar=tensor([0.1902, 0.1625, 0.0174, 0.0252, 0.0283, 0.0688, 0.0276, 0.0776], device='cuda:1'), in_proj_covar=tensor([0.0222, 0.0185, 0.0133, 0.0138, 0.0140, 0.0179, 0.0153, 0.0189], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-17 05:54:54,927 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3268, 4.7323, 4.5596, 4.5713, 4.5779, 4.3995, 4.8028, 4.8204], device='cuda:1'), covar=tensor([0.0206, 0.0208, 0.0192, 0.0300, 0.0314, 0.0263, 0.0257, 0.0212], device='cuda:1'), in_proj_covar=tensor([0.0150, 0.0217, 0.0180, 0.0194, 0.0165, 0.0216, 0.0143, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-04-17 05:55:07,123 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-04-17 05:55:19,534 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.767e+02 2.575e+02 3.029e+02 3.545e+02 5.639e+02, threshold=6.057e+02, percent-clipped=0.0 2023-04-17 05:55:21,191 INFO [train.py:893] (1/4) Epoch 27, batch 250, loss[loss=0.1611, simple_loss=0.2286, pruned_loss=0.04677, over 13534.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.233, pruned_loss=0.05589, over 1888328.38 frames. ], batch size: 85, lr: 4.86e-03, grad_scale: 32.0 2023-04-17 05:55:57,165 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70466.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:56:06,778 INFO [train.py:893] (1/4) Epoch 27, batch 300, loss[loss=0.1674, simple_loss=0.2407, pruned_loss=0.04699, over 13526.00 frames. ], tot_loss[loss=0.1725, simple_loss=0.2336, pruned_loss=0.05569, over 2052542.92 frames. ], batch size: 83, lr: 4.86e-03, grad_scale: 32.0 2023-04-17 05:56:35,017 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 2023-04-17 05:56:50,517 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.718e+02 2.582e+02 3.015e+02 3.678e+02 5.260e+02, threshold=6.030e+02, percent-clipped=0.0 2023-04-17 05:56:52,135 INFO [train.py:893] (1/4) Epoch 27, batch 350, loss[loss=0.2004, simple_loss=0.2618, pruned_loss=0.06952, over 13545.00 frames. ], tot_loss[loss=0.1719, simple_loss=0.2334, pruned_loss=0.05524, over 2192786.91 frames. ], batch size: 85, lr: 4.86e-03, grad_scale: 32.0 2023-04-17 05:57:02,442 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.77 vs. limit=2.0 2023-04-17 05:57:32,350 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7139, 3.5900, 3.7397, 2.2635, 4.0483, 3.8520, 3.8620, 4.0542], device='cuda:1'), covar=tensor([0.0316, 0.0198, 0.0182, 0.1384, 0.0200, 0.0309, 0.0182, 0.0135], device='cuda:1'), in_proj_covar=tensor([0.0098, 0.0061, 0.0087, 0.0106, 0.0104, 0.0116, 0.0087, 0.0082], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:57:38,642 INFO [train.py:893] (1/4) Epoch 27, batch 400, loss[loss=0.1495, simple_loss=0.2201, pruned_loss=0.03945, over 13464.00 frames. ], tot_loss[loss=0.1716, simple_loss=0.2333, pruned_loss=0.05497, over 2301086.16 frames. ], batch size: 79, lr: 4.85e-03, grad_scale: 32.0 2023-04-17 05:57:50,402 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2500, 3.6156, 3.4268, 3.9273, 2.1439, 3.0468, 3.7841, 2.1499], device='cuda:1'), covar=tensor([0.0149, 0.0440, 0.0815, 0.0485, 0.1733, 0.0959, 0.0491, 0.1774], device='cuda:1'), in_proj_covar=tensor([0.0142, 0.0194, 0.0210, 0.0255, 0.0188, 0.0204, 0.0184, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 05:58:01,269 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.1350, 2.1620, 3.8165, 3.6816, 3.6654, 3.0408, 3.4530, 2.9225], device='cuda:1'), covar=tensor([0.1856, 0.1317, 0.0178, 0.0252, 0.0245, 0.0655, 0.0281, 0.0763], device='cuda:1'), in_proj_covar=tensor([0.0220, 0.0183, 0.0132, 0.0137, 0.0139, 0.0178, 0.0150, 0.0187], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-17 05:58:15,008 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=70616.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 05:58:22,822 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.774e+02 2.559e+02 3.057e+02 3.646e+02 7.469e+02, threshold=6.114e+02, percent-clipped=3.0 2023-04-17 05:58:24,530 INFO [train.py:893] (1/4) Epoch 27, batch 450, loss[loss=0.1551, simple_loss=0.2123, pruned_loss=0.04893, over 13492.00 frames. ], tot_loss[loss=0.1724, simple_loss=0.2344, pruned_loss=0.05524, over 2383650.22 frames. ], batch size: 70, lr: 4.85e-03, grad_scale: 32.0 2023-04-17 05:58:50,665 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-17 05:58:57,010 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9204, 2.1585, 1.7219, 3.7906, 4.2099, 3.1591, 4.1804, 3.9698], device='cuda:1'), covar=tensor([0.0167, 0.1375, 0.1602, 0.0185, 0.0180, 0.0727, 0.0141, 0.0161], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0093, 0.0101, 0.0085, 0.0072, 0.0083, 0.0059, 0.0082], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 05:59:06,784 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7026, 2.4070, 2.4299, 2.8051, 2.0529, 2.8788, 2.7145, 2.2179], device='cuda:1'), covar=tensor([0.0086, 0.0234, 0.0176, 0.0173, 0.0261, 0.0139, 0.0203, 0.0238], device='cuda:1'), in_proj_covar=tensor([0.0113, 0.0126, 0.0131, 0.0131, 0.0140, 0.0118, 0.0115, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 05:59:09,753 INFO [train.py:893] (1/4) Epoch 27, batch 500, loss[loss=0.1711, simple_loss=0.239, pruned_loss=0.05158, over 13423.00 frames. ], tot_loss[loss=0.1726, simple_loss=0.2351, pruned_loss=0.05507, over 2448352.86 frames. ], batch size: 95, lr: 4.85e-03, grad_scale: 16.0 2023-04-17 05:59:14,418 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-04-17 05:59:29,260 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1710, 4.4184, 3.2878, 2.9868, 3.3558, 2.6689, 4.5525, 2.5714], device='cuda:1'), covar=tensor([0.1768, 0.0330, 0.1341, 0.2437, 0.0823, 0.3710, 0.0241, 0.4072], device='cuda:1'), in_proj_covar=tensor([0.0322, 0.0299, 0.0331, 0.0352, 0.0270, 0.0342, 0.0225, 0.0378], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 05:59:52,937 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.766e+02 2.540e+02 3.001e+02 3.649e+02 1.046e+03, threshold=6.001e+02, percent-clipped=2.0 2023-04-17 05:59:53,803 INFO [train.py:893] (1/4) Epoch 27, batch 550, loss[loss=0.1982, simple_loss=0.2577, pruned_loss=0.06942, over 11912.00 frames. ], tot_loss[loss=0.1722, simple_loss=0.2344, pruned_loss=0.05499, over 2491786.51 frames. ], batch size: 157, lr: 4.85e-03, grad_scale: 16.0 2023-04-17 06:00:28,938 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70766.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:00:38,860 INFO [train.py:893] (1/4) Epoch 27, batch 600, loss[loss=0.1602, simple_loss=0.2203, pruned_loss=0.0501, over 13531.00 frames. ], tot_loss[loss=0.1703, simple_loss=0.2324, pruned_loss=0.0541, over 2531496.63 frames. ], batch size: 85, lr: 4.85e-03, grad_scale: 16.0 2023-04-17 06:01:13,293 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70814.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:01:15,925 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2469, 3.5528, 3.3969, 3.9310, 2.2070, 3.0114, 3.7672, 2.1934], device='cuda:1'), covar=tensor([0.0140, 0.0467, 0.0762, 0.0470, 0.1637, 0.0994, 0.0480, 0.1711], device='cuda:1'), in_proj_covar=tensor([0.0142, 0.0194, 0.0210, 0.0256, 0.0189, 0.0205, 0.0184, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:01:23,609 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.781e+02 2.508e+02 2.864e+02 3.369e+02 6.579e+02, threshold=5.729e+02, percent-clipped=1.0 2023-04-17 06:01:24,451 INFO [train.py:893] (1/4) Epoch 27, batch 650, loss[loss=0.1638, simple_loss=0.2318, pruned_loss=0.04789, over 11987.00 frames. ], tot_loss[loss=0.1699, simple_loss=0.2318, pruned_loss=0.05396, over 2555692.14 frames. ], batch size: 157, lr: 4.85e-03, grad_scale: 16.0 2023-04-17 06:02:10,095 INFO [train.py:893] (1/4) Epoch 27, batch 700, loss[loss=0.2075, simple_loss=0.2728, pruned_loss=0.07109, over 13496.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2308, pruned_loss=0.05302, over 2575582.30 frames. ], batch size: 93, lr: 4.84e-03, grad_scale: 16.0 2023-04-17 06:02:22,653 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-04-17 06:02:37,666 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4680, 4.3987, 4.3402, 2.9100, 4.7834, 4.5765, 4.6037, 4.8341], device='cuda:1'), covar=tensor([0.0322, 0.0176, 0.0210, 0.1299, 0.0204, 0.0306, 0.0176, 0.0123], device='cuda:1'), in_proj_covar=tensor([0.0099, 0.0061, 0.0087, 0.0106, 0.0104, 0.0116, 0.0087, 0.0083], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:02:45,840 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=70916.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:02:55,323 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.829e+02 2.619e+02 3.031e+02 3.545e+02 7.685e+02, threshold=6.063e+02, percent-clipped=3.0 2023-04-17 06:02:55,347 INFO [train.py:893] (1/4) Epoch 27, batch 750, loss[loss=0.1752, simple_loss=0.2277, pruned_loss=0.06139, over 11981.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2304, pruned_loss=0.05319, over 2593710.13 frames. ], batch size: 157, lr: 4.84e-03, grad_scale: 8.0 2023-04-17 06:03:08,563 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3325, 2.9892, 2.8814, 3.3740, 2.7241, 3.4300, 3.3104, 2.9331], device='cuda:1'), covar=tensor([0.0126, 0.0149, 0.0141, 0.0158, 0.0189, 0.0122, 0.0192, 0.0142], device='cuda:1'), in_proj_covar=tensor([0.0113, 0.0126, 0.0132, 0.0131, 0.0141, 0.0119, 0.0115, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 06:03:27,945 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=70964.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:03:39,861 INFO [train.py:893] (1/4) Epoch 27, batch 800, loss[loss=0.1666, simple_loss=0.2319, pruned_loss=0.05059, over 13257.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2312, pruned_loss=0.05351, over 2610535.29 frames. ], batch size: 124, lr: 4.84e-03, grad_scale: 8.0 2023-04-17 06:03:46,519 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4884, 4.1895, 4.3447, 2.8679, 4.7529, 4.4977, 4.5787, 4.7174], device='cuda:1'), covar=tensor([0.0225, 0.0160, 0.0159, 0.0980, 0.0143, 0.0251, 0.0125, 0.0101], device='cuda:1'), in_proj_covar=tensor([0.0099, 0.0061, 0.0088, 0.0106, 0.0104, 0.0116, 0.0087, 0.0083], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:03:53,119 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=70992.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:03:54,008 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2541, 3.6123, 3.3147, 4.1933, 2.0830, 2.7496, 3.7051, 2.2256], device='cuda:1'), covar=tensor([0.0122, 0.0509, 0.0793, 0.0472, 0.1758, 0.1241, 0.0573, 0.1788], device='cuda:1'), in_proj_covar=tensor([0.0142, 0.0194, 0.0210, 0.0257, 0.0189, 0.0205, 0.0183, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:04:24,208 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.874e+02 2.677e+02 3.139e+02 3.587e+02 7.209e+02, threshold=6.278e+02, percent-clipped=2.0 2023-04-17 06:04:24,232 INFO [train.py:893] (1/4) Epoch 27, batch 850, loss[loss=0.182, simple_loss=0.2384, pruned_loss=0.06278, over 13355.00 frames. ], tot_loss[loss=0.1707, simple_loss=0.233, pruned_loss=0.05425, over 2625016.64 frames. ], batch size: 67, lr: 4.84e-03, grad_scale: 8.0 2023-04-17 06:04:27,667 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71030.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:04:48,526 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71053.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:04:53,676 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-04-17 06:05:09,351 INFO [train.py:893] (1/4) Epoch 27, batch 900, loss[loss=0.1613, simple_loss=0.2292, pruned_loss=0.04665, over 13442.00 frames. ], tot_loss[loss=0.1708, simple_loss=0.2329, pruned_loss=0.05441, over 2635241.58 frames. ], batch size: 103, lr: 4.84e-03, grad_scale: 8.0 2023-04-17 06:05:22,922 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71091.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:05:38,403 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-17 06:05:41,613 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9778, 3.9979, 4.0355, 2.3449, 4.4081, 4.1092, 4.1919, 4.4036], device='cuda:1'), covar=tensor([0.0376, 0.0188, 0.0184, 0.1432, 0.0221, 0.0350, 0.0196, 0.0144], device='cuda:1'), in_proj_covar=tensor([0.0099, 0.0061, 0.0087, 0.0105, 0.0104, 0.0116, 0.0086, 0.0082], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:05:49,405 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.42 vs. limit=2.0 2023-04-17 06:05:55,246 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.070e+02 2.531e+02 3.075e+02 3.695e+02 7.717e+02, threshold=6.149e+02, percent-clipped=1.0 2023-04-17 06:05:55,271 INFO [train.py:893] (1/4) Epoch 27, batch 950, loss[loss=0.1591, simple_loss=0.2218, pruned_loss=0.04822, over 13390.00 frames. ], tot_loss[loss=0.1704, simple_loss=0.2319, pruned_loss=0.05445, over 2644535.99 frames. ], batch size: 113, lr: 4.84e-03, grad_scale: 8.0 2023-04-17 06:06:41,762 INFO [train.py:893] (1/4) Epoch 27, batch 1000, loss[loss=0.1752, simple_loss=0.2377, pruned_loss=0.05631, over 13472.00 frames. ], tot_loss[loss=0.169, simple_loss=0.23, pruned_loss=0.05397, over 2647081.21 frames. ], batch size: 100, lr: 4.83e-03, grad_scale: 8.0 2023-04-17 06:07:22,297 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0444, 4.3488, 3.3058, 2.9687, 3.1422, 2.6385, 4.3943, 2.5750], device='cuda:1'), covar=tensor([0.1912, 0.0374, 0.1260, 0.2352, 0.0883, 0.3594, 0.0286, 0.4344], device='cuda:1'), in_proj_covar=tensor([0.0321, 0.0298, 0.0332, 0.0353, 0.0269, 0.0340, 0.0224, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 06:07:26,798 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.962e+02 2.549e+02 2.975e+02 3.461e+02 6.038e+02, threshold=5.950e+02, percent-clipped=0.0 2023-04-17 06:07:26,822 INFO [train.py:893] (1/4) Epoch 27, batch 1050, loss[loss=0.1568, simple_loss=0.2132, pruned_loss=0.05022, over 13545.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2287, pruned_loss=0.05284, over 2654076.99 frames. ], batch size: 72, lr: 4.83e-03, grad_scale: 8.0 2023-04-17 06:07:31,224 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9846, 4.1413, 2.9644, 3.7794, 4.0489, 2.8070, 3.6729, 2.9177], device='cuda:1'), covar=tensor([0.0295, 0.0233, 0.0921, 0.0352, 0.0234, 0.1146, 0.0477, 0.1083], device='cuda:1'), in_proj_covar=tensor([0.0167, 0.0185, 0.0179, 0.0233, 0.0143, 0.0164, 0.0164, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:08:05,440 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7530, 2.3243, 2.4386, 2.8195, 2.1051, 2.8758, 2.7565, 2.2783], device='cuda:1'), covar=tensor([0.0080, 0.0222, 0.0163, 0.0156, 0.0243, 0.0144, 0.0185, 0.0218], device='cuda:1'), in_proj_covar=tensor([0.0113, 0.0127, 0.0133, 0.0133, 0.0142, 0.0120, 0.0115, 0.0119], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 06:08:12,363 INFO [train.py:893] (1/4) Epoch 27, batch 1100, loss[loss=0.1546, simple_loss=0.2212, pruned_loss=0.04406, over 13542.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2289, pruned_loss=0.05249, over 2654032.68 frames. ], batch size: 72, lr: 4.83e-03, grad_scale: 8.0 2023-04-17 06:08:16,171 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.46 vs. limit=2.0 2023-04-17 06:08:44,892 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8405, 3.9242, 2.8218, 3.5316, 3.8506, 2.4540, 3.4894, 2.4840], device='cuda:1'), covar=tensor([0.0320, 0.0202, 0.1050, 0.0410, 0.0289, 0.1412, 0.0551, 0.1387], device='cuda:1'), in_proj_covar=tensor([0.0166, 0.0183, 0.0178, 0.0231, 0.0142, 0.0163, 0.0163, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:08:57,474 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.923e+02 2.567e+02 2.971e+02 3.412e+02 6.882e+02, threshold=5.943e+02, percent-clipped=1.0 2023-04-17 06:08:57,499 INFO [train.py:893] (1/4) Epoch 27, batch 1150, loss[loss=0.2023, simple_loss=0.2597, pruned_loss=0.07247, over 13517.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2287, pruned_loss=0.05202, over 2657598.10 frames. ], batch size: 85, lr: 4.83e-03, grad_scale: 8.0 2023-04-17 06:09:16,407 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71348.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:09:16,661 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-17 06:09:35,973 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5147, 4.8363, 4.5071, 4.5563, 4.6013, 4.9397, 4.6903, 4.6533], device='cuda:1'), covar=tensor([0.0304, 0.0282, 0.0336, 0.0812, 0.0290, 0.0231, 0.0289, 0.0193], device='cuda:1'), in_proj_covar=tensor([0.0171, 0.0167, 0.0188, 0.0273, 0.0186, 0.0204, 0.0185, 0.0115], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 06:09:42,215 INFO [train.py:893] (1/4) Epoch 27, batch 1200, loss[loss=0.1465, simple_loss=0.2012, pruned_loss=0.04583, over 13158.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.2288, pruned_loss=0.05207, over 2651395.64 frames. ], batch size: 58, lr: 4.83e-03, grad_scale: 8.0 2023-04-17 06:09:50,476 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71386.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:10:09,333 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-17 06:10:20,562 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-17 06:10:25,079 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9101, 3.9664, 2.9103, 3.6615, 3.8729, 2.5140, 3.5933, 2.5719], device='cuda:1'), covar=tensor([0.0325, 0.0224, 0.0998, 0.0456, 0.0297, 0.1357, 0.0479, 0.1358], device='cuda:1'), in_proj_covar=tensor([0.0167, 0.0185, 0.0180, 0.0233, 0.0143, 0.0164, 0.0164, 0.0163], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:10:28,585 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.042e+02 2.662e+02 2.989e+02 3.476e+02 6.704e+02, threshold=5.979e+02, percent-clipped=2.0 2023-04-17 06:10:28,609 INFO [train.py:893] (1/4) Epoch 27, batch 1250, loss[loss=0.1515, simple_loss=0.2065, pruned_loss=0.04824, over 12793.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2297, pruned_loss=0.05286, over 2653086.67 frames. ], batch size: 52, lr: 4.83e-03, grad_scale: 8.0 2023-04-17 06:11:12,982 INFO [train.py:893] (1/4) Epoch 27, batch 1300, loss[loss=0.1839, simple_loss=0.2457, pruned_loss=0.06103, over 13360.00 frames. ], tot_loss[loss=0.1684, simple_loss=0.2307, pruned_loss=0.05311, over 2654717.10 frames. ], batch size: 62, lr: 4.82e-03, grad_scale: 8.0 2023-04-17 06:11:58,684 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.824e+02 2.626e+02 3.134e+02 3.874e+02 8.782e+02, threshold=6.268e+02, percent-clipped=3.0 2023-04-17 06:11:58,707 INFO [train.py:893] (1/4) Epoch 27, batch 1350, loss[loss=0.1768, simple_loss=0.2354, pruned_loss=0.05911, over 13570.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.2312, pruned_loss=0.05329, over 2658185.65 frames. ], batch size: 89, lr: 4.82e-03, grad_scale: 8.0 2023-04-17 06:12:43,420 INFO [train.py:893] (1/4) Epoch 27, batch 1400, loss[loss=0.16, simple_loss=0.2258, pruned_loss=0.04707, over 13522.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2306, pruned_loss=0.05315, over 2658921.59 frames. ], batch size: 98, lr: 4.82e-03, grad_scale: 8.0 2023-04-17 06:12:45,518 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5333, 3.9180, 3.6934, 4.2504, 2.4627, 3.4202, 4.1274, 2.4949], device='cuda:1'), covar=tensor([0.0165, 0.0417, 0.0730, 0.0467, 0.1560, 0.0914, 0.0411, 0.1504], device='cuda:1'), in_proj_covar=tensor([0.0141, 0.0192, 0.0210, 0.0255, 0.0187, 0.0203, 0.0182, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:13:05,440 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=5.15 vs. limit=5.0 2023-04-17 06:13:20,754 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71616.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:13:30,174 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.758e+02 2.491e+02 2.947e+02 3.481e+02 7.529e+02, threshold=5.893e+02, percent-clipped=1.0 2023-04-17 06:13:30,198 INFO [train.py:893] (1/4) Epoch 27, batch 1450, loss[loss=0.1627, simple_loss=0.2303, pruned_loss=0.04757, over 13472.00 frames. ], tot_loss[loss=0.1687, simple_loss=0.231, pruned_loss=0.05323, over 2663146.98 frames. ], batch size: 100, lr: 4.82e-03, grad_scale: 8.0 2023-04-17 06:13:49,947 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71648.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:14:14,954 INFO [train.py:893] (1/4) Epoch 27, batch 1500, loss[loss=0.1701, simple_loss=0.2239, pruned_loss=0.05817, over 13374.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2308, pruned_loss=0.0527, over 2667581.34 frames. ], batch size: 67, lr: 4.82e-03, grad_scale: 8.0 2023-04-17 06:14:15,314 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71677.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:14:23,944 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=71686.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:14:32,755 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=71696.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:14:32,850 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4937, 4.7186, 4.4476, 4.5778, 4.5917, 4.9022, 4.6564, 4.5918], device='cuda:1'), covar=tensor([0.0387, 0.0397, 0.0441, 0.0779, 0.0364, 0.0271, 0.0388, 0.0279], device='cuda:1'), in_proj_covar=tensor([0.0171, 0.0166, 0.0187, 0.0271, 0.0185, 0.0203, 0.0184, 0.0114], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 06:14:36,599 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-04-17 06:15:00,682 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.808e+02 2.588e+02 3.144e+02 3.766e+02 5.385e+02, threshold=6.289e+02, percent-clipped=0.0 2023-04-17 06:15:00,706 INFO [train.py:893] (1/4) Epoch 27, batch 1550, loss[loss=0.1519, simple_loss=0.2169, pruned_loss=0.04349, over 13374.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2313, pruned_loss=0.05297, over 2665316.41 frames. ], batch size: 77, lr: 4.82e-03, grad_scale: 8.0 2023-04-17 06:15:06,435 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=71734.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:15:23,577 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7315, 3.4628, 2.8010, 3.0435, 2.8819, 2.2047, 3.5790, 2.0684], device='cuda:1'), covar=tensor([0.0666, 0.0613, 0.0522, 0.0502, 0.0678, 0.1896, 0.0979, 0.1317], device='cuda:1'), in_proj_covar=tensor([0.0147, 0.0148, 0.0138, 0.0119, 0.0150, 0.0192, 0.0186, 0.0145], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 06:15:44,126 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-04-17 06:15:46,374 INFO [train.py:893] (1/4) Epoch 27, batch 1600, loss[loss=0.1447, simple_loss=0.2008, pruned_loss=0.04426, over 12524.00 frames. ], tot_loss[loss=0.1693, simple_loss=0.2319, pruned_loss=0.05331, over 2663124.88 frames. ], batch size: 51, lr: 4.81e-03, grad_scale: 8.0 2023-04-17 06:16:06,221 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9172, 4.3233, 4.1323, 4.1586, 4.1854, 3.9448, 4.3711, 4.3921], device='cuda:1'), covar=tensor([0.0254, 0.0272, 0.0259, 0.0367, 0.0317, 0.0313, 0.0275, 0.0288], device='cuda:1'), in_proj_covar=tensor([0.0148, 0.0216, 0.0179, 0.0193, 0.0164, 0.0216, 0.0142, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-04-17 06:16:31,537 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.806e+02 2.588e+02 3.045e+02 3.762e+02 5.857e+02, threshold=6.090e+02, percent-clipped=0.0 2023-04-17 06:16:31,561 INFO [train.py:893] (1/4) Epoch 27, batch 1650, loss[loss=0.138, simple_loss=0.2114, pruned_loss=0.03227, over 13487.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2331, pruned_loss=0.05321, over 2663905.03 frames. ], batch size: 81, lr: 4.81e-03, grad_scale: 8.0 2023-04-17 06:16:59,351 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4033, 3.1712, 3.8341, 2.6802, 2.4155, 2.5732, 4.1611, 4.2444], device='cuda:1'), covar=tensor([0.1246, 0.1848, 0.0370, 0.1957, 0.1853, 0.1797, 0.0254, 0.0227], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0282, 0.0204, 0.0229, 0.0226, 0.0190, 0.0222, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:17:16,695 INFO [train.py:893] (1/4) Epoch 27, batch 1700, loss[loss=0.167, simple_loss=0.232, pruned_loss=0.05097, over 13487.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2335, pruned_loss=0.05305, over 2663551.41 frames. ], batch size: 93, lr: 4.81e-03, grad_scale: 8.0 2023-04-17 06:17:27,017 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7310, 2.8722, 3.1877, 4.2725, 3.7666, 4.3119, 3.5753, 2.9734], device='cuda:1'), covar=tensor([0.0270, 0.0768, 0.0682, 0.0061, 0.0234, 0.0059, 0.0530, 0.0728], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0152, 0.0171, 0.0107, 0.0125, 0.0103, 0.0174, 0.0152], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:17:54,141 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=71919.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:17:56,555 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1440, 4.4303, 4.2118, 4.2462, 4.2989, 4.5998, 4.3775, 4.2484], device='cuda:1'), covar=tensor([0.0303, 0.0268, 0.0294, 0.0853, 0.0265, 0.0248, 0.0286, 0.0287], device='cuda:1'), in_proj_covar=tensor([0.0172, 0.0166, 0.0188, 0.0274, 0.0187, 0.0204, 0.0185, 0.0115], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 06:18:02,450 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.742e+02 2.543e+02 2.961e+02 3.560e+02 7.883e+02, threshold=5.923e+02, percent-clipped=1.0 2023-04-17 06:18:02,474 INFO [train.py:893] (1/4) Epoch 27, batch 1750, loss[loss=0.159, simple_loss=0.2213, pruned_loss=0.04839, over 13467.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2317, pruned_loss=0.05215, over 2660759.00 frames. ], batch size: 106, lr: 4.81e-03, grad_scale: 8.0 2023-04-17 06:18:32,992 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4541, 3.0961, 3.8124, 2.7469, 2.6108, 2.7121, 4.2300, 4.2758], device='cuda:1'), covar=tensor([0.1245, 0.2227, 0.0430, 0.1957, 0.1703, 0.1718, 0.0299, 0.0245], device='cuda:1'), in_proj_covar=tensor([0.0255, 0.0281, 0.0205, 0.0228, 0.0225, 0.0189, 0.0221, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:18:43,471 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=71972.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:18:47,442 INFO [train.py:893] (1/4) Epoch 27, batch 1800, loss[loss=0.1493, simple_loss=0.2186, pruned_loss=0.04, over 13532.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2304, pruned_loss=0.05132, over 2662308.07 frames. ], batch size: 83, lr: 4.81e-03, grad_scale: 8.0 2023-04-17 06:18:50,209 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=71980.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:19:37,156 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.833e+02 2.380e+02 2.800e+02 3.278e+02 8.080e+02, threshold=5.600e+02, percent-clipped=1.0 2023-04-17 06:19:37,180 INFO [train.py:893] (1/4) Epoch 27, batch 1850, loss[loss=0.1458, simple_loss=0.2109, pruned_loss=0.04039, over 13531.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2295, pruned_loss=0.05095, over 2660378.42 frames. ], batch size: 83, lr: 4.81e-03, grad_scale: 8.0 2023-04-17 06:19:40,592 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-17 06:19:58,506 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9181, 2.5888, 2.5833, 3.0784, 2.3524, 3.0412, 3.0090, 2.5429], device='cuda:1'), covar=tensor([0.0112, 0.0237, 0.0203, 0.0173, 0.0252, 0.0145, 0.0177, 0.0199], device='cuda:1'), in_proj_covar=tensor([0.0114, 0.0128, 0.0133, 0.0133, 0.0143, 0.0119, 0.0116, 0.0118], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 06:20:13,063 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5656, 4.7733, 4.5671, 4.5421, 4.6489, 4.9475, 4.7564, 4.6407], device='cuda:1'), covar=tensor([0.0391, 0.0313, 0.0465, 0.0929, 0.0373, 0.0256, 0.0468, 0.0234], device='cuda:1'), in_proj_covar=tensor([0.0172, 0.0166, 0.0189, 0.0273, 0.0187, 0.0203, 0.0184, 0.0115], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 06:20:21,890 INFO [train.py:893] (1/4) Epoch 27, batch 1900, loss[loss=0.1802, simple_loss=0.2395, pruned_loss=0.06042, over 13181.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2287, pruned_loss=0.05085, over 2661882.43 frames. ], batch size: 132, lr: 4.80e-03, grad_scale: 8.0 2023-04-17 06:20:47,087 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8323, 3.5717, 2.8951, 3.2670, 2.9650, 2.1785, 3.5964, 2.0589], device='cuda:1'), covar=tensor([0.0690, 0.0637, 0.0498, 0.0421, 0.0676, 0.2014, 0.1065, 0.1336], device='cuda:1'), in_proj_covar=tensor([0.0148, 0.0149, 0.0140, 0.0120, 0.0151, 0.0194, 0.0188, 0.0148], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 06:21:07,576 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2023-04-17 06:21:07,740 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.840e+02 2.498e+02 3.028e+02 3.455e+02 6.047e+02, threshold=6.055e+02, percent-clipped=2.0 2023-04-17 06:21:07,764 INFO [train.py:893] (1/4) Epoch 27, batch 1950, loss[loss=0.1633, simple_loss=0.2276, pruned_loss=0.04956, over 13512.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2283, pruned_loss=0.05065, over 2665687.38 frames. ], batch size: 83, lr: 4.80e-03, grad_scale: 8.0 2023-04-17 06:21:10,405 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72129.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:21:10,416 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.2783, 3.2248, 2.7718, 2.1330, 2.3842, 2.8171, 2.8851, 3.3962], device='cuda:1'), covar=tensor([0.0890, 0.0274, 0.0546, 0.1277, 0.0489, 0.0458, 0.0570, 0.0214], device='cuda:1'), in_proj_covar=tensor([0.0177, 0.0153, 0.0130, 0.0214, 0.0118, 0.0171, 0.0182, 0.0140], device='cuda:1'), out_proj_covar=tensor([1.3066e-04, 1.1388e-04, 9.9944e-05, 1.5801e-04, 8.4852e-05, 1.2948e-04, 1.3696e-04, 1.0267e-04], device='cuda:1') 2023-04-17 06:21:50,730 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.56 vs. limit=2.0 2023-04-17 06:21:53,868 INFO [train.py:893] (1/4) Epoch 27, batch 2000, loss[loss=0.1841, simple_loss=0.2557, pruned_loss=0.05627, over 13386.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2295, pruned_loss=0.05137, over 2663266.83 frames. ], batch size: 109, lr: 4.80e-03, grad_scale: 8.0 2023-04-17 06:22:00,306 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-17 06:22:05,643 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72190.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:22:30,585 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2716, 2.6483, 2.2179, 4.2170, 4.6621, 3.5201, 4.5719, 4.3760], device='cuda:1'), covar=tensor([0.0098, 0.0872, 0.0999, 0.0091, 0.0070, 0.0413, 0.0070, 0.0084], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0093, 0.0101, 0.0085, 0.0073, 0.0083, 0.0059, 0.0083], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 06:22:30,730 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=2.54 vs. limit=5.0 2023-04-17 06:22:36,158 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1502, 2.6104, 2.1840, 4.1114, 4.5069, 3.4054, 4.4410, 4.2310], device='cuda:1'), covar=tensor([0.0089, 0.0860, 0.0940, 0.0084, 0.0065, 0.0447, 0.0063, 0.0083], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0093, 0.0101, 0.0085, 0.0073, 0.0083, 0.0059, 0.0083], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 06:22:39,173 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.946e+02 2.675e+02 3.139e+02 3.825e+02 7.364e+02, threshold=6.278e+02, percent-clipped=1.0 2023-04-17 06:22:39,198 INFO [train.py:893] (1/4) Epoch 27, batch 2050, loss[loss=0.1919, simple_loss=0.2528, pruned_loss=0.06547, over 13465.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2306, pruned_loss=0.05222, over 2664696.57 frames. ], batch size: 100, lr: 4.80e-03, grad_scale: 8.0 2023-04-17 06:22:41,074 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.1009, 2.1460, 3.8388, 3.7188, 3.6898, 2.9480, 3.4415, 2.9004], device='cuda:1'), covar=tensor([0.1980, 0.1344, 0.0185, 0.0199, 0.0224, 0.0723, 0.0289, 0.0780], device='cuda:1'), in_proj_covar=tensor([0.0219, 0.0184, 0.0132, 0.0137, 0.0140, 0.0177, 0.0152, 0.0187], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-17 06:22:56,285 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4286, 2.7722, 2.3084, 4.3957, 4.7609, 3.6882, 4.6746, 4.4770], device='cuda:1'), covar=tensor([0.0084, 0.0840, 0.0961, 0.0084, 0.0081, 0.0405, 0.0071, 0.0079], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0093, 0.0101, 0.0085, 0.0073, 0.0083, 0.0059, 0.0083], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 06:23:19,849 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72272.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:23:22,923 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72275.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:23:24,346 INFO [train.py:893] (1/4) Epoch 27, batch 2100, loss[loss=0.1718, simple_loss=0.2337, pruned_loss=0.05493, over 13539.00 frames. ], tot_loss[loss=0.1682, simple_loss=0.2312, pruned_loss=0.05258, over 2662637.33 frames. ], batch size: 85, lr: 4.80e-03, grad_scale: 8.0 2023-04-17 06:24:03,251 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72320.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:24:08,368 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72326.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:24:08,861 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.635e+02 2.231e+02 2.747e+02 3.271e+02 6.169e+02, threshold=5.495e+02, percent-clipped=0.0 2023-04-17 06:24:08,885 INFO [train.py:893] (1/4) Epoch 27, batch 2150, loss[loss=0.1581, simple_loss=0.2239, pruned_loss=0.04614, over 13096.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2313, pruned_loss=0.05222, over 2664680.75 frames. ], batch size: 142, lr: 4.80e-03, grad_scale: 8.0 2023-04-17 06:24:26,041 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.3297, 4.7981, 4.7104, 4.7795, 4.6084, 4.6472, 5.2690, 4.8596], device='cuda:1'), covar=tensor([0.0633, 0.1262, 0.1939, 0.2411, 0.0945, 0.1710, 0.0821, 0.1152], device='cuda:1'), in_proj_covar=tensor([0.0351, 0.0404, 0.0493, 0.0494, 0.0315, 0.0372, 0.0460, 0.0370], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 06:24:30,178 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.6367, 2.4709, 2.0649, 1.4228, 1.8246, 1.9726, 2.1644, 2.5776], device='cuda:1'), covar=tensor([0.0898, 0.0287, 0.0640, 0.1563, 0.0181, 0.0574, 0.0689, 0.0258], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0156, 0.0132, 0.0217, 0.0119, 0.0174, 0.0185, 0.0143], device='cuda:1'), out_proj_covar=tensor([1.3337e-04, 1.1641e-04, 1.0192e-04, 1.6035e-04, 8.5514e-05, 1.3192e-04, 1.3907e-04, 1.0515e-04], device='cuda:1') 2023-04-17 06:24:33,508 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-04-17 06:24:49,787 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.55 vs. limit=5.0 2023-04-17 06:24:54,766 INFO [train.py:893] (1/4) Epoch 27, batch 2200, loss[loss=0.1373, simple_loss=0.2008, pruned_loss=0.03694, over 13539.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2299, pruned_loss=0.05133, over 2663871.63 frames. ], batch size: 72, lr: 4.79e-03, grad_scale: 8.0 2023-04-17 06:25:03,433 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72387.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:25:21,377 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0924, 2.0275, 3.7450, 3.6275, 3.6067, 2.9495, 3.3721, 2.8878], device='cuda:1'), covar=tensor([0.2164, 0.1480, 0.0182, 0.0201, 0.0240, 0.0720, 0.0307, 0.0774], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0185, 0.0134, 0.0137, 0.0139, 0.0177, 0.0153, 0.0189], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-17 06:25:24,476 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.3108, 5.1272, 5.4003, 5.1133, 5.6470, 5.1877, 5.6513, 5.6059], device='cuda:1'), covar=tensor([0.0438, 0.0593, 0.0561, 0.0553, 0.0493, 0.0745, 0.0421, 0.0430], device='cuda:1'), in_proj_covar=tensor([0.0311, 0.0310, 0.0322, 0.0241, 0.0460, 0.0361, 0.0302, 0.0318], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:25:39,703 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.840e+02 2.441e+02 2.891e+02 3.478e+02 5.512e+02, threshold=5.782e+02, percent-clipped=2.0 2023-04-17 06:25:39,728 INFO [train.py:893] (1/4) Epoch 27, batch 2250, loss[loss=0.1371, simple_loss=0.2052, pruned_loss=0.03447, over 13568.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.228, pruned_loss=0.05021, over 2665142.95 frames. ], batch size: 78, lr: 4.79e-03, grad_scale: 8.0 2023-04-17 06:26:25,344 INFO [train.py:893] (1/4) Epoch 27, batch 2300, loss[loss=0.1956, simple_loss=0.2531, pruned_loss=0.0691, over 13536.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2283, pruned_loss=0.05055, over 2666752.16 frames. ], batch size: 87, lr: 4.79e-03, grad_scale: 8.0 2023-04-17 06:26:32,103 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72485.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:26:48,316 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0695, 3.9549, 4.0555, 2.3329, 4.2773, 4.1357, 4.0590, 4.2845], device='cuda:1'), covar=tensor([0.0270, 0.0165, 0.0141, 0.1286, 0.0160, 0.0241, 0.0164, 0.0118], device='cuda:1'), in_proj_covar=tensor([0.0100, 0.0062, 0.0089, 0.0107, 0.0104, 0.0117, 0.0088, 0.0084], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:26:58,880 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0713, 2.5006, 2.0550, 4.0382, 4.4048, 3.2868, 4.3495, 4.1302], device='cuda:1'), covar=tensor([0.0083, 0.0953, 0.1013, 0.0086, 0.0067, 0.0465, 0.0075, 0.0086], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0092, 0.0100, 0.0085, 0.0072, 0.0082, 0.0059, 0.0082], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 06:27:04,522 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8603, 4.3040, 4.0969, 4.0972, 4.2026, 3.9719, 4.3819, 4.4453], device='cuda:1'), covar=tensor([0.0278, 0.0257, 0.0258, 0.0345, 0.0266, 0.0273, 0.0283, 0.0241], device='cuda:1'), in_proj_covar=tensor([0.0149, 0.0215, 0.0180, 0.0193, 0.0164, 0.0215, 0.0144, 0.0195], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-04-17 06:27:11,482 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.738e+02 2.357e+02 2.755e+02 3.330e+02 4.557e+02, threshold=5.510e+02, percent-clipped=0.0 2023-04-17 06:27:11,507 INFO [train.py:893] (1/4) Epoch 27, batch 2350, loss[loss=0.1529, simple_loss=0.2218, pruned_loss=0.04198, over 13244.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2281, pruned_loss=0.05052, over 2666065.05 frames. ], batch size: 117, lr: 4.79e-03, grad_scale: 8.0 2023-04-17 06:27:32,824 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-17 06:27:44,270 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4250, 2.7381, 2.3251, 4.3810, 4.8013, 3.6715, 4.7377, 4.4466], device='cuda:1'), covar=tensor([0.0094, 0.0915, 0.1018, 0.0098, 0.0086, 0.0427, 0.0076, 0.0095], device='cuda:1'), in_proj_covar=tensor([0.0079, 0.0093, 0.0101, 0.0085, 0.0072, 0.0082, 0.0059, 0.0082], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 06:27:54,814 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72575.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:27:56,315 INFO [train.py:893] (1/4) Epoch 27, batch 2400, loss[loss=0.1581, simple_loss=0.2221, pruned_loss=0.04706, over 13525.00 frames. ], tot_loss[loss=0.1631, simple_loss=0.2263, pruned_loss=0.04996, over 2665075.23 frames. ], batch size: 87, lr: 4.79e-03, grad_scale: 8.0 2023-04-17 06:28:39,772 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72623.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:28:42,929 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.028e+02 2.421e+02 2.989e+02 3.536e+02 6.414e+02, threshold=5.978e+02, percent-clipped=1.0 2023-04-17 06:28:42,953 INFO [train.py:893] (1/4) Epoch 27, batch 2450, loss[loss=0.1634, simple_loss=0.2309, pruned_loss=0.04798, over 13398.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2272, pruned_loss=0.05031, over 2662016.70 frames. ], batch size: 113, lr: 4.79e-03, grad_scale: 8.0 2023-04-17 06:29:27,624 INFO [train.py:893] (1/4) Epoch 27, batch 2500, loss[loss=0.1887, simple_loss=0.2379, pruned_loss=0.06974, over 13541.00 frames. ], tot_loss[loss=0.1629, simple_loss=0.2262, pruned_loss=0.04978, over 2664779.05 frames. ], batch size: 72, lr: 4.78e-03, grad_scale: 8.0 2023-04-17 06:29:33,448 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=72682.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:30:02,265 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9192, 3.7540, 3.0072, 3.4458, 3.0961, 2.3751, 3.8100, 2.1881], device='cuda:1'), covar=tensor([0.0695, 0.0559, 0.0538, 0.0380, 0.0671, 0.1830, 0.0966, 0.1349], device='cuda:1'), in_proj_covar=tensor([0.0149, 0.0150, 0.0141, 0.0121, 0.0152, 0.0195, 0.0189, 0.0149], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 06:30:15,076 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.688e+02 2.583e+02 2.896e+02 3.437e+02 6.200e+02, threshold=5.793e+02, percent-clipped=1.0 2023-04-17 06:30:15,100 INFO [train.py:893] (1/4) Epoch 27, batch 2550, loss[loss=0.1415, simple_loss=0.2111, pruned_loss=0.03602, over 13466.00 frames. ], tot_loss[loss=0.1631, simple_loss=0.2266, pruned_loss=0.04986, over 2663443.35 frames. ], batch size: 79, lr: 4.78e-03, grad_scale: 8.0 2023-04-17 06:30:38,924 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-17 06:30:57,966 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2242, 2.6782, 2.0595, 4.1297, 4.5346, 3.3101, 4.4998, 4.2418], device='cuda:1'), covar=tensor([0.0094, 0.0879, 0.1102, 0.0096, 0.0083, 0.0511, 0.0070, 0.0088], device='cuda:1'), in_proj_covar=tensor([0.0080, 0.0093, 0.0101, 0.0086, 0.0073, 0.0083, 0.0060, 0.0083], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 06:31:00,068 INFO [train.py:893] (1/4) Epoch 27, batch 2600, loss[loss=0.1782, simple_loss=0.2333, pruned_loss=0.06155, over 12069.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2269, pruned_loss=0.05039, over 2662106.15 frames. ], batch size: 157, lr: 4.78e-03, grad_scale: 8.0 2023-04-17 06:31:07,058 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72785.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:31:27,588 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.58 vs. limit=2.0 2023-04-17 06:31:34,660 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3609, 3.7065, 3.5692, 4.0898, 2.2331, 3.1495, 3.9065, 2.2985], device='cuda:1'), covar=tensor([0.0153, 0.0555, 0.0774, 0.0481, 0.1728, 0.0977, 0.0479, 0.1698], device='cuda:1'), in_proj_covar=tensor([0.0140, 0.0190, 0.0208, 0.0253, 0.0186, 0.0203, 0.0181, 0.0177], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:31:34,749 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-04-17 06:31:40,334 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.031e+02 2.584e+02 3.071e+02 3.664e+02 5.210e+02, threshold=6.141e+02, percent-clipped=0.0 2023-04-17 06:31:40,358 INFO [train.py:893] (1/4) Epoch 27, batch 2650, loss[loss=0.1798, simple_loss=0.2396, pruned_loss=0.06005, over 13491.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2277, pruned_loss=0.05099, over 2660265.01 frames. ], batch size: 93, lr: 4.78e-03, grad_scale: 8.0 2023-04-17 06:31:45,639 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=72833.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:32:37,800 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-17 06:32:47,349 INFO [train.py:893] (1/4) Epoch 28, batch 0, loss[loss=0.1619, simple_loss=0.2244, pruned_loss=0.0497, over 13533.00 frames. ], tot_loss[loss=0.1619, simple_loss=0.2244, pruned_loss=0.0497, over 13533.00 frames. ], batch size: 87, lr: 4.69e-03, grad_scale: 8.0 2023-04-17 06:32:47,350 INFO [train.py:918] (1/4) Computing validation loss 2023-04-17 06:32:58,941 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.7830, 4.7212, 4.8042, 4.7499, 5.0581, 4.6681, 5.0376, 5.0434], device='cuda:1'), covar=tensor([0.0386, 0.0568, 0.0613, 0.0493, 0.0495, 0.0747, 0.0431, 0.0387], device='cuda:1'), in_proj_covar=tensor([0.0313, 0.0311, 0.0323, 0.0242, 0.0462, 0.0363, 0.0303, 0.0319], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:33:09,317 INFO [train.py:927] (1/4) Epoch 28, validation: loss=0.1317, simple_loss=0.1914, pruned_loss=0.03593, over 2446609.00 frames. 2023-04-17 06:33:09,318 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12831MB 2023-04-17 06:33:15,549 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.44 vs. limit=5.0 2023-04-17 06:33:27,827 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72894.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:33:56,890 INFO [train.py:893] (1/4) Epoch 28, batch 50, loss[loss=0.1688, simple_loss=0.2278, pruned_loss=0.05491, over 13516.00 frames. ], tot_loss[loss=0.1615, simple_loss=0.2222, pruned_loss=0.05044, over 603489.72 frames. ], batch size: 83, lr: 4.69e-03, grad_scale: 16.0 2023-04-17 06:33:57,590 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-04-17 06:33:57,751 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.022e+02 2.571e+02 2.970e+02 3.456e+02 5.243e+02, threshold=5.941e+02, percent-clipped=0.0 2023-04-17 06:34:19,634 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-17 06:34:19,634 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-17 06:34:19,634 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-17 06:34:19,650 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-17 06:34:19,658 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-17 06:34:19,677 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-17 06:34:19,689 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-17 06:34:22,241 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=72955.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:34:30,510 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=72964.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 06:34:40,021 INFO [train.py:893] (1/4) Epoch 28, batch 100, loss[loss=0.1658, simple_loss=0.2271, pruned_loss=0.05227, over 13514.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.225, pruned_loss=0.05229, over 1059742.08 frames. ], batch size: 70, lr: 4.69e-03, grad_scale: 16.0 2023-04-17 06:34:46,037 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=72982.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:34:50,205 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0757, 4.0906, 2.9774, 3.7811, 4.0275, 2.7282, 3.6534, 2.7719], device='cuda:1'), covar=tensor([0.0297, 0.0240, 0.1034, 0.0392, 0.0252, 0.1214, 0.0541, 0.1227], device='cuda:1'), in_proj_covar=tensor([0.0169, 0.0188, 0.0182, 0.0238, 0.0144, 0.0164, 0.0166, 0.0163], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:35:14,233 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-04-17 06:35:27,357 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73025.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 06:35:27,878 INFO [train.py:893] (1/4) Epoch 28, batch 150, loss[loss=0.1698, simple_loss=0.2336, pruned_loss=0.05304, over 13444.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2285, pruned_loss=0.05403, over 1409099.75 frames. ], batch size: 103, lr: 4.69e-03, grad_scale: 16.0 2023-04-17 06:35:28,682 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.685e+02 2.652e+02 3.088e+02 3.819e+02 5.892e+02, threshold=6.175e+02, percent-clipped=0.0 2023-04-17 06:35:31,397 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73030.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:36:12,588 INFO [train.py:893] (1/4) Epoch 28, batch 200, loss[loss=0.1723, simple_loss=0.2387, pruned_loss=0.05298, over 13467.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.2298, pruned_loss=0.05367, over 1683147.73 frames. ], batch size: 100, lr: 4.68e-03, grad_scale: 16.0 2023-04-17 06:36:16,182 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9832, 2.9025, 2.5473, 1.8321, 2.0572, 2.4326, 2.5659, 3.1524], device='cuda:1'), covar=tensor([0.1029, 0.0410, 0.0585, 0.1574, 0.0590, 0.0637, 0.0729, 0.0234], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0157, 0.0134, 0.0219, 0.0121, 0.0176, 0.0186, 0.0144], device='cuda:1'), out_proj_covar=tensor([1.3415e-04, 1.1748e-04, 1.0331e-04, 1.6233e-04, 8.6829e-05, 1.3324e-04, 1.4012e-04, 1.0572e-04], device='cuda:1') 2023-04-17 06:36:19,371 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5056, 2.4005, 2.9177, 3.9989, 3.5978, 4.0693, 3.2379, 2.4612], device='cuda:1'), covar=tensor([0.0367, 0.0836, 0.0802, 0.0084, 0.0262, 0.0070, 0.0622, 0.0897], device='cuda:1'), in_proj_covar=tensor([0.0129, 0.0152, 0.0172, 0.0108, 0.0127, 0.0104, 0.0175, 0.0152], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:36:29,945 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5362, 3.8539, 3.6630, 4.2848, 2.5045, 3.2404, 4.0986, 2.3479], device='cuda:1'), covar=tensor([0.0151, 0.0545, 0.0812, 0.0545, 0.1601, 0.1064, 0.0470, 0.1882], device='cuda:1'), in_proj_covar=tensor([0.0142, 0.0193, 0.0211, 0.0257, 0.0188, 0.0206, 0.0184, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:36:59,295 INFO [train.py:893] (1/4) Epoch 28, batch 250, loss[loss=0.1484, simple_loss=0.2058, pruned_loss=0.04551, over 13384.00 frames. ], tot_loss[loss=0.1697, simple_loss=0.2309, pruned_loss=0.0543, over 1898420.51 frames. ], batch size: 62, lr: 4.68e-03, grad_scale: 16.0 2023-04-17 06:37:00,079 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.971e+02 2.642e+02 3.255e+02 4.009e+02 7.837e+02, threshold=6.511e+02, percent-clipped=3.0 2023-04-17 06:37:13,209 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 2023-04-17 06:37:21,893 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4504, 4.7500, 4.4719, 4.5296, 4.5540, 4.9217, 4.7124, 4.6289], device='cuda:1'), covar=tensor([0.0337, 0.0266, 0.0305, 0.0997, 0.0288, 0.0260, 0.0297, 0.0224], device='cuda:1'), in_proj_covar=tensor([0.0175, 0.0168, 0.0191, 0.0277, 0.0190, 0.0207, 0.0187, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 06:37:24,431 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0389, 3.7694, 3.8910, 2.4750, 4.3033, 4.0778, 4.0397, 4.2281], device='cuda:1'), covar=tensor([0.0260, 0.0216, 0.0171, 0.1126, 0.0135, 0.0238, 0.0146, 0.0122], device='cuda:1'), in_proj_covar=tensor([0.0100, 0.0062, 0.0089, 0.0106, 0.0104, 0.0117, 0.0087, 0.0084], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:37:35,170 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9901, 4.3790, 4.2584, 4.2078, 4.2166, 4.0527, 4.4435, 4.4954], device='cuda:1'), covar=tensor([0.0241, 0.0242, 0.0207, 0.0301, 0.0286, 0.0265, 0.0279, 0.0224], device='cuda:1'), in_proj_covar=tensor([0.0151, 0.0216, 0.0181, 0.0195, 0.0167, 0.0217, 0.0145, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-04-17 06:37:43,143 INFO [train.py:893] (1/4) Epoch 28, batch 300, loss[loss=0.1872, simple_loss=0.2433, pruned_loss=0.06554, over 13050.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2314, pruned_loss=0.05429, over 2062072.67 frames. ], batch size: 142, lr: 4.68e-03, grad_scale: 16.0 2023-04-17 06:38:03,374 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7456, 3.8009, 2.8706, 2.5698, 2.6840, 2.3559, 3.8358, 2.1735], device='cuda:1'), covar=tensor([0.1774, 0.0387, 0.1487, 0.2431, 0.0981, 0.3628, 0.0325, 0.4417], device='cuda:1'), in_proj_covar=tensor([0.0317, 0.0296, 0.0330, 0.0350, 0.0268, 0.0340, 0.0223, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 06:38:29,131 INFO [train.py:893] (1/4) Epoch 28, batch 350, loss[loss=0.1523, simple_loss=0.2243, pruned_loss=0.04014, over 13493.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.2315, pruned_loss=0.05403, over 2197033.27 frames. ], batch size: 93, lr: 4.68e-03, grad_scale: 16.0 2023-04-17 06:38:29,972 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.701e+02 2.601e+02 2.960e+02 3.409e+02 5.967e+02, threshold=5.920e+02, percent-clipped=0.0 2023-04-17 06:38:51,596 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73250.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:39:15,987 INFO [train.py:893] (1/4) Epoch 28, batch 400, loss[loss=0.1689, simple_loss=0.2315, pruned_loss=0.05309, over 13540.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2309, pruned_loss=0.05371, over 2299841.64 frames. ], batch size: 78, lr: 4.68e-03, grad_scale: 16.0 2023-04-17 06:39:30,165 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73293.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:39:55,155 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73320.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 06:39:59,893 INFO [train.py:893] (1/4) Epoch 28, batch 450, loss[loss=0.1954, simple_loss=0.2585, pruned_loss=0.06613, over 13361.00 frames. ], tot_loss[loss=0.171, simple_loss=0.2332, pruned_loss=0.05433, over 2377844.31 frames. ], batch size: 118, lr: 4.68e-03, grad_scale: 16.0 2023-04-17 06:40:00,717 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.805e+02 2.473e+02 2.892e+02 3.329e+02 8.217e+02, threshold=5.784e+02, percent-clipped=1.0 2023-04-17 06:40:18,788 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-04-17 06:40:26,115 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73354.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:40:26,622 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-17 06:40:46,165 INFO [train.py:893] (1/4) Epoch 28, batch 500, loss[loss=0.1881, simple_loss=0.2416, pruned_loss=0.06726, over 13184.00 frames. ], tot_loss[loss=0.1701, simple_loss=0.2327, pruned_loss=0.05378, over 2442910.60 frames. ], batch size: 58, lr: 4.67e-03, grad_scale: 16.0 2023-04-17 06:41:21,014 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4405, 4.9556, 4.7302, 4.7151, 4.6936, 4.5887, 4.9970, 5.0113], device='cuda:1'), covar=tensor([0.0233, 0.0199, 0.0205, 0.0332, 0.0324, 0.0234, 0.0252, 0.0210], device='cuda:1'), in_proj_covar=tensor([0.0151, 0.0217, 0.0182, 0.0197, 0.0168, 0.0218, 0.0146, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-04-17 06:41:30,426 INFO [train.py:893] (1/4) Epoch 28, batch 550, loss[loss=0.1745, simple_loss=0.244, pruned_loss=0.05248, over 13409.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2329, pruned_loss=0.05357, over 2492610.59 frames. ], batch size: 113, lr: 4.67e-03, grad_scale: 16.0 2023-04-17 06:41:31,240 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.685e+02 2.577e+02 2.942e+02 3.367e+02 6.254e+02, threshold=5.883e+02, percent-clipped=2.0 2023-04-17 06:41:48,752 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.87 vs. limit=5.0 2023-04-17 06:41:50,902 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3307, 4.3097, 3.0161, 3.8103, 4.2192, 2.7601, 3.8625, 2.9385], device='cuda:1'), covar=tensor([0.0265, 0.0324, 0.1082, 0.0356, 0.0249, 0.1284, 0.0478, 0.1288], device='cuda:1'), in_proj_covar=tensor([0.0168, 0.0185, 0.0179, 0.0235, 0.0143, 0.0162, 0.0164, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:42:14,504 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3588, 2.7980, 2.8487, 3.3752, 2.7294, 3.3990, 3.3214, 2.8364], device='cuda:1'), covar=tensor([0.0077, 0.0264, 0.0177, 0.0162, 0.0214, 0.0127, 0.0173, 0.0165], device='cuda:1'), in_proj_covar=tensor([0.0116, 0.0130, 0.0135, 0.0135, 0.0145, 0.0122, 0.0117, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 06:42:15,786 INFO [train.py:893] (1/4) Epoch 28, batch 600, loss[loss=0.1546, simple_loss=0.2175, pruned_loss=0.04584, over 12035.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2317, pruned_loss=0.05339, over 2525555.85 frames. ], batch size: 158, lr: 4.67e-03, grad_scale: 16.0 2023-04-17 06:43:01,679 INFO [train.py:893] (1/4) Epoch 28, batch 650, loss[loss=0.1557, simple_loss=0.2221, pruned_loss=0.04462, over 13280.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2304, pruned_loss=0.05283, over 2555374.30 frames. ], batch size: 124, lr: 4.67e-03, grad_scale: 16.0 2023-04-17 06:43:02,441 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.755e+02 2.525e+02 2.817e+02 3.484e+02 5.990e+02, threshold=5.634e+02, percent-clipped=2.0 2023-04-17 06:43:22,199 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73550.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:43:22,356 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.2474, 2.9943, 3.5276, 2.6942, 2.4761, 2.5825, 3.9327, 3.9546], device='cuda:1'), covar=tensor([0.1236, 0.1921, 0.0381, 0.1723, 0.1596, 0.1519, 0.0311, 0.0204], device='cuda:1'), in_proj_covar=tensor([0.0254, 0.0281, 0.0204, 0.0228, 0.0224, 0.0188, 0.0219, 0.0194], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:43:45,697 INFO [train.py:893] (1/4) Epoch 28, batch 700, loss[loss=0.1807, simple_loss=0.2391, pruned_loss=0.06116, over 13558.00 frames. ], tot_loss[loss=0.1671, simple_loss=0.2297, pruned_loss=0.05231, over 2577024.99 frames. ], batch size: 78, lr: 4.67e-03, grad_scale: 16.0 2023-04-17 06:44:00,347 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73590.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:44:06,745 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73598.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:44:27,196 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73620.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 06:44:32,625 INFO [train.py:893] (1/4) Epoch 28, batch 750, loss[loss=0.1655, simple_loss=0.2295, pruned_loss=0.05076, over 13363.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2303, pruned_loss=0.05299, over 2599669.68 frames. ], batch size: 118, lr: 4.67e-03, grad_scale: 16.0 2023-04-17 06:44:33,386 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.905e+02 2.661e+02 3.020e+02 3.513e+02 5.839e+02, threshold=6.040e+02, percent-clipped=2.0 2023-04-17 06:44:46,251 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.1494, 3.0585, 2.5118, 2.0532, 2.1776, 2.5832, 2.7592, 3.2651], device='cuda:1'), covar=tensor([0.1055, 0.0341, 0.0800, 0.1405, 0.0451, 0.0502, 0.0685, 0.0263], device='cuda:1'), in_proj_covar=tensor([0.0182, 0.0155, 0.0133, 0.0219, 0.0121, 0.0174, 0.0185, 0.0143], device='cuda:1'), out_proj_covar=tensor([1.3408e-04, 1.1611e-04, 1.0209e-04, 1.6210e-04, 8.7139e-05, 1.3193e-04, 1.3912e-04, 1.0504e-04], device='cuda:1') 2023-04-17 06:44:53,325 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.4976, 4.9975, 4.8922, 5.0142, 4.8722, 4.7886, 5.4755, 5.1105], device='cuda:1'), covar=tensor([0.0647, 0.1145, 0.1946, 0.2400, 0.0843, 0.1617, 0.0834, 0.1029], device='cuda:1'), in_proj_covar=tensor([0.0348, 0.0399, 0.0486, 0.0488, 0.0314, 0.0368, 0.0453, 0.0367], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 06:44:53,339 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73649.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:44:55,065 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73651.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:45:10,589 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73668.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 06:45:16,852 INFO [train.py:893] (1/4) Epoch 28, batch 800, loss[loss=0.1948, simple_loss=0.2539, pruned_loss=0.06783, over 13262.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2309, pruned_loss=0.05301, over 2615692.38 frames. ], batch size: 124, lr: 4.66e-03, grad_scale: 16.0 2023-04-17 06:45:36,587 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5990, 3.5247, 2.7832, 3.2591, 3.4917, 2.4460, 3.1582, 2.5184], device='cuda:1'), covar=tensor([0.0258, 0.0194, 0.0811, 0.0425, 0.0263, 0.1071, 0.0486, 0.1146], device='cuda:1'), in_proj_covar=tensor([0.0169, 0.0186, 0.0179, 0.0235, 0.0142, 0.0163, 0.0164, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:46:03,562 INFO [train.py:893] (1/4) Epoch 28, batch 850, loss[loss=0.1567, simple_loss=0.223, pruned_loss=0.0452, over 13464.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2316, pruned_loss=0.05337, over 2623229.46 frames. ], batch size: 79, lr: 4.66e-03, grad_scale: 16.0 2023-04-17 06:46:04,402 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.682e+02 2.586e+02 3.000e+02 3.524e+02 7.256e+02, threshold=5.999e+02, percent-clipped=2.0 2023-04-17 06:46:08,907 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=73732.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:46:48,914 INFO [train.py:893] (1/4) Epoch 28, batch 900, loss[loss=0.1714, simple_loss=0.2363, pruned_loss=0.05321, over 13503.00 frames. ], tot_loss[loss=0.1702, simple_loss=0.2323, pruned_loss=0.05404, over 2633400.96 frames. ], batch size: 93, lr: 4.66e-03, grad_scale: 16.0 2023-04-17 06:47:04,419 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=73793.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:47:18,674 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-17 06:47:34,602 INFO [train.py:893] (1/4) Epoch 28, batch 950, loss[loss=0.172, simple_loss=0.2301, pruned_loss=0.0569, over 11903.00 frames. ], tot_loss[loss=0.17, simple_loss=0.2316, pruned_loss=0.05415, over 2634630.65 frames. ], batch size: 157, lr: 4.66e-03, grad_scale: 16.0 2023-04-17 06:47:35,321 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.005e+02 2.551e+02 2.954e+02 3.450e+02 1.093e+03, threshold=5.908e+02, percent-clipped=4.0 2023-04-17 06:48:20,951 INFO [train.py:893] (1/4) Epoch 28, batch 1000, loss[loss=0.1885, simple_loss=0.2491, pruned_loss=0.06394, over 13451.00 frames. ], tot_loss[loss=0.1685, simple_loss=0.2298, pruned_loss=0.05366, over 2642235.63 frames. ], batch size: 95, lr: 4.66e-03, grad_scale: 16.0 2023-04-17 06:49:04,434 INFO [train.py:893] (1/4) Epoch 28, batch 1050, loss[loss=0.1663, simple_loss=0.2155, pruned_loss=0.05855, over 12751.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2288, pruned_loss=0.05313, over 2646798.17 frames. ], batch size: 52, lr: 4.66e-03, grad_scale: 16.0 2023-04-17 06:49:05,888 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.818e+02 2.540e+02 2.971e+02 3.541e+02 8.453e+02, threshold=5.943e+02, percent-clipped=1.0 2023-04-17 06:49:23,862 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=73946.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:49:26,444 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=73949.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:49:43,797 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7550, 2.4611, 2.1338, 1.4149, 1.6338, 1.8858, 2.1725, 2.5897], device='cuda:1'), covar=tensor([0.0898, 0.0319, 0.0643, 0.1574, 0.0214, 0.0554, 0.0686, 0.0275], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0153, 0.0131, 0.0216, 0.0119, 0.0172, 0.0182, 0.0141], device='cuda:1'), out_proj_covar=tensor([1.3261e-04, 1.1414e-04, 1.0059e-04, 1.5968e-04, 8.5614e-05, 1.3025e-04, 1.3724e-04, 1.0363e-04], device='cuda:1') 2023-04-17 06:49:51,402 INFO [train.py:893] (1/4) Epoch 28, batch 1100, loss[loss=0.1372, simple_loss=0.2016, pruned_loss=0.03641, over 13367.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2287, pruned_loss=0.05254, over 2647397.30 frames. ], batch size: 67, lr: 4.66e-03, grad_scale: 16.0 2023-04-17 06:50:10,155 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=73997.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:50:10,388 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7859, 4.0083, 3.0751, 2.8285, 2.8045, 2.4950, 4.0791, 2.3533], device='cuda:1'), covar=tensor([0.2006, 0.0363, 0.1474, 0.2405, 0.0992, 0.3758, 0.0309, 0.4408], device='cuda:1'), in_proj_covar=tensor([0.0321, 0.0300, 0.0334, 0.0355, 0.0271, 0.0343, 0.0226, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 06:50:23,977 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3442, 4.6288, 4.3760, 4.4298, 4.4297, 4.7972, 4.5501, 4.4989], device='cuda:1'), covar=tensor([0.0311, 0.0264, 0.0325, 0.0800, 0.0292, 0.0219, 0.0329, 0.0271], device='cuda:1'), in_proj_covar=tensor([0.0175, 0.0168, 0.0192, 0.0276, 0.0191, 0.0208, 0.0188, 0.0116], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 06:50:38,562 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3168, 3.9945, 4.2827, 2.7519, 4.6449, 4.3704, 4.3688, 4.5465], device='cuda:1'), covar=tensor([0.0245, 0.0157, 0.0124, 0.0978, 0.0134, 0.0241, 0.0142, 0.0091], device='cuda:1'), in_proj_covar=tensor([0.0099, 0.0061, 0.0089, 0.0106, 0.0105, 0.0118, 0.0087, 0.0083], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:50:40,628 INFO [train.py:893] (1/4) Epoch 28, batch 1150, loss[loss=0.1479, simple_loss=0.2148, pruned_loss=0.04049, over 13555.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2281, pruned_loss=0.05163, over 2650876.62 frames. ], batch size: 78, lr: 4.65e-03, grad_scale: 16.0 2023-04-17 06:50:41,406 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.778e+02 2.470e+02 2.846e+02 3.350e+02 5.021e+02, threshold=5.693e+02, percent-clipped=0.0 2023-04-17 06:51:15,791 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74064.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:51:25,527 INFO [train.py:893] (1/4) Epoch 28, batch 1200, loss[loss=0.1544, simple_loss=0.214, pruned_loss=0.04739, over 13365.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2279, pruned_loss=0.0513, over 2653147.97 frames. ], batch size: 73, lr: 4.65e-03, grad_scale: 16.0 2023-04-17 06:51:36,789 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74088.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:51:51,950 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-17 06:51:58,137 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7129, 3.7235, 2.7356, 3.4284, 3.7092, 2.4731, 3.3996, 2.4806], device='cuda:1'), covar=tensor([0.0296, 0.0230, 0.1031, 0.0366, 0.0246, 0.1208, 0.0494, 0.1311], device='cuda:1'), in_proj_covar=tensor([0.0171, 0.0189, 0.0183, 0.0239, 0.0144, 0.0166, 0.0167, 0.0164], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:52:02,130 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4114, 3.2183, 3.8571, 2.8543, 2.4764, 2.6758, 4.1419, 4.3139], device='cuda:1'), covar=tensor([0.1232, 0.1806, 0.0408, 0.1676, 0.1642, 0.1621, 0.0294, 0.0190], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0283, 0.0205, 0.0230, 0.0224, 0.0190, 0.0221, 0.0196], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:52:04,038 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-17 06:52:10,844 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74125.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:52:11,346 INFO [train.py:893] (1/4) Epoch 28, batch 1250, loss[loss=0.1616, simple_loss=0.2349, pruned_loss=0.04419, over 13342.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2293, pruned_loss=0.05222, over 2644945.79 frames. ], batch size: 109, lr: 4.65e-03, grad_scale: 16.0 2023-04-17 06:52:12,122 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.727e+02 2.579e+02 2.969e+02 3.696e+02 6.007e+02, threshold=5.937e+02, percent-clipped=2.0 2023-04-17 06:52:51,752 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3493, 3.7362, 3.4935, 4.1656, 2.2724, 3.0557, 3.9892, 2.3115], device='cuda:1'), covar=tensor([0.0203, 0.0558, 0.0964, 0.0709, 0.1844, 0.1227, 0.0466, 0.1927], device='cuda:1'), in_proj_covar=tensor([0.0142, 0.0193, 0.0212, 0.0257, 0.0188, 0.0207, 0.0185, 0.0180], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:52:57,752 INFO [train.py:893] (1/4) Epoch 28, batch 1300, loss[loss=0.155, simple_loss=0.2196, pruned_loss=0.04517, over 13552.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2302, pruned_loss=0.05236, over 2647627.66 frames. ], batch size: 72, lr: 4.65e-03, grad_scale: 16.0 2023-04-17 06:53:15,688 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-04-17 06:53:43,806 INFO [train.py:893] (1/4) Epoch 28, batch 1350, loss[loss=0.1895, simple_loss=0.2522, pruned_loss=0.0634, over 13571.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2306, pruned_loss=0.05245, over 2652195.39 frames. ], batch size: 89, lr: 4.65e-03, grad_scale: 16.0 2023-04-17 06:53:44,618 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.857e+02 2.634e+02 3.033e+02 3.647e+02 6.795e+02, threshold=6.067e+02, percent-clipped=2.0 2023-04-17 06:53:57,435 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74240.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:54:02,961 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74246.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:54:30,465 INFO [train.py:893] (1/4) Epoch 28, batch 1400, loss[loss=0.1595, simple_loss=0.2196, pruned_loss=0.04965, over 13496.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2301, pruned_loss=0.05213, over 2656551.51 frames. ], batch size: 70, lr: 4.65e-03, grad_scale: 16.0 2023-04-17 06:54:46,479 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74294.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:54:53,969 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74301.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:54:56,406 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2617, 4.5992, 4.4554, 4.3915, 4.5393, 4.1723, 4.7155, 4.7325], device='cuda:1'), covar=tensor([0.0287, 0.0380, 0.0283, 0.0471, 0.0303, 0.0383, 0.0353, 0.0322], device='cuda:1'), in_proj_covar=tensor([0.0150, 0.0216, 0.0180, 0.0195, 0.0166, 0.0216, 0.0144, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-04-17 06:55:15,845 INFO [train.py:893] (1/4) Epoch 28, batch 1450, loss[loss=0.1673, simple_loss=0.2301, pruned_loss=0.05226, over 13347.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2301, pruned_loss=0.05221, over 2658320.30 frames. ], batch size: 67, lr: 4.64e-03, grad_scale: 16.0 2023-04-17 06:55:16,670 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.844e+02 2.544e+02 3.026e+02 3.432e+02 5.527e+02, threshold=6.053e+02, percent-clipped=0.0 2023-04-17 06:55:18,982 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.72 vs. limit=5.0 2023-04-17 06:55:41,248 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6012, 2.3973, 2.9104, 3.9946, 3.6409, 4.0828, 3.1891, 2.5517], device='cuda:1'), covar=tensor([0.0253, 0.0868, 0.0783, 0.0073, 0.0229, 0.0067, 0.0654, 0.0896], device='cuda:1'), in_proj_covar=tensor([0.0128, 0.0152, 0.0172, 0.0109, 0.0126, 0.0104, 0.0175, 0.0152], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 06:56:02,057 INFO [train.py:893] (1/4) Epoch 28, batch 1500, loss[loss=0.1644, simple_loss=0.2232, pruned_loss=0.05277, over 13529.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2305, pruned_loss=0.05219, over 2663075.98 frames. ], batch size: 83, lr: 4.64e-03, grad_scale: 16.0 2023-04-17 06:56:12,866 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74388.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:56:15,867 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.92 vs. limit=2.0 2023-04-17 06:56:16,435 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-04-17 06:56:42,330 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74420.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:56:47,109 INFO [train.py:893] (1/4) Epoch 28, batch 1550, loss[loss=0.1558, simple_loss=0.2172, pruned_loss=0.04718, over 13528.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2298, pruned_loss=0.05187, over 2664382.29 frames. ], batch size: 76, lr: 4.64e-03, grad_scale: 16.0 2023-04-17 06:56:48,572 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.854e+02 2.451e+02 2.902e+02 3.422e+02 5.266e+02, threshold=5.804e+02, percent-clipped=0.0 2023-04-17 06:56:57,504 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74436.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:57:33,951 INFO [train.py:893] (1/4) Epoch 28, batch 1600, loss[loss=0.1504, simple_loss=0.2077, pruned_loss=0.04658, over 13201.00 frames. ], tot_loss[loss=0.1674, simple_loss=0.2305, pruned_loss=0.05215, over 2662275.88 frames. ], batch size: 58, lr: 4.64e-03, grad_scale: 16.0 2023-04-17 06:58:12,908 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8671, 3.7599, 3.0021, 3.3242, 3.0516, 2.3925, 3.7979, 2.2608], device='cuda:1'), covar=tensor([0.0690, 0.0459, 0.0489, 0.0406, 0.0682, 0.1757, 0.0846, 0.1227], device='cuda:1'), in_proj_covar=tensor([0.0148, 0.0149, 0.0140, 0.0121, 0.0153, 0.0193, 0.0191, 0.0148], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 06:58:19,735 INFO [train.py:893] (1/4) Epoch 28, batch 1650, loss[loss=0.1879, simple_loss=0.2603, pruned_loss=0.05777, over 13344.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2316, pruned_loss=0.05201, over 2663562.44 frames. ], batch size: 118, lr: 4.64e-03, grad_scale: 16.0 2023-04-17 06:58:20,547 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.816e+02 2.550e+02 3.022e+02 3.774e+02 7.035e+02, threshold=6.044e+02, percent-clipped=3.0 2023-04-17 06:58:55,490 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0771, 4.4233, 4.1293, 4.2200, 4.2272, 4.4990, 4.3763, 4.0944], device='cuda:1'), covar=tensor([0.0285, 0.0240, 0.0338, 0.0698, 0.0265, 0.0246, 0.0293, 0.0388], device='cuda:1'), in_proj_covar=tensor([0.0174, 0.0167, 0.0192, 0.0275, 0.0191, 0.0206, 0.0186, 0.0117], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 06:59:04,957 INFO [train.py:893] (1/4) Epoch 28, batch 1700, loss[loss=0.1415, simple_loss=0.2076, pruned_loss=0.03763, over 13368.00 frames. ], tot_loss[loss=0.1672, simple_loss=0.2314, pruned_loss=0.0515, over 2663903.51 frames. ], batch size: 62, lr: 4.64e-03, grad_scale: 16.0 2023-04-17 06:59:23,200 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74595.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:59:23,939 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74596.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 06:59:51,394 INFO [train.py:893] (1/4) Epoch 28, batch 1750, loss[loss=0.1728, simple_loss=0.235, pruned_loss=0.05534, over 13521.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2302, pruned_loss=0.0508, over 2665824.94 frames. ], batch size: 91, lr: 4.64e-03, grad_scale: 16.0 2023-04-17 06:59:52,265 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.683e+02 2.479e+02 3.029e+02 3.465e+02 5.673e+02, threshold=6.058e+02, percent-clipped=0.0 2023-04-17 07:00:01,673 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3613, 2.2250, 2.6686, 3.6761, 3.4052, 3.7896, 3.0075, 2.2421], device='cuda:1'), covar=tensor([0.0270, 0.0911, 0.0798, 0.0109, 0.0266, 0.0078, 0.0686, 0.0982], device='cuda:1'), in_proj_covar=tensor([0.0127, 0.0152, 0.0171, 0.0108, 0.0126, 0.0104, 0.0175, 0.0152], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:00:19,724 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74656.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:00:38,087 INFO [train.py:893] (1/4) Epoch 28, batch 1800, loss[loss=0.1603, simple_loss=0.227, pruned_loss=0.04677, over 13094.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.229, pruned_loss=0.05017, over 2664732.12 frames. ], batch size: 142, lr: 4.63e-03, grad_scale: 16.0 2023-04-17 07:01:18,091 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74720.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:01:23,276 INFO [train.py:893] (1/4) Epoch 28, batch 1850, loss[loss=0.1515, simple_loss=0.2163, pruned_loss=0.04338, over 13547.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2279, pruned_loss=0.05, over 2664139.43 frames. ], batch size: 87, lr: 4.63e-03, grad_scale: 16.0 2023-04-17 07:01:24,077 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.611e+02 2.423e+02 2.814e+02 3.269e+02 7.266e+02, threshold=5.629e+02, percent-clipped=1.0 2023-04-17 07:01:24,166 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-17 07:02:01,884 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74768.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:02:09,151 INFO [train.py:893] (1/4) Epoch 28, batch 1900, loss[loss=0.1719, simple_loss=0.2327, pruned_loss=0.05553, over 13255.00 frames. ], tot_loss[loss=0.1646, simple_loss=0.2281, pruned_loss=0.05056, over 2666456.09 frames. ], batch size: 132, lr: 4.63e-03, grad_scale: 16.0 2023-04-17 07:02:16,907 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.80 vs. limit=5.0 2023-04-17 07:02:54,447 INFO [train.py:893] (1/4) Epoch 28, batch 1950, loss[loss=0.1968, simple_loss=0.2463, pruned_loss=0.07368, over 11835.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2285, pruned_loss=0.05122, over 2666277.13 frames. ], batch size: 157, lr: 4.63e-03, grad_scale: 16.0 2023-04-17 07:02:55,262 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.498e+02 2.666e+02 3.093e+02 3.546e+02 8.280e+02, threshold=6.185e+02, percent-clipped=1.0 2023-04-17 07:03:02,871 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-04-17 07:03:28,592 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7316, 2.4162, 3.2039, 4.2072, 3.8306, 4.2619, 3.5639, 2.7197], device='cuda:1'), covar=tensor([0.0319, 0.1219, 0.0765, 0.0088, 0.0245, 0.0088, 0.0587, 0.0907], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0152, 0.0172, 0.0108, 0.0126, 0.0103, 0.0174, 0.0151], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:03:31,988 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.90 vs. limit=2.0 2023-04-17 07:03:40,122 INFO [train.py:893] (1/4) Epoch 28, batch 2000, loss[loss=0.2017, simple_loss=0.267, pruned_loss=0.06821, over 13430.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2297, pruned_loss=0.05176, over 2668471.72 frames. ], batch size: 95, lr: 4.63e-03, grad_scale: 16.0 2023-04-17 07:03:44,949 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-17 07:03:59,089 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=74896.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:04:10,699 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74909.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:04:26,422 INFO [train.py:893] (1/4) Epoch 28, batch 2050, loss[loss=0.1847, simple_loss=0.2349, pruned_loss=0.06728, over 13206.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.2314, pruned_loss=0.05219, over 2669640.79 frames. ], batch size: 58, lr: 4.63e-03, grad_scale: 32.0 2023-04-17 07:04:27,211 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.636e+02 2.577e+02 2.966e+02 3.569e+02 6.675e+02, threshold=5.932e+02, percent-clipped=3.0 2023-04-17 07:04:42,665 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=74944.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:04:48,462 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=74951.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:05:01,398 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=74964.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:05:06,347 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=74970.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:05:08,762 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8772, 3.5805, 2.9264, 3.2759, 2.8881, 2.1908, 3.6603, 2.1399], device='cuda:1'), covar=tensor([0.0662, 0.0618, 0.0474, 0.0424, 0.0690, 0.1998, 0.1017, 0.1310], device='cuda:1'), in_proj_covar=tensor([0.0148, 0.0151, 0.0140, 0.0122, 0.0152, 0.0194, 0.0190, 0.0149], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 07:05:11,608 INFO [train.py:893] (1/4) Epoch 28, batch 2100, loss[loss=0.1794, simple_loss=0.2382, pruned_loss=0.06027, over 13542.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2308, pruned_loss=0.05163, over 2671636.73 frames. ], batch size: 98, lr: 4.62e-03, grad_scale: 32.0 2023-04-17 07:05:56,407 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75025.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:05:56,922 INFO [train.py:893] (1/4) Epoch 28, batch 2150, loss[loss=0.1402, simple_loss=0.204, pruned_loss=0.03824, over 13448.00 frames. ], tot_loss[loss=0.167, simple_loss=0.231, pruned_loss=0.05145, over 2669224.26 frames. ], batch size: 65, lr: 4.62e-03, grad_scale: 32.0 2023-04-17 07:05:58,384 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.932e+02 2.614e+02 2.974e+02 3.546e+02 5.707e+02, threshold=5.948e+02, percent-clipped=0.0 2023-04-17 07:06:00,417 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1208, 4.1720, 3.0722, 3.8225, 4.0725, 2.7110, 3.7939, 2.7315], device='cuda:1'), covar=tensor([0.0304, 0.0281, 0.0953, 0.0392, 0.0299, 0.1199, 0.0476, 0.1237], device='cuda:1'), in_proj_covar=tensor([0.0169, 0.0187, 0.0179, 0.0238, 0.0144, 0.0163, 0.0165, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:06:21,256 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.8599, 4.2008, 3.7401, 4.7528, 2.2677, 3.1884, 4.3012, 2.5965], device='cuda:1'), covar=tensor([0.0111, 0.0471, 0.0915, 0.0532, 0.1992, 0.1247, 0.0456, 0.1816], device='cuda:1'), in_proj_covar=tensor([0.0145, 0.0196, 0.0214, 0.0260, 0.0190, 0.0209, 0.0186, 0.0181], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:06:37,889 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.95 vs. limit=2.0 2023-04-17 07:06:42,118 INFO [train.py:893] (1/4) Epoch 28, batch 2200, loss[loss=0.1641, simple_loss=0.2259, pruned_loss=0.05117, over 13524.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2297, pruned_loss=0.05089, over 2666495.27 frames. ], batch size: 85, lr: 4.62e-03, grad_scale: 16.0 2023-04-17 07:07:10,760 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2249, 4.2548, 3.0279, 3.8206, 4.2029, 2.7326, 3.8495, 2.9092], device='cuda:1'), covar=tensor([0.0287, 0.0240, 0.1051, 0.0381, 0.0239, 0.1241, 0.0439, 0.1244], device='cuda:1'), in_proj_covar=tensor([0.0169, 0.0189, 0.0181, 0.0239, 0.0144, 0.0163, 0.0166, 0.0161], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:07:28,230 INFO [train.py:893] (1/4) Epoch 28, batch 2250, loss[loss=0.1765, simple_loss=0.2436, pruned_loss=0.05466, over 13468.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.229, pruned_loss=0.05067, over 2667389.56 frames. ], batch size: 100, lr: 4.62e-03, grad_scale: 16.0 2023-04-17 07:07:29,797 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.746e+02 2.493e+02 2.845e+02 3.262e+02 6.199e+02, threshold=5.691e+02, percent-clipped=1.0 2023-04-17 07:08:13,255 INFO [train.py:893] (1/4) Epoch 28, batch 2300, loss[loss=0.1525, simple_loss=0.2173, pruned_loss=0.04383, over 13470.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2277, pruned_loss=0.05021, over 2664939.23 frames. ], batch size: 79, lr: 4.62e-03, grad_scale: 16.0 2023-04-17 07:08:55,885 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7238, 3.6709, 2.7811, 2.5253, 2.6248, 2.3694, 3.7694, 2.1486], device='cuda:1'), covar=tensor([0.1792, 0.0433, 0.1561, 0.2705, 0.1005, 0.3624, 0.0367, 0.4565], device='cuda:1'), in_proj_covar=tensor([0.0322, 0.0300, 0.0334, 0.0354, 0.0271, 0.0343, 0.0227, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 07:08:58,909 INFO [train.py:893] (1/4) Epoch 28, batch 2350, loss[loss=0.1582, simple_loss=0.2206, pruned_loss=0.04794, over 13202.00 frames. ], tot_loss[loss=0.1632, simple_loss=0.2271, pruned_loss=0.0497, over 2665883.15 frames. ], batch size: 132, lr: 4.62e-03, grad_scale: 16.0 2023-04-17 07:09:00,521 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.654e+02 2.368e+02 2.701e+02 3.226e+02 5.813e+02, threshold=5.403e+02, percent-clipped=1.0 2023-04-17 07:09:00,856 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75228.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:09:02,609 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5208, 3.8554, 3.6270, 4.3267, 2.3307, 3.2693, 4.0300, 2.3724], device='cuda:1'), covar=tensor([0.0144, 0.0476, 0.0840, 0.0457, 0.1716, 0.1015, 0.0486, 0.1810], device='cuda:1'), in_proj_covar=tensor([0.0143, 0.0193, 0.0211, 0.0256, 0.0187, 0.0206, 0.0183, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:09:11,503 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 2023-04-17 07:09:22,440 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-17 07:09:22,644 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75251.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:09:31,892 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5454, 4.4415, 4.4636, 2.8826, 4.8111, 4.5737, 4.5609, 4.7051], device='cuda:1'), covar=tensor([0.0251, 0.0129, 0.0166, 0.1004, 0.0157, 0.0254, 0.0161, 0.0118], device='cuda:1'), in_proj_covar=tensor([0.0101, 0.0063, 0.0091, 0.0108, 0.0107, 0.0120, 0.0090, 0.0085], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:09:34,957 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75265.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:09:39,855 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=75270.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:09:44,434 INFO [train.py:893] (1/4) Epoch 28, batch 2400, loss[loss=0.1897, simple_loss=0.249, pruned_loss=0.06514, over 13357.00 frames. ], tot_loss[loss=0.1628, simple_loss=0.2261, pruned_loss=0.04971, over 2666897.19 frames. ], batch size: 109, lr: 4.62e-03, grad_scale: 16.0 2023-04-17 07:09:57,052 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75289.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:10:05,936 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75299.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:10:22,782 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4717, 2.4122, 2.7810, 3.8320, 3.5127, 3.9425, 3.0865, 2.3507], device='cuda:1'), covar=tensor([0.0270, 0.0868, 0.0800, 0.0081, 0.0250, 0.0068, 0.0707, 0.0986], device='cuda:1'), in_proj_covar=tensor([0.0127, 0.0152, 0.0171, 0.0108, 0.0126, 0.0103, 0.0174, 0.0152], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:10:25,025 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75320.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:10:27,572 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9993, 4.2384, 4.1941, 3.7621, 4.0345, 4.3575, 4.3339, 4.1692], device='cuda:1'), covar=tensor([0.0315, 0.0329, 0.0355, 0.1092, 0.0366, 0.0329, 0.0344, 0.0462], device='cuda:1'), in_proj_covar=tensor([0.0172, 0.0168, 0.0190, 0.0273, 0.0189, 0.0206, 0.0185, 0.0116], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 07:10:29,680 INFO [train.py:893] (1/4) Epoch 28, batch 2450, loss[loss=0.1852, simple_loss=0.2476, pruned_loss=0.06142, over 13285.00 frames. ], tot_loss[loss=0.163, simple_loss=0.2264, pruned_loss=0.04984, over 2666162.66 frames. ], batch size: 124, lr: 4.61e-03, grad_scale: 16.0 2023-04-17 07:10:31,309 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.808e+02 2.390e+02 2.893e+02 3.505e+02 5.299e+02, threshold=5.785e+02, percent-clipped=0.0 2023-04-17 07:10:34,829 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=75331.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 07:10:58,181 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.6430, 5.2014, 5.0142, 5.2185, 5.0215, 4.9639, 5.6555, 5.2494], device='cuda:1'), covar=tensor([0.0815, 0.1131, 0.2186, 0.2204, 0.0771, 0.1833, 0.0862, 0.1232], device='cuda:1'), in_proj_covar=tensor([0.0359, 0.0409, 0.0501, 0.0504, 0.0322, 0.0378, 0.0466, 0.0379], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:11:05,633 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3393, 2.5551, 2.3059, 4.2488, 4.6690, 3.4337, 4.5605, 4.4095], device='cuda:1'), covar=tensor([0.0099, 0.0965, 0.0983, 0.0097, 0.0079, 0.0462, 0.0078, 0.0090], device='cuda:1'), in_proj_covar=tensor([0.0081, 0.0095, 0.0102, 0.0087, 0.0074, 0.0084, 0.0061, 0.0085], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 07:11:16,471 INFO [train.py:893] (1/4) Epoch 28, batch 2500, loss[loss=0.1852, simple_loss=0.2412, pruned_loss=0.06461, over 13056.00 frames. ], tot_loss[loss=0.1633, simple_loss=0.2267, pruned_loss=0.04993, over 2664254.50 frames. ], batch size: 142, lr: 4.61e-03, grad_scale: 16.0 2023-04-17 07:12:01,180 INFO [train.py:893] (1/4) Epoch 28, batch 2550, loss[loss=0.1697, simple_loss=0.2249, pruned_loss=0.05727, over 13365.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2273, pruned_loss=0.05015, over 2665281.85 frames. ], batch size: 67, lr: 4.61e-03, grad_scale: 16.0 2023-04-17 07:12:03,482 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.759e+02 2.463e+02 2.650e+02 3.401e+02 6.679e+02, threshold=5.300e+02, percent-clipped=1.0 2023-04-17 07:12:05,293 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.5876, 5.1440, 4.9465, 5.0998, 4.9254, 4.9111, 5.5736, 5.1492], device='cuda:1'), covar=tensor([0.0683, 0.1057, 0.2123, 0.2144, 0.0845, 0.1639, 0.0761, 0.1037], device='cuda:1'), in_proj_covar=tensor([0.0360, 0.0410, 0.0503, 0.0505, 0.0324, 0.0380, 0.0466, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:12:11,194 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-04-17 07:12:25,393 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-17 07:12:47,077 INFO [train.py:893] (1/4) Epoch 28, batch 2600, loss[loss=0.1839, simple_loss=0.2369, pruned_loss=0.06544, over 13455.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2273, pruned_loss=0.05031, over 2664071.61 frames. ], batch size: 79, lr: 4.61e-03, grad_scale: 16.0 2023-04-17 07:13:15,030 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1759, 2.5678, 2.3028, 4.1111, 4.5652, 3.3702, 4.4712, 4.2636], device='cuda:1'), covar=tensor([0.0106, 0.0918, 0.0952, 0.0104, 0.0086, 0.0485, 0.0082, 0.0096], device='cuda:1'), in_proj_covar=tensor([0.0081, 0.0094, 0.0101, 0.0086, 0.0074, 0.0084, 0.0061, 0.0084], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 07:13:28,710 INFO [train.py:893] (1/4) Epoch 28, batch 2650, loss[loss=0.1516, simple_loss=0.2185, pruned_loss=0.04231, over 13501.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2278, pruned_loss=0.051, over 2658976.57 frames. ], batch size: 70, lr: 4.61e-03, grad_scale: 16.0 2023-04-17 07:13:30,203 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.967e+02 2.635e+02 3.052e+02 3.899e+02 8.488e+02, threshold=6.104e+02, percent-clipped=5.0 2023-04-17 07:13:57,803 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1702, 2.4509, 2.0287, 4.0829, 4.5080, 3.2970, 4.4040, 4.2709], device='cuda:1'), covar=tensor([0.0103, 0.1070, 0.1185, 0.0097, 0.0078, 0.0524, 0.0099, 0.0084], device='cuda:1'), in_proj_covar=tensor([0.0081, 0.0095, 0.0102, 0.0087, 0.0074, 0.0085, 0.0062, 0.0085], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 07:13:58,450 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75565.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:14:25,190 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-17 07:14:35,392 INFO [train.py:893] (1/4) Epoch 29, batch 0, loss[loss=0.1577, simple_loss=0.2193, pruned_loss=0.04806, over 13536.00 frames. ], tot_loss[loss=0.1577, simple_loss=0.2193, pruned_loss=0.04806, over 13536.00 frames. ], batch size: 72, lr: 4.52e-03, grad_scale: 16.0 2023-04-17 07:14:35,392 INFO [train.py:918] (1/4) Computing validation loss 2023-04-17 07:14:56,942 INFO [train.py:927] (1/4) Epoch 29, validation: loss=0.1316, simple_loss=0.1914, pruned_loss=0.03588, over 2446609.00 frames. 2023-04-17 07:14:56,943 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12831MB 2023-04-17 07:15:06,287 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75584.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:15:21,661 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-04-17 07:15:31,947 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75613.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:15:38,494 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75620.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:15:42,446 INFO [train.py:893] (1/4) Epoch 29, batch 50, loss[loss=0.1657, simple_loss=0.2236, pruned_loss=0.05391, over 13537.00 frames. ], tot_loss[loss=0.1589, simple_loss=0.2203, pruned_loss=0.0487, over 604446.87 frames. ], batch size: 83, lr: 4.52e-03, grad_scale: 16.0 2023-04-17 07:15:43,489 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=75626.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 07:15:44,925 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.808e+02 2.542e+02 2.969e+02 3.744e+02 7.692e+02, threshold=5.938e+02, percent-clipped=2.0 2023-04-17 07:15:52,149 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.62 vs. limit=2.0 2023-04-17 07:15:58,590 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9981, 2.1569, 1.8488, 3.8464, 4.2587, 3.0713, 4.1774, 3.9970], device='cuda:1'), covar=tensor([0.0166, 0.1426, 0.1532, 0.0161, 0.0211, 0.0805, 0.0187, 0.0156], device='cuda:1'), in_proj_covar=tensor([0.0081, 0.0095, 0.0102, 0.0086, 0.0074, 0.0084, 0.0061, 0.0084], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 07:16:06,953 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-17 07:16:06,953 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-17 07:16:06,953 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-17 07:16:06,961 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-17 07:16:06,980 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-17 07:16:06,993 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-17 07:16:07,012 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-17 07:16:21,734 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75668.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:16:27,291 INFO [train.py:893] (1/4) Epoch 29, batch 100, loss[loss=0.1843, simple_loss=0.2409, pruned_loss=0.06381, over 13536.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2259, pruned_loss=0.05154, over 1059972.70 frames. ], batch size: 83, lr: 4.52e-03, grad_scale: 16.0 2023-04-17 07:17:13,301 INFO [train.py:893] (1/4) Epoch 29, batch 150, loss[loss=0.1737, simple_loss=0.2429, pruned_loss=0.05226, over 13288.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2288, pruned_loss=0.05323, over 1405805.71 frames. ], batch size: 124, lr: 4.52e-03, grad_scale: 16.0 2023-04-17 07:17:16,371 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.802e+02 2.595e+02 2.945e+02 3.538e+02 7.026e+02, threshold=5.889e+02, percent-clipped=1.0 2023-04-17 07:17:59,198 INFO [train.py:893] (1/4) Epoch 29, batch 200, loss[loss=0.1866, simple_loss=0.2405, pruned_loss=0.06628, over 13505.00 frames. ], tot_loss[loss=0.1689, simple_loss=0.23, pruned_loss=0.05392, over 1675603.68 frames. ], batch size: 70, lr: 4.52e-03, grad_scale: 16.0 2023-04-17 07:18:45,525 INFO [train.py:893] (1/4) Epoch 29, batch 250, loss[loss=0.165, simple_loss=0.2278, pruned_loss=0.05114, over 13391.00 frames. ], tot_loss[loss=0.1698, simple_loss=0.231, pruned_loss=0.05428, over 1899750.50 frames. ], batch size: 113, lr: 4.52e-03, grad_scale: 16.0 2023-04-17 07:18:48,007 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.006e+02 2.662e+02 3.045e+02 3.757e+02 6.774e+02, threshold=6.090e+02, percent-clipped=3.0 2023-04-17 07:19:06,288 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-04-17 07:19:30,081 INFO [train.py:893] (1/4) Epoch 29, batch 300, loss[loss=0.1666, simple_loss=0.2352, pruned_loss=0.04898, over 13377.00 frames. ], tot_loss[loss=0.1692, simple_loss=0.2312, pruned_loss=0.05357, over 2069635.33 frames. ], batch size: 113, lr: 4.52e-03, grad_scale: 16.0 2023-04-17 07:19:38,373 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75884.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:19:42,022 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-04-17 07:19:47,044 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.43 vs. limit=2.0 2023-04-17 07:19:48,345 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6062, 3.5313, 2.7284, 3.0017, 2.9320, 2.1554, 3.5553, 2.0362], device='cuda:1'), covar=tensor([0.0764, 0.0480, 0.0564, 0.0472, 0.0642, 0.1901, 0.1080, 0.1300], device='cuda:1'), in_proj_covar=tensor([0.0148, 0.0151, 0.0141, 0.0122, 0.0153, 0.0194, 0.0192, 0.0150], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 07:20:15,639 INFO [train.py:893] (1/4) Epoch 29, batch 350, loss[loss=0.1768, simple_loss=0.2435, pruned_loss=0.05509, over 13430.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2314, pruned_loss=0.05339, over 2202582.87 frames. ], batch size: 106, lr: 4.51e-03, grad_scale: 16.0 2023-04-17 07:20:16,777 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=75926.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 07:20:18,050 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.812e+02 2.506e+02 3.011e+02 3.603e+02 7.167e+02, threshold=6.022e+02, percent-clipped=1.0 2023-04-17 07:20:22,823 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75932.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:20:59,766 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=75974.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:21:00,425 INFO [train.py:893] (1/4) Epoch 29, batch 400, loss[loss=0.1454, simple_loss=0.2083, pruned_loss=0.04123, over 13376.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.2305, pruned_loss=0.0529, over 2304784.47 frames. ], batch size: 67, lr: 4.51e-03, grad_scale: 16.0 2023-04-17 07:21:49,780 INFO [train.py:893] (1/4) Epoch 29, batch 450, loss[loss=0.1524, simple_loss=0.2233, pruned_loss=0.04076, over 13497.00 frames. ], tot_loss[loss=0.1691, simple_loss=0.2315, pruned_loss=0.05334, over 2382724.32 frames. ], batch size: 81, lr: 4.51e-03, grad_scale: 16.0 2023-04-17 07:21:52,278 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.887e+02 2.551e+02 2.867e+02 3.275e+02 5.566e+02, threshold=5.734e+02, percent-clipped=0.0 2023-04-17 07:22:14,354 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-17 07:22:21,186 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3551, 3.0890, 3.8390, 2.7227, 2.5526, 2.6471, 4.0879, 4.2095], device='cuda:1'), covar=tensor([0.1214, 0.1895, 0.0400, 0.1836, 0.1590, 0.1523, 0.0286, 0.0239], device='cuda:1'), in_proj_covar=tensor([0.0256, 0.0283, 0.0205, 0.0228, 0.0223, 0.0189, 0.0221, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:22:33,986 INFO [train.py:893] (1/4) Epoch 29, batch 500, loss[loss=0.1608, simple_loss=0.2285, pruned_loss=0.04653, over 13339.00 frames. ], tot_loss[loss=0.1681, simple_loss=0.231, pruned_loss=0.05262, over 2443580.01 frames. ], batch size: 118, lr: 4.51e-03, grad_scale: 16.0 2023-04-17 07:22:39,215 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7685, 3.2118, 3.1986, 3.6051, 2.2962, 2.8159, 3.3787, 2.0331], device='cuda:1'), covar=tensor([0.0150, 0.0636, 0.0746, 0.0583, 0.1448, 0.0992, 0.0605, 0.1845], device='cuda:1'), in_proj_covar=tensor([0.0143, 0.0193, 0.0210, 0.0257, 0.0187, 0.0206, 0.0184, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:23:00,577 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6193, 3.5546, 2.7260, 3.2421, 3.5841, 2.3126, 3.3163, 2.4447], device='cuda:1'), covar=tensor([0.0302, 0.0233, 0.0983, 0.0364, 0.0308, 0.1253, 0.0536, 0.1237], device='cuda:1'), in_proj_covar=tensor([0.0169, 0.0187, 0.0179, 0.0237, 0.0144, 0.0163, 0.0165, 0.0160], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:23:19,785 INFO [train.py:893] (1/4) Epoch 29, batch 550, loss[loss=0.1552, simple_loss=0.213, pruned_loss=0.04869, over 13348.00 frames. ], tot_loss[loss=0.1678, simple_loss=0.2309, pruned_loss=0.0524, over 2491301.15 frames. ], batch size: 62, lr: 4.51e-03, grad_scale: 16.0 2023-04-17 07:23:22,214 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.913e+02 2.461e+02 2.951e+02 3.383e+02 6.216e+02, threshold=5.902e+02, percent-clipped=2.0 2023-04-17 07:23:40,109 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.8633, 4.2059, 3.8399, 4.8021, 2.5118, 3.3104, 4.2907, 2.5835], device='cuda:1'), covar=tensor([0.0133, 0.0474, 0.0770, 0.0469, 0.1646, 0.1103, 0.0490, 0.1658], device='cuda:1'), in_proj_covar=tensor([0.0143, 0.0192, 0.0210, 0.0257, 0.0187, 0.0205, 0.0183, 0.0178], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:24:03,455 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6722, 3.4570, 2.7899, 3.0612, 2.8653, 2.2801, 3.5243, 2.0309], device='cuda:1'), covar=tensor([0.0767, 0.0643, 0.0521, 0.0476, 0.0726, 0.1744, 0.0822, 0.1321], device='cuda:1'), in_proj_covar=tensor([0.0147, 0.0151, 0.0140, 0.0121, 0.0153, 0.0192, 0.0190, 0.0148], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 07:24:04,828 INFO [train.py:893] (1/4) Epoch 29, batch 600, loss[loss=0.1539, simple_loss=0.2123, pruned_loss=0.04773, over 13523.00 frames. ], tot_loss[loss=0.1669, simple_loss=0.2299, pruned_loss=0.052, over 2528222.71 frames. ], batch size: 72, lr: 4.51e-03, grad_scale: 16.0 2023-04-17 07:24:15,402 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0529, 2.0719, 3.9227, 3.7396, 3.7083, 2.9213, 3.5734, 2.8851], device='cuda:1'), covar=tensor([0.2159, 0.1522, 0.0164, 0.0271, 0.0277, 0.0836, 0.0278, 0.0867], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0189, 0.0135, 0.0140, 0.0144, 0.0181, 0.0153, 0.0191], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-17 07:24:50,354 INFO [train.py:893] (1/4) Epoch 29, batch 650, loss[loss=0.1534, simple_loss=0.2209, pruned_loss=0.04298, over 13436.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2291, pruned_loss=0.05178, over 2554098.48 frames. ], batch size: 106, lr: 4.51e-03, grad_scale: 16.0 2023-04-17 07:24:53,530 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.828e+02 2.554e+02 2.954e+02 3.493e+02 1.019e+03, threshold=5.908e+02, percent-clipped=2.0 2023-04-17 07:25:08,682 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-04-17 07:25:31,156 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76269.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 07:25:37,231 INFO [train.py:893] (1/4) Epoch 29, batch 700, loss[loss=0.1681, simple_loss=0.2204, pruned_loss=0.05791, over 13521.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.228, pruned_loss=0.05119, over 2574539.32 frames. ], batch size: 91, lr: 4.50e-03, grad_scale: 16.0 2023-04-17 07:25:50,591 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2726, 4.7070, 4.4833, 4.4769, 4.4999, 4.3325, 4.7369, 4.7837], device='cuda:1'), covar=tensor([0.0225, 0.0222, 0.0239, 0.0347, 0.0292, 0.0270, 0.0267, 0.0223], device='cuda:1'), in_proj_covar=tensor([0.0151, 0.0214, 0.0181, 0.0195, 0.0167, 0.0216, 0.0144, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-04-17 07:25:53,809 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76294.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:26:01,881 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9686, 1.9026, 3.8245, 3.6973, 3.6399, 2.8791, 3.4716, 2.8988], device='cuda:1'), covar=tensor([0.2231, 0.1640, 0.0190, 0.0183, 0.0255, 0.0820, 0.0285, 0.0823], device='cuda:1'), in_proj_covar=tensor([0.0221, 0.0189, 0.0135, 0.0140, 0.0143, 0.0180, 0.0153, 0.0190], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-17 07:26:20,342 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.1354, 4.9990, 5.1508, 5.0046, 5.4647, 4.9824, 5.4632, 5.3694], device='cuda:1'), covar=tensor([0.0418, 0.0549, 0.0657, 0.0538, 0.0502, 0.0836, 0.0454, 0.0541], device='cuda:1'), in_proj_covar=tensor([0.0318, 0.0317, 0.0330, 0.0244, 0.0466, 0.0368, 0.0307, 0.0326], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-04-17 07:26:20,408 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76323.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 07:26:21,233 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0650, 2.4995, 2.0964, 3.9075, 4.3684, 3.2650, 4.2776, 4.1396], device='cuda:1'), covar=tensor([0.0102, 0.0998, 0.1071, 0.0100, 0.0087, 0.0508, 0.0085, 0.0092], device='cuda:1'), in_proj_covar=tensor([0.0081, 0.0094, 0.0102, 0.0087, 0.0074, 0.0085, 0.0061, 0.0084], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 07:26:21,754 INFO [train.py:893] (1/4) Epoch 29, batch 750, loss[loss=0.1613, simple_loss=0.2279, pruned_loss=0.04731, over 13255.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2281, pruned_loss=0.05123, over 2596055.05 frames. ], batch size: 117, lr: 4.50e-03, grad_scale: 16.0 2023-04-17 07:26:24,125 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.818e+02 2.465e+02 2.925e+02 3.423e+02 5.602e+02, threshold=5.850e+02, percent-clipped=0.0 2023-04-17 07:26:26,870 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76330.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 07:26:39,889 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4666, 2.5477, 2.9331, 3.9098, 3.5736, 3.9832, 3.0429, 2.6621], device='cuda:1'), covar=tensor([0.0260, 0.0758, 0.0710, 0.0068, 0.0247, 0.0069, 0.0696, 0.0727], device='cuda:1'), in_proj_covar=tensor([0.0126, 0.0151, 0.0168, 0.0107, 0.0125, 0.0103, 0.0172, 0.0150], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:26:44,109 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.3067, 4.8667, 4.7086, 4.8579, 4.6570, 4.6816, 5.2631, 4.8364], device='cuda:1'), covar=tensor([0.0675, 0.1179, 0.1970, 0.2239, 0.0995, 0.1599, 0.0791, 0.1084], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0403, 0.0498, 0.0498, 0.0319, 0.0372, 0.0459, 0.0375], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:26:49,881 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76355.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:27:08,051 INFO [train.py:893] (1/4) Epoch 29, batch 800, loss[loss=0.1792, simple_loss=0.2363, pruned_loss=0.06102, over 13539.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2297, pruned_loss=0.05197, over 2614177.44 frames. ], batch size: 85, lr: 4.50e-03, grad_scale: 16.0 2023-04-17 07:27:16,349 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76384.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 07:27:29,139 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.8551, 4.6644, 4.9134, 4.8042, 5.1535, 4.6717, 5.1564, 5.0906], device='cuda:1'), covar=tensor([0.0415, 0.0573, 0.0647, 0.0558, 0.0562, 0.0792, 0.0485, 0.0506], device='cuda:1'), in_proj_covar=tensor([0.0318, 0.0318, 0.0331, 0.0245, 0.0467, 0.0368, 0.0307, 0.0326], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-04-17 07:27:53,012 INFO [train.py:893] (1/4) Epoch 29, batch 850, loss[loss=0.1767, simple_loss=0.2332, pruned_loss=0.0601, over 12986.00 frames. ], tot_loss[loss=0.1677, simple_loss=0.2306, pruned_loss=0.05243, over 2626207.98 frames. ], batch size: 142, lr: 4.50e-03, grad_scale: 16.0 2023-04-17 07:27:56,137 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.913e+02 2.473e+02 2.862e+02 3.521e+02 5.957e+02, threshold=5.724e+02, percent-clipped=2.0 2023-04-17 07:27:59,838 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.2452, 4.3873, 3.3289, 3.0390, 3.3124, 2.7425, 4.5387, 2.6441], device='cuda:1'), covar=tensor([0.1650, 0.0402, 0.1389, 0.2145, 0.0804, 0.3227, 0.0283, 0.3860], device='cuda:1'), in_proj_covar=tensor([0.0320, 0.0301, 0.0334, 0.0355, 0.0271, 0.0343, 0.0229, 0.0380], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 07:28:21,564 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8072, 2.5294, 2.5694, 2.9366, 2.1494, 2.9070, 2.8665, 2.4154], device='cuda:1'), covar=tensor([0.0085, 0.0184, 0.0146, 0.0128, 0.0236, 0.0135, 0.0148, 0.0181], device='cuda:1'), in_proj_covar=tensor([0.0114, 0.0130, 0.0133, 0.0134, 0.0143, 0.0122, 0.0116, 0.0120], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 07:28:38,691 INFO [train.py:893] (1/4) Epoch 29, batch 900, loss[loss=0.1651, simple_loss=0.2216, pruned_loss=0.05424, over 13220.00 frames. ], tot_loss[loss=0.1686, simple_loss=0.231, pruned_loss=0.05313, over 2636469.83 frames. ], batch size: 132, lr: 4.50e-03, grad_scale: 16.0 2023-04-17 07:29:10,026 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-17 07:29:15,320 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.58 vs. limit=5.0 2023-04-17 07:29:24,502 INFO [train.py:893] (1/4) Epoch 29, batch 950, loss[loss=0.1496, simple_loss=0.2119, pruned_loss=0.04364, over 13558.00 frames. ], tot_loss[loss=0.1679, simple_loss=0.23, pruned_loss=0.05289, over 2645425.98 frames. ], batch size: 78, lr: 4.50e-03, grad_scale: 16.0 2023-04-17 07:29:27,604 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.840e+02 2.700e+02 3.084e+02 3.655e+02 6.677e+02, threshold=6.169e+02, percent-clipped=1.0 2023-04-17 07:29:58,065 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.1348, 3.0838, 2.8388, 2.1613, 2.2713, 2.7125, 2.7774, 3.2626], device='cuda:1'), covar=tensor([0.0923, 0.0326, 0.0563, 0.1209, 0.0479, 0.0596, 0.0667, 0.0291], device='cuda:1'), in_proj_covar=tensor([0.0180, 0.0156, 0.0131, 0.0213, 0.0118, 0.0173, 0.0182, 0.0143], device='cuda:1'), out_proj_covar=tensor([1.3268e-04, 1.1597e-04, 1.0060e-04, 1.5761e-04, 8.4761e-05, 1.3045e-04, 1.3719e-04, 1.0522e-04], device='cuda:1') 2023-04-17 07:30:09,169 INFO [train.py:893] (1/4) Epoch 29, batch 1000, loss[loss=0.1831, simple_loss=0.233, pruned_loss=0.06665, over 12214.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2277, pruned_loss=0.05212, over 2645823.30 frames. ], batch size: 158, lr: 4.49e-03, grad_scale: 16.0 2023-04-17 07:30:15,303 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.4488, 4.9281, 4.7708, 4.9439, 4.7832, 4.7533, 5.3656, 4.9445], device='cuda:1'), covar=tensor([0.0686, 0.1285, 0.1988, 0.2178, 0.0954, 0.1578, 0.0875, 0.1121], device='cuda:1'), in_proj_covar=tensor([0.0356, 0.0405, 0.0499, 0.0502, 0.0321, 0.0374, 0.0463, 0.0377], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:30:55,835 INFO [train.py:893] (1/4) Epoch 29, batch 1050, loss[loss=0.1367, simple_loss=0.195, pruned_loss=0.03921, over 13184.00 frames. ], tot_loss[loss=0.1636, simple_loss=0.2258, pruned_loss=0.05067, over 2651109.75 frames. ], batch size: 58, lr: 4.49e-03, grad_scale: 16.0 2023-04-17 07:30:56,021 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76625.0, num_to_drop=1, layers_to_drop={2} 2023-04-17 07:30:57,836 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7462, 3.4277, 4.2860, 2.9868, 2.8004, 2.9194, 4.5417, 4.6314], device='cuda:1'), covar=tensor([0.1082, 0.1857, 0.0317, 0.1919, 0.1734, 0.1660, 0.0243, 0.0258], device='cuda:1'), in_proj_covar=tensor([0.0258, 0.0285, 0.0208, 0.0230, 0.0226, 0.0190, 0.0223, 0.0199], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:30:58,260 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.799e+02 2.390e+02 2.885e+02 3.502e+02 6.067e+02, threshold=5.770e+02, percent-clipped=0.0 2023-04-17 07:31:17,257 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76650.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:31:40,181 INFO [train.py:893] (1/4) Epoch 29, batch 1100, loss[loss=0.1525, simple_loss=0.2236, pruned_loss=0.04075, over 13535.00 frames. ], tot_loss[loss=0.1639, simple_loss=0.2267, pruned_loss=0.05059, over 2652844.07 frames. ], batch size: 91, lr: 4.49e-03, grad_scale: 16.0 2023-04-17 07:31:43,799 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=76679.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 07:32:26,122 INFO [train.py:893] (1/4) Epoch 29, batch 1150, loss[loss=0.1679, simple_loss=0.2348, pruned_loss=0.05052, over 13171.00 frames. ], tot_loss[loss=0.1641, simple_loss=0.2271, pruned_loss=0.05053, over 2654438.40 frames. ], batch size: 132, lr: 4.49e-03, grad_scale: 16.0 2023-04-17 07:32:28,190 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1277, 4.2119, 3.0278, 3.8218, 4.0606, 2.7617, 3.7992, 2.8399], device='cuda:1'), covar=tensor([0.0294, 0.0204, 0.1008, 0.0393, 0.0296, 0.1200, 0.0452, 0.1173], device='cuda:1'), in_proj_covar=tensor([0.0171, 0.0188, 0.0181, 0.0240, 0.0145, 0.0165, 0.0166, 0.0163], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:32:28,619 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.716e+02 2.373e+02 2.855e+02 3.407e+02 7.423e+02, threshold=5.711e+02, percent-clipped=5.0 2023-04-17 07:33:11,208 INFO [train.py:893] (1/4) Epoch 29, batch 1200, loss[loss=0.1645, simple_loss=0.2279, pruned_loss=0.05056, over 13375.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2275, pruned_loss=0.05029, over 2656076.19 frames. ], batch size: 113, lr: 4.49e-03, grad_scale: 16.0 2023-04-17 07:33:37,413 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4240, 3.7640, 3.5448, 4.3305, 2.2146, 2.8280, 3.8591, 2.3314], device='cuda:1'), covar=tensor([0.0127, 0.0490, 0.0797, 0.0394, 0.1761, 0.1246, 0.0570, 0.1704], device='cuda:1'), in_proj_covar=tensor([0.0144, 0.0193, 0.0212, 0.0258, 0.0188, 0.0206, 0.0183, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:33:39,316 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-17 07:33:50,789 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-17 07:33:53,325 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4397, 4.7451, 4.4135, 4.5004, 4.5454, 4.8562, 4.6676, 4.5391], device='cuda:1'), covar=tensor([0.0304, 0.0262, 0.0375, 0.0828, 0.0280, 0.0245, 0.0328, 0.0198], device='cuda:1'), in_proj_covar=tensor([0.0176, 0.0169, 0.0193, 0.0278, 0.0191, 0.0209, 0.0187, 0.0116], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 07:33:56,391 INFO [train.py:893] (1/4) Epoch 29, batch 1250, loss[loss=0.1685, simple_loss=0.2355, pruned_loss=0.05068, over 13531.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2281, pruned_loss=0.05084, over 2654932.06 frames. ], batch size: 72, lr: 4.49e-03, grad_scale: 16.0 2023-04-17 07:33:58,835 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.935e+02 2.494e+02 2.806e+02 3.320e+02 5.743e+02, threshold=5.612e+02, percent-clipped=1.0 2023-04-17 07:34:08,842 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=76839.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:34:36,741 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 2023-04-17 07:34:41,807 INFO [train.py:893] (1/4) Epoch 29, batch 1300, loss[loss=0.1963, simple_loss=0.2527, pruned_loss=0.06999, over 13535.00 frames. ], tot_loss[loss=0.1663, simple_loss=0.2295, pruned_loss=0.05152, over 2655738.47 frames. ], batch size: 87, lr: 4.49e-03, grad_scale: 16.0 2023-04-17 07:35:04,194 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=76900.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:35:26,454 INFO [train.py:893] (1/4) Epoch 29, batch 1350, loss[loss=0.1768, simple_loss=0.2416, pruned_loss=0.05596, over 11819.00 frames. ], tot_loss[loss=0.1667, simple_loss=0.2304, pruned_loss=0.05148, over 2659793.52 frames. ], batch size: 158, lr: 4.48e-03, grad_scale: 16.0 2023-04-17 07:35:26,673 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76925.0, num_to_drop=1, layers_to_drop={2} 2023-04-17 07:35:29,570 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.832e+02 2.606e+02 2.968e+02 3.524e+02 4.932e+02, threshold=5.935e+02, percent-clipped=0.0 2023-04-17 07:35:50,022 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76950.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:35:50,808 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.1500, 4.9672, 5.1837, 4.9902, 5.4417, 4.9833, 5.4511, 5.3741], device='cuda:1'), covar=tensor([0.0394, 0.0538, 0.0603, 0.0540, 0.0477, 0.0829, 0.0397, 0.0479], device='cuda:1'), in_proj_covar=tensor([0.0319, 0.0316, 0.0330, 0.0246, 0.0469, 0.0368, 0.0308, 0.0326], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-04-17 07:36:10,279 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=76973.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 07:36:11,675 INFO [train.py:893] (1/4) Epoch 29, batch 1400, loss[loss=0.1759, simple_loss=0.2281, pruned_loss=0.06186, over 13242.00 frames. ], tot_loss[loss=0.1668, simple_loss=0.2302, pruned_loss=0.05165, over 2658332.55 frames. ], batch size: 58, lr: 4.48e-03, grad_scale: 16.0 2023-04-17 07:36:15,990 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=76979.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 07:36:33,349 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=76998.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:36:57,268 INFO [train.py:893] (1/4) Epoch 29, batch 1450, loss[loss=0.1651, simple_loss=0.229, pruned_loss=0.05063, over 13543.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2299, pruned_loss=0.05161, over 2662755.87 frames. ], batch size: 87, lr: 4.48e-03, grad_scale: 16.0 2023-04-17 07:36:59,086 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77027.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 07:37:00,375 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.817e+02 2.615e+02 3.065e+02 3.635e+02 5.318e+02, threshold=6.130e+02, percent-clipped=0.0 2023-04-17 07:37:43,765 INFO [train.py:893] (1/4) Epoch 29, batch 1500, loss[loss=0.1617, simple_loss=0.2289, pruned_loss=0.04719, over 13519.00 frames. ], tot_loss[loss=0.1662, simple_loss=0.2298, pruned_loss=0.05133, over 2662492.00 frames. ], batch size: 70, lr: 4.48e-03, grad_scale: 32.0 2023-04-17 07:38:02,570 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77095.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:38:03,418 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.8429, 3.6478, 2.9857, 3.3646, 2.9956, 2.3406, 3.7907, 2.2992], device='cuda:1'), covar=tensor([0.0674, 0.0581, 0.0479, 0.0402, 0.0709, 0.1818, 0.0871, 0.1242], device='cuda:1'), in_proj_covar=tensor([0.0148, 0.0153, 0.0143, 0.0123, 0.0155, 0.0195, 0.0192, 0.0150], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 07:38:29,877 INFO [train.py:893] (1/4) Epoch 29, batch 1550, loss[loss=0.1852, simple_loss=0.2488, pruned_loss=0.06073, over 13397.00 frames. ], tot_loss[loss=0.1664, simple_loss=0.23, pruned_loss=0.05142, over 2659845.25 frames. ], batch size: 113, lr: 4.48e-03, grad_scale: 32.0 2023-04-17 07:38:32,484 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.779e+02 2.614e+02 2.982e+02 3.394e+02 6.130e+02, threshold=5.963e+02, percent-clipped=1.0 2023-04-17 07:38:42,669 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5826, 3.9823, 3.6521, 4.4104, 2.4126, 3.2755, 4.1265, 2.4547], device='cuda:1'), covar=tensor([0.0244, 0.0461, 0.0929, 0.0635, 0.1685, 0.1051, 0.0448, 0.1902], device='cuda:1'), in_proj_covar=tensor([0.0145, 0.0192, 0.0212, 0.0258, 0.0187, 0.0205, 0.0183, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:38:46,719 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5093, 4.6614, 3.3809, 4.2042, 4.4608, 3.1133, 4.0324, 3.4239], device='cuda:1'), covar=tensor([0.0271, 0.0229, 0.0900, 0.0506, 0.0180, 0.1090, 0.0374, 0.1052], device='cuda:1'), in_proj_covar=tensor([0.0171, 0.0188, 0.0180, 0.0239, 0.0145, 0.0164, 0.0166, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:38:58,509 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77156.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 07:39:14,433 INFO [train.py:893] (1/4) Epoch 29, batch 1600, loss[loss=0.1685, simple_loss=0.2389, pruned_loss=0.04904, over 13453.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2304, pruned_loss=0.05137, over 2661537.93 frames. ], batch size: 79, lr: 4.48e-03, grad_scale: 32.0 2023-04-17 07:39:33,767 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77195.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:39:48,459 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77211.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:40:01,913 INFO [train.py:893] (1/4) Epoch 29, batch 1650, loss[loss=0.1621, simple_loss=0.2205, pruned_loss=0.05184, over 13416.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2296, pruned_loss=0.05034, over 2663134.29 frames. ], batch size: 65, lr: 4.48e-03, grad_scale: 32.0 2023-04-17 07:40:04,411 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.838e+02 2.449e+02 2.816e+02 3.387e+02 9.099e+02, threshold=5.633e+02, percent-clipped=2.0 2023-04-17 07:40:43,885 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77272.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:40:44,639 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77273.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:40:46,620 INFO [train.py:893] (1/4) Epoch 29, batch 1700, loss[loss=0.1656, simple_loss=0.2347, pruned_loss=0.04828, over 13489.00 frames. ], tot_loss[loss=0.1648, simple_loss=0.2295, pruned_loss=0.05001, over 2668555.16 frames. ], batch size: 81, lr: 4.47e-03, grad_scale: 32.0 2023-04-17 07:40:49,154 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77277.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:41:15,548 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-04-17 07:41:32,660 INFO [train.py:893] (1/4) Epoch 29, batch 1750, loss[loss=0.1379, simple_loss=0.2012, pruned_loss=0.03733, over 13374.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.229, pruned_loss=0.04982, over 2668699.76 frames. ], batch size: 62, lr: 4.47e-03, grad_scale: 32.0 2023-04-17 07:41:35,060 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.926e+02 2.374e+02 2.741e+02 3.275e+02 6.260e+02, threshold=5.482e+02, percent-clipped=1.0 2023-04-17 07:41:40,278 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77334.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:41:44,165 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77338.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:42:17,494 INFO [train.py:893] (1/4) Epoch 29, batch 1800, loss[loss=0.1697, simple_loss=0.2381, pruned_loss=0.05063, over 13466.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.2283, pruned_loss=0.0494, over 2668690.17 frames. ], batch size: 103, lr: 4.47e-03, grad_scale: 32.0 2023-04-17 07:42:20,135 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.9110, 3.6569, 2.9841, 3.4926, 3.7497, 2.6972, 3.5952, 2.5263], device='cuda:1'), covar=tensor([0.0239, 0.0189, 0.0813, 0.0332, 0.0259, 0.0999, 0.0414, 0.1138], device='cuda:1'), in_proj_covar=tensor([0.0171, 0.0189, 0.0180, 0.0239, 0.0146, 0.0164, 0.0165, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:42:30,039 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.8196, 4.2089, 3.9040, 4.6434, 2.5863, 3.6430, 4.4370, 2.7847], device='cuda:1'), covar=tensor([0.0160, 0.0394, 0.0813, 0.0617, 0.1637, 0.0876, 0.0364, 0.1516], device='cuda:1'), in_proj_covar=tensor([0.0145, 0.0192, 0.0212, 0.0258, 0.0187, 0.0205, 0.0182, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:42:34,346 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.94 vs. limit=2.0 2023-04-17 07:43:02,751 INFO [train.py:893] (1/4) Epoch 29, batch 1850, loss[loss=0.186, simple_loss=0.2517, pruned_loss=0.06016, over 13532.00 frames. ], tot_loss[loss=0.1634, simple_loss=0.2281, pruned_loss=0.04933, over 2665097.88 frames. ], batch size: 98, lr: 4.47e-03, grad_scale: 32.0 2023-04-17 07:43:03,054 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7055, 3.4715, 2.7821, 3.0378, 2.8249, 2.2668, 3.5828, 2.1299], device='cuda:1'), covar=tensor([0.0768, 0.0527, 0.0504, 0.0511, 0.0701, 0.1821, 0.0859, 0.1340], device='cuda:1'), in_proj_covar=tensor([0.0149, 0.0153, 0.0142, 0.0124, 0.0155, 0.0196, 0.0192, 0.0151], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 07:43:05,115 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.962e+02 2.432e+02 2.941e+02 3.505e+02 5.866e+02, threshold=5.882e+02, percent-clipped=2.0 2023-04-17 07:43:06,709 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-17 07:43:26,596 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77451.0, num_to_drop=1, layers_to_drop={2} 2023-04-17 07:43:36,504 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 2023-04-17 07:43:48,648 INFO [train.py:893] (1/4) Epoch 29, batch 1900, loss[loss=0.157, simple_loss=0.2206, pruned_loss=0.04668, over 11972.00 frames. ], tot_loss[loss=0.1633, simple_loss=0.2274, pruned_loss=0.04964, over 2660958.14 frames. ], batch size: 157, lr: 4.47e-03, grad_scale: 32.0 2023-04-17 07:43:56,559 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8234, 2.9966, 3.4451, 4.3832, 3.8686, 4.4004, 3.6371, 3.0762], device='cuda:1'), covar=tensor([0.0366, 0.0842, 0.0623, 0.0055, 0.0261, 0.0070, 0.0569, 0.0772], device='cuda:1'), in_proj_covar=tensor([0.0129, 0.0154, 0.0172, 0.0109, 0.0128, 0.0106, 0.0176, 0.0153], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:43:58,198 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.7682, 4.1724, 3.8953, 4.5692, 2.6221, 3.6038, 4.4015, 2.6228], device='cuda:1'), covar=tensor([0.0203, 0.0444, 0.0758, 0.0477, 0.1545, 0.0886, 0.0364, 0.1699], device='cuda:1'), in_proj_covar=tensor([0.0145, 0.0193, 0.0211, 0.0258, 0.0187, 0.0205, 0.0183, 0.0179], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:44:06,111 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77495.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:44:33,587 INFO [train.py:893] (1/4) Epoch 29, batch 1950, loss[loss=0.1759, simple_loss=0.2385, pruned_loss=0.05665, over 13522.00 frames. ], tot_loss[loss=0.1627, simple_loss=0.2265, pruned_loss=0.04941, over 2661118.68 frames. ], batch size: 98, lr: 4.47e-03, grad_scale: 32.0 2023-04-17 07:44:36,672 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.998e+02 2.561e+02 3.002e+02 3.542e+02 6.432e+02, threshold=6.003e+02, percent-clipped=1.0 2023-04-17 07:44:50,766 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77543.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:45:12,491 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77567.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:45:18,913 INFO [train.py:893] (1/4) Epoch 29, batch 2000, loss[loss=0.1895, simple_loss=0.2473, pruned_loss=0.06579, over 13445.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2283, pruned_loss=0.05007, over 2664021.73 frames. ], batch size: 106, lr: 4.47e-03, grad_scale: 32.0 2023-04-17 07:45:23,987 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-17 07:45:30,555 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7207, 3.8277, 2.6477, 3.2586, 2.9362, 2.1691, 3.9304, 2.0923], device='cuda:1'), covar=tensor([0.0931, 0.0467, 0.0885, 0.0542, 0.0809, 0.2460, 0.0852, 0.1707], device='cuda:1'), in_proj_covar=tensor([0.0148, 0.0153, 0.0142, 0.0124, 0.0154, 0.0195, 0.0191, 0.0150], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 07:45:55,247 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77616.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:46:03,103 INFO [train.py:893] (1/4) Epoch 29, batch 2050, loss[loss=0.1452, simple_loss=0.214, pruned_loss=0.0382, over 13498.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2293, pruned_loss=0.05024, over 2664079.09 frames. ], batch size: 70, lr: 4.46e-03, grad_scale: 32.0 2023-04-17 07:46:06,323 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.846e+02 2.483e+02 2.949e+02 3.550e+02 5.640e+02, threshold=5.898e+02, percent-clipped=0.0 2023-04-17 07:46:07,365 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77629.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:46:10,640 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77633.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:46:47,891 INFO [train.py:893] (1/4) Epoch 29, batch 2100, loss[loss=0.1442, simple_loss=0.2112, pruned_loss=0.0386, over 13335.00 frames. ], tot_loss[loss=0.1653, simple_loss=0.2295, pruned_loss=0.05051, over 2661461.94 frames. ], batch size: 67, lr: 4.46e-03, grad_scale: 32.0 2023-04-17 07:46:49,842 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=77677.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:46:49,860 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77677.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:47:10,070 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5449, 3.9425, 2.7128, 3.6501, 3.8676, 2.6361, 3.3050, 3.0123], device='cuda:1'), covar=tensor([0.0520, 0.0468, 0.1044, 0.0344, 0.0289, 0.1134, 0.0675, 0.0905], device='cuda:1'), in_proj_covar=tensor([0.0172, 0.0190, 0.0181, 0.0240, 0.0146, 0.0164, 0.0167, 0.0163], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:47:34,416 INFO [train.py:893] (1/4) Epoch 29, batch 2150, loss[loss=0.1752, simple_loss=0.2457, pruned_loss=0.05237, over 13492.00 frames. ], tot_loss[loss=0.1651, simple_loss=0.2297, pruned_loss=0.05026, over 2662989.88 frames. ], batch size: 81, lr: 4.46e-03, grad_scale: 32.0 2023-04-17 07:47:36,702 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.954e+02 2.489e+02 2.950e+02 3.593e+02 6.767e+02, threshold=5.899e+02, percent-clipped=1.0 2023-04-17 07:47:45,409 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=77738.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:47:47,138 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.96 vs. limit=2.0 2023-04-17 07:47:54,893 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0880, 3.9333, 4.0223, 2.4527, 4.3693, 4.1489, 4.0975, 4.3550], device='cuda:1'), covar=tensor([0.0280, 0.0158, 0.0146, 0.1221, 0.0158, 0.0280, 0.0162, 0.0108], device='cuda:1'), in_proj_covar=tensor([0.0102, 0.0063, 0.0092, 0.0109, 0.0108, 0.0121, 0.0090, 0.0085], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:47:57,993 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77751.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 07:47:59,109 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.35 vs. limit=2.0 2023-04-17 07:48:19,626 INFO [train.py:893] (1/4) Epoch 29, batch 2200, loss[loss=0.1618, simple_loss=0.2247, pruned_loss=0.04941, over 13375.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.2291, pruned_loss=0.05016, over 2659039.81 frames. ], batch size: 73, lr: 4.46e-03, grad_scale: 32.0 2023-04-17 07:48:41,186 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77799.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:49:05,599 INFO [train.py:893] (1/4) Epoch 29, batch 2250, loss[loss=0.1611, simple_loss=0.2283, pruned_loss=0.04696, over 13435.00 frames. ], tot_loss[loss=0.1627, simple_loss=0.2269, pruned_loss=0.04929, over 2660583.79 frames. ], batch size: 103, lr: 4.46e-03, grad_scale: 32.0 2023-04-17 07:49:08,043 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.824e+02 2.331e+02 2.696e+02 3.278e+02 5.521e+02, threshold=5.391e+02, percent-clipped=0.0 2023-04-17 07:49:20,761 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.76 vs. limit=2.0 2023-04-17 07:49:32,878 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6603, 2.4160, 2.3902, 2.7395, 2.0793, 2.7387, 2.7361, 2.2699], device='cuda:1'), covar=tensor([0.0089, 0.0219, 0.0166, 0.0153, 0.0240, 0.0149, 0.0165, 0.0233], device='cuda:1'), in_proj_covar=tensor([0.0115, 0.0130, 0.0134, 0.0136, 0.0145, 0.0124, 0.0117, 0.0121], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 07:49:43,214 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77867.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:49:50,168 INFO [train.py:893] (1/4) Epoch 29, batch 2300, loss[loss=0.1819, simple_loss=0.2449, pruned_loss=0.05941, over 13247.00 frames. ], tot_loss[loss=0.1625, simple_loss=0.2267, pruned_loss=0.04912, over 2661566.70 frames. ], batch size: 124, lr: 4.46e-03, grad_scale: 32.0 2023-04-17 07:50:27,413 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77915.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:50:35,384 INFO [train.py:893] (1/4) Epoch 29, batch 2350, loss[loss=0.1265, simple_loss=0.19, pruned_loss=0.03149, over 13484.00 frames. ], tot_loss[loss=0.1626, simple_loss=0.2269, pruned_loss=0.04918, over 2665585.33 frames. ], batch size: 70, lr: 4.46e-03, grad_scale: 32.0 2023-04-17 07:50:38,563 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.953e+02 2.391e+02 2.757e+02 3.275e+02 5.484e+02, threshold=5.514e+02, percent-clipped=1.0 2023-04-17 07:50:39,657 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77929.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:50:42,795 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=77933.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:50:59,339 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-17 07:51:08,395 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.1706, 4.3017, 3.1649, 3.8194, 3.3800, 2.4839, 4.3450, 2.4585], device='cuda:1'), covar=tensor([0.0736, 0.0322, 0.0755, 0.0302, 0.0655, 0.2171, 0.0645, 0.1507], device='cuda:1'), in_proj_covar=tensor([0.0149, 0.0153, 0.0142, 0.0124, 0.0155, 0.0196, 0.0193, 0.0150], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 07:51:18,169 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=77972.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:51:20,415 INFO [train.py:893] (1/4) Epoch 29, batch 2400, loss[loss=0.1526, simple_loss=0.2185, pruned_loss=0.04333, over 13511.00 frames. ], tot_loss[loss=0.162, simple_loss=0.2261, pruned_loss=0.04896, over 2668299.98 frames. ], batch size: 91, lr: 4.45e-03, grad_scale: 32.0 2023-04-17 07:51:23,584 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77977.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:51:26,989 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=77981.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:51:27,979 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7030, 2.5703, 3.1245, 4.2111, 3.7400, 4.2999, 3.3373, 2.7960], device='cuda:1'), covar=tensor([0.0288, 0.0978, 0.0765, 0.0066, 0.0295, 0.0063, 0.0718, 0.0836], device='cuda:1'), in_proj_covar=tensor([0.0129, 0.0154, 0.0173, 0.0109, 0.0129, 0.0106, 0.0176, 0.0153], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0002, 0.0002, 0.0002, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:52:10,233 INFO [train.py:893] (1/4) Epoch 29, batch 2450, loss[loss=0.1749, simple_loss=0.2362, pruned_loss=0.05677, over 13080.00 frames. ], tot_loss[loss=0.1625, simple_loss=0.2264, pruned_loss=0.04927, over 2668646.50 frames. ], batch size: 142, lr: 4.45e-03, grad_scale: 32.0 2023-04-17 07:52:12,746 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.799e+02 2.344e+02 2.892e+02 3.560e+02 6.451e+02, threshold=5.784e+02, percent-clipped=2.0 2023-04-17 07:52:17,968 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78033.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:52:56,052 INFO [train.py:893] (1/4) Epoch 29, batch 2500, loss[loss=0.158, simple_loss=0.2215, pruned_loss=0.04724, over 13513.00 frames. ], tot_loss[loss=0.1628, simple_loss=0.2269, pruned_loss=0.04937, over 2671049.48 frames. ], batch size: 70, lr: 4.45e-03, grad_scale: 32.0 2023-04-17 07:53:07,637 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5761, 3.9972, 3.8119, 4.3828, 2.5309, 3.3157, 4.1541, 2.4623], device='cuda:1'), covar=tensor([0.0205, 0.0385, 0.0817, 0.0504, 0.1587, 0.0970, 0.0420, 0.1711], device='cuda:1'), in_proj_covar=tensor([0.0146, 0.0194, 0.0212, 0.0260, 0.0188, 0.0206, 0.0185, 0.0180], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:53:41,531 INFO [train.py:893] (1/4) Epoch 29, batch 2550, loss[loss=0.1701, simple_loss=0.2372, pruned_loss=0.05152, over 13558.00 frames. ], tot_loss[loss=0.1632, simple_loss=0.2271, pruned_loss=0.04963, over 2669429.36 frames. ], batch size: 89, lr: 4.45e-03, grad_scale: 32.0 2023-04-17 07:53:43,946 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.828e+02 2.449e+02 2.838e+02 3.334e+02 6.416e+02, threshold=5.677e+02, percent-clipped=2.0 2023-04-17 07:53:52,147 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.5317, 4.8520, 4.5190, 4.5594, 4.6109, 4.9146, 4.7491, 4.6151], device='cuda:1'), covar=tensor([0.0300, 0.0239, 0.0314, 0.0826, 0.0275, 0.0223, 0.0289, 0.0209], device='cuda:1'), in_proj_covar=tensor([0.0174, 0.0168, 0.0191, 0.0274, 0.0190, 0.0207, 0.0186, 0.0115], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 07:54:05,001 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-17 07:54:24,811 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6387, 3.6661, 2.7844, 3.2658, 3.6224, 2.4754, 3.3215, 2.5298], device='cuda:1'), covar=tensor([0.0325, 0.0233, 0.0959, 0.0387, 0.0282, 0.1176, 0.0550, 0.1229], device='cuda:1'), in_proj_covar=tensor([0.0172, 0.0188, 0.0181, 0.0239, 0.0145, 0.0164, 0.0166, 0.0163], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:54:26,956 INFO [train.py:893] (1/4) Epoch 29, batch 2600, loss[loss=0.1355, simple_loss=0.1822, pruned_loss=0.0444, over 9316.00 frames. ], tot_loss[loss=0.1628, simple_loss=0.2265, pruned_loss=0.04955, over 2663717.55 frames. ], batch size: 37, lr: 4.45e-03, grad_scale: 32.0 2023-04-17 07:54:28,420 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.71 vs. limit=5.0 2023-04-17 07:54:33,733 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3761, 4.5532, 3.1589, 3.9862, 4.3783, 2.9372, 3.9743, 3.0508], device='cuda:1'), covar=tensor([0.0305, 0.0249, 0.1018, 0.0495, 0.0219, 0.1227, 0.0394, 0.1497], device='cuda:1'), in_proj_covar=tensor([0.0172, 0.0189, 0.0181, 0.0239, 0.0145, 0.0164, 0.0166, 0.0164], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:54:59,949 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 2023-04-17 07:55:08,656 INFO [train.py:893] (1/4) Epoch 29, batch 2650, loss[loss=0.1515, simple_loss=0.215, pruned_loss=0.04403, over 13512.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2274, pruned_loss=0.05025, over 2660736.63 frames. ], batch size: 70, lr: 4.45e-03, grad_scale: 32.0 2023-04-17 07:55:10,857 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.961e+02 2.802e+02 3.099e+02 3.581e+02 5.823e+02, threshold=6.198e+02, percent-clipped=1.0 2023-04-17 07:55:13,879 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78232.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:55:44,001 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78272.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:56:06,338 WARNING [train.py:1054] (1/4) Exclude cut with ID db3105debd557279a777dee8a84ecb323b3b63f83e1d939111fe3f232d72bf04767da13a8383344d428a5a9ad767c044bce3837e3623d98f38591cac317d2932-common_voice_en_21838523 from training. Duration: 21.6 2023-04-17 07:56:14,651 INFO [train.py:893] (1/4) Epoch 30, batch 0, loss[loss=0.1784, simple_loss=0.2382, pruned_loss=0.05931, over 13234.00 frames. ], tot_loss[loss=0.1784, simple_loss=0.2382, pruned_loss=0.05931, over 13234.00 frames. ], batch size: 132, lr: 4.37e-03, grad_scale: 32.0 2023-04-17 07:56:14,652 INFO [train.py:918] (1/4) Computing validation loss 2023-04-17 07:56:20,173 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0623, 4.3437, 3.1606, 2.8957, 2.9904, 2.5848, 4.4753, 2.4508], device='cuda:1'), covar=tensor([0.2002, 0.0380, 0.1673, 0.2706, 0.1030, 0.3987, 0.0270, 0.5021], device='cuda:1'), in_proj_covar=tensor([0.0324, 0.0305, 0.0340, 0.0362, 0.0275, 0.0347, 0.0233, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 07:56:34,672 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4936, 3.6307, 4.6208, 4.2592, 3.5009, 3.3965, 4.9481, 5.1138], device='cuda:1'), covar=tensor([0.0789, 0.1657, 0.0328, 0.0930, 0.1180, 0.1505, 0.0149, 0.0302], device='cuda:1'), in_proj_covar=tensor([0.0259, 0.0286, 0.0209, 0.0231, 0.0227, 0.0192, 0.0224, 0.0200], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:56:36,838 INFO [train.py:927] (1/4) Epoch 30, validation: loss=0.1309, simple_loss=0.1907, pruned_loss=0.03552, over 2446609.00 frames. 2023-04-17 07:56:36,839 INFO [train.py:928] (1/4) Maximum memory allocated so far is 12831MB 2023-04-17 07:56:54,953 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78293.0, num_to_drop=1, layers_to_drop={3} 2023-04-17 07:56:56,484 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0546, 4.4691, 4.2971, 4.2762, 4.3237, 4.1001, 4.5413, 4.5442], device='cuda:1'), covar=tensor([0.0250, 0.0233, 0.0274, 0.0362, 0.0305, 0.0314, 0.0247, 0.0269], device='cuda:1'), in_proj_covar=tensor([0.0154, 0.0218, 0.0183, 0.0200, 0.0169, 0.0220, 0.0148, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-04-17 07:57:19,893 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78320.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:57:23,062 INFO [train.py:893] (1/4) Epoch 30, batch 50, loss[loss=0.1572, simple_loss=0.2171, pruned_loss=0.04866, over 13534.00 frames. ], tot_loss[loss=0.1597, simple_loss=0.2209, pruned_loss=0.04919, over 597102.80 frames. ], batch size: 72, lr: 4.37e-03, grad_scale: 32.0 2023-04-17 07:57:23,415 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7795, 3.8283, 2.8117, 3.4647, 3.7535, 2.4814, 3.4864, 2.6661], device='cuda:1'), covar=tensor([0.0311, 0.0241, 0.0966, 0.0453, 0.0280, 0.1180, 0.0477, 0.1236], device='cuda:1'), in_proj_covar=tensor([0.0172, 0.0189, 0.0182, 0.0239, 0.0146, 0.0164, 0.0166, 0.0164], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:57:26,331 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.743e+02 2.510e+02 3.075e+02 3.612e+02 7.972e+02, threshold=6.151e+02, percent-clipped=2.0 2023-04-17 07:57:31,304 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78333.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:57:47,346 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233310 from training. Duration: 33.876 2023-04-17 07:57:47,346 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233451 from training. Duration: 20.556 2023-04-17 07:57:47,346 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27233710 from training. Duration: 49.932 2023-04-17 07:57:47,353 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27289387 from training. Duration: 27.792 2023-04-17 07:57:47,367 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27312193 from training. Duration: 23.22 2023-04-17 07:57:47,387 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27367777 from training. Duration: 27.0 2023-04-17 07:57:48,120 WARNING [train.py:1054] (1/4) Exclude cut with ID 416dc03bdc0085ce8ea41524449ef5421eed43264a1c42a4fc14300111a73fbe13396d018c6972fa5ec859c678bb15e077675dc399e12f637812c624d6e993dc-common_voice_en_27371830 from training. Duration: 23.688 2023-04-17 07:58:08,160 INFO [train.py:893] (1/4) Epoch 30, batch 100, loss[loss=0.1648, simple_loss=0.2252, pruned_loss=0.05218, over 13174.00 frames. ], tot_loss[loss=0.163, simple_loss=0.2239, pruned_loss=0.05102, over 1053010.37 frames. ], batch size: 132, lr: 4.37e-03, grad_scale: 32.0 2023-04-17 07:58:11,488 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3736, 4.1375, 4.1956, 2.6647, 4.6218, 4.3482, 4.3444, 4.5537], device='cuda:1'), covar=tensor([0.0219, 0.0134, 0.0151, 0.0991, 0.0134, 0.0266, 0.0144, 0.0102], device='cuda:1'), in_proj_covar=tensor([0.0102, 0.0063, 0.0091, 0.0108, 0.0108, 0.0121, 0.0090, 0.0085], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 07:58:15,423 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78381.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:58:54,010 INFO [train.py:893] (1/4) Epoch 30, batch 150, loss[loss=0.1471, simple_loss=0.2196, pruned_loss=0.03732, over 13451.00 frames. ], tot_loss[loss=0.1647, simple_loss=0.226, pruned_loss=0.05171, over 1399701.33 frames. ], batch size: 106, lr: 4.37e-03, grad_scale: 32.0 2023-04-17 07:58:58,106 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.972e+02 2.511e+02 2.997e+02 3.677e+02 5.041e+02, threshold=5.994e+02, percent-clipped=0.0 2023-04-17 07:59:27,679 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78460.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 07:59:40,912 INFO [train.py:893] (1/4) Epoch 30, batch 200, loss[loss=0.1592, simple_loss=0.2253, pruned_loss=0.04661, over 13529.00 frames. ], tot_loss[loss=0.1656, simple_loss=0.2274, pruned_loss=0.05193, over 1670486.25 frames. ], batch size: 76, lr: 4.36e-03, grad_scale: 32.0 2023-04-17 08:00:07,782 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.78 vs. limit=5.0 2023-04-17 08:00:23,028 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78521.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:00:25,835 INFO [train.py:893] (1/4) Epoch 30, batch 250, loss[loss=0.1627, simple_loss=0.2205, pruned_loss=0.05249, over 13197.00 frames. ], tot_loss[loss=0.166, simple_loss=0.2281, pruned_loss=0.05192, over 1891679.28 frames. ], batch size: 132, lr: 4.36e-03, grad_scale: 32.0 2023-04-17 08:00:27,066 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.5909, 3.3422, 3.9999, 2.8979, 2.7546, 2.8071, 4.3262, 4.4490], device='cuda:1'), covar=tensor([0.1186, 0.1791, 0.0390, 0.1778, 0.1578, 0.1614, 0.0286, 0.0221], device='cuda:1'), in_proj_covar=tensor([0.0259, 0.0287, 0.0209, 0.0232, 0.0228, 0.0192, 0.0224, 0.0201], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 08:00:29,093 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.795e+02 2.445e+02 2.983e+02 3.638e+02 6.878e+02, threshold=5.966e+02, percent-clipped=2.0 2023-04-17 08:00:43,416 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.8012, 4.6252, 4.8589, 4.7510, 5.1081, 4.6184, 5.1160, 5.0780], device='cuda:1'), covar=tensor([0.0439, 0.0543, 0.0618, 0.0541, 0.0536, 0.0819, 0.0459, 0.0409], device='cuda:1'), in_proj_covar=tensor([0.0320, 0.0320, 0.0333, 0.0247, 0.0475, 0.0371, 0.0310, 0.0328], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-04-17 08:00:54,651 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78556.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:01:11,323 INFO [train.py:893] (1/4) Epoch 30, batch 300, loss[loss=0.1648, simple_loss=0.2304, pruned_loss=0.04962, over 11810.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2279, pruned_loss=0.05212, over 2054343.39 frames. ], batch size: 157, lr: 4.36e-03, grad_scale: 32.0 2023-04-17 08:01:14,069 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.1318, 4.6690, 4.5179, 4.6866, 4.4516, 4.5029, 5.0825, 4.6547], device='cuda:1'), covar=tensor([0.0675, 0.1234, 0.2225, 0.2194, 0.1036, 0.1550, 0.0792, 0.1205], device='cuda:1'), in_proj_covar=tensor([0.0361, 0.0412, 0.0508, 0.0509, 0.0327, 0.0381, 0.0470, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 08:01:23,808 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78588.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 08:01:50,870 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=78617.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:01:54,998 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.0007, 4.8586, 5.0674, 4.8953, 5.3099, 4.7750, 5.3153, 5.2830], device='cuda:1'), covar=tensor([0.0448, 0.0561, 0.0635, 0.0606, 0.0517, 0.0877, 0.0460, 0.0465], device='cuda:1'), in_proj_covar=tensor([0.0320, 0.0320, 0.0333, 0.0247, 0.0477, 0.0371, 0.0311, 0.0329], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-04-17 08:01:56,448 INFO [train.py:893] (1/4) Epoch 30, batch 350, loss[loss=0.1421, simple_loss=0.2124, pruned_loss=0.03592, over 13555.00 frames. ], tot_loss[loss=0.1665, simple_loss=0.2285, pruned_loss=0.0523, over 2184712.59 frames. ], batch size: 78, lr: 4.36e-03, grad_scale: 32.0 2023-04-17 08:02:00,472 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.006e+02 2.600e+02 2.996e+02 3.438e+02 5.374e+02, threshold=5.992e+02, percent-clipped=0.0 2023-04-17 08:02:42,442 INFO [train.py:893] (1/4) Epoch 30, batch 400, loss[loss=0.1828, simple_loss=0.2297, pruned_loss=0.06798, over 13407.00 frames. ], tot_loss[loss=0.1659, simple_loss=0.2282, pruned_loss=0.0518, over 2294717.50 frames. ], batch size: 62, lr: 4.36e-03, grad_scale: 32.0 2023-04-17 08:03:27,771 INFO [train.py:893] (1/4) Epoch 30, batch 450, loss[loss=0.1871, simple_loss=0.2478, pruned_loss=0.06313, over 13257.00 frames. ], tot_loss[loss=0.1673, simple_loss=0.2304, pruned_loss=0.05213, over 2379728.19 frames. ], batch size: 124, lr: 4.36e-03, grad_scale: 32.0 2023-04-17 08:03:32,442 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.010e+02 2.627e+02 2.942e+02 3.377e+02 5.601e+02, threshold=5.884e+02, percent-clipped=0.0 2023-04-17 08:03:53,002 WARNING [train.py:1054] (1/4) Exclude cut with ID c3c070732f2c104d269cc54d36a8684bdb15120f468e145af8bc6bcd0d942d561341b6d9e6c6f1cdd7f1a56d60c8536c3afea58ec24eaf7eb9629a8f96995f2d-common_voice_en_619791 from training. Duration: 20.496 2023-04-17 08:04:13,296 INFO [train.py:893] (1/4) Epoch 30, batch 500, loss[loss=0.155, simple_loss=0.2353, pruned_loss=0.03731, over 13396.00 frames. ], tot_loss[loss=0.1676, simple_loss=0.2312, pruned_loss=0.05197, over 2439378.87 frames. ], batch size: 88, lr: 4.36e-03, grad_scale: 32.0 2023-04-17 08:04:52,643 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78816.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:04:56,970 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-04-17 08:05:00,450 INFO [train.py:893] (1/4) Epoch 30, batch 550, loss[loss=0.1354, simple_loss=0.1949, pruned_loss=0.03798, over 13341.00 frames. ], tot_loss[loss=0.1666, simple_loss=0.2303, pruned_loss=0.05139, over 2491666.92 frames. ], batch size: 67, lr: 4.35e-03, grad_scale: 32.0 2023-04-17 08:05:03,825 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.765e+02 2.525e+02 2.925e+02 3.340e+02 5.483e+02, threshold=5.851e+02, percent-clipped=0.0 2023-04-17 08:05:25,321 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4867, 3.1733, 3.8823, 2.8131, 2.6140, 2.6922, 4.1676, 4.3472], device='cuda:1'), covar=tensor([0.1145, 0.1856, 0.0412, 0.1893, 0.1683, 0.1673, 0.0307, 0.0217], device='cuda:1'), in_proj_covar=tensor([0.0260, 0.0288, 0.0209, 0.0233, 0.0228, 0.0192, 0.0225, 0.0202], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 08:05:45,190 INFO [train.py:893] (1/4) Epoch 30, batch 600, loss[loss=0.1544, simple_loss=0.2196, pruned_loss=0.04461, over 13530.00 frames. ], tot_loss[loss=0.1661, simple_loss=0.2296, pruned_loss=0.05133, over 2521287.30 frames. ], batch size: 76, lr: 4.35e-03, grad_scale: 32.0 2023-04-17 08:05:47,485 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-04-17 08:05:58,905 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=78888.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:06:09,510 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2471, 4.0067, 4.1092, 2.4858, 4.5375, 4.2842, 4.2711, 4.4756], device='cuda:1'), covar=tensor([0.0259, 0.0162, 0.0141, 0.1166, 0.0127, 0.0265, 0.0151, 0.0123], device='cuda:1'), in_proj_covar=tensor([0.0101, 0.0063, 0.0090, 0.0107, 0.0107, 0.0121, 0.0090, 0.0085], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 08:06:19,983 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=78912.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:06:31,204 INFO [train.py:893] (1/4) Epoch 30, batch 650, loss[loss=0.1764, simple_loss=0.2371, pruned_loss=0.05782, over 13054.00 frames. ], tot_loss[loss=0.1657, simple_loss=0.2287, pruned_loss=0.05135, over 2553415.15 frames. ], batch size: 142, lr: 4.35e-03, grad_scale: 32.0 2023-04-17 08:06:34,572 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.693e+02 2.526e+02 3.052e+02 3.686e+02 4.958e+02, threshold=6.105e+02, percent-clipped=0.0 2023-04-17 08:06:42,047 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=78936.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:06:56,083 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-04-17 08:07:08,099 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=78965.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:07:13,916 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.35 vs. limit=5.0 2023-04-17 08:07:15,824 INFO [train.py:893] (1/4) Epoch 30, batch 700, loss[loss=0.1653, simple_loss=0.229, pruned_loss=0.05077, over 13283.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2274, pruned_loss=0.05058, over 2578860.97 frames. ], batch size: 124, lr: 4.35e-03, grad_scale: 32.0 2023-04-17 08:08:02,127 INFO [train.py:893] (1/4) Epoch 30, batch 750, loss[loss=0.1726, simple_loss=0.24, pruned_loss=0.05257, over 13417.00 frames. ], tot_loss[loss=0.165, simple_loss=0.2283, pruned_loss=0.05081, over 2601443.69 frames. ], batch size: 113, lr: 4.35e-03, grad_scale: 32.0 2023-04-17 08:08:04,075 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79026.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:08:05,378 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.768e+02 2.504e+02 2.968e+02 3.500e+02 7.955e+02, threshold=5.936e+02, percent-clipped=1.0 2023-04-17 08:08:12,116 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.2409, 5.0685, 5.2950, 5.0869, 5.5459, 5.0544, 5.5451, 5.5480], device='cuda:1'), covar=tensor([0.0416, 0.0605, 0.0620, 0.0601, 0.0578, 0.0895, 0.0494, 0.0456], device='cuda:1'), in_proj_covar=tensor([0.0320, 0.0322, 0.0333, 0.0246, 0.0478, 0.0372, 0.0313, 0.0331], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-04-17 08:08:48,615 INFO [train.py:893] (1/4) Epoch 30, batch 800, loss[loss=0.1872, simple_loss=0.2558, pruned_loss=0.0593, over 13399.00 frames. ], tot_loss[loss=0.167, simple_loss=0.2303, pruned_loss=0.05187, over 2614152.84 frames. ], batch size: 113, lr: 4.35e-03, grad_scale: 64.0 2023-04-17 08:08:54,715 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8502, 4.1338, 3.9072, 3.9709, 4.0633, 4.2625, 4.1382, 3.8922], device='cuda:1'), covar=tensor([0.0283, 0.0250, 0.0320, 0.0743, 0.0229, 0.0209, 0.0258, 0.0515], device='cuda:1'), in_proj_covar=tensor([0.0175, 0.0168, 0.0193, 0.0274, 0.0191, 0.0207, 0.0187, 0.0116], device='cuda:1'), out_proj_covar=tensor([0.0004, 0.0004, 0.0004, 0.0006, 0.0004, 0.0005, 0.0004, 0.0003], device='cuda:1') 2023-04-17 08:09:26,258 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79116.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:09:33,369 INFO [train.py:893] (1/4) Epoch 30, batch 850, loss[loss=0.1722, simple_loss=0.2352, pruned_loss=0.05458, over 13293.00 frames. ], tot_loss[loss=0.168, simple_loss=0.2314, pruned_loss=0.05228, over 2628984.20 frames. ], batch size: 124, lr: 4.35e-03, grad_scale: 32.0 2023-04-17 08:09:37,515 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.639e+02 2.521e+02 3.034e+02 3.558e+02 4.951e+02, threshold=6.069e+02, percent-clipped=0.0 2023-04-17 08:10:09,929 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79164.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:10:18,666 INFO [train.py:893] (1/4) Epoch 30, batch 900, loss[loss=0.1471, simple_loss=0.2119, pruned_loss=0.04117, over 13228.00 frames. ], tot_loss[loss=0.1683, simple_loss=0.2314, pruned_loss=0.0526, over 2637848.69 frames. ], batch size: 132, lr: 4.35e-03, grad_scale: 32.0 2023-04-17 08:10:27,886 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.0669, 2.1211, 3.7714, 3.6376, 3.5367, 2.9868, 3.3561, 2.9382], device='cuda:1'), covar=tensor([0.1876, 0.1312, 0.0158, 0.0214, 0.0283, 0.0659, 0.0291, 0.0736], device='cuda:1'), in_proj_covar=tensor([0.0217, 0.0185, 0.0134, 0.0138, 0.0140, 0.0178, 0.0150, 0.0187], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:1') 2023-04-17 08:10:50,338 WARNING [train.py:1054] (1/4) Exclude cut with ID 04b63653596f689ef7216c2603c0f6523a707a0b3a695f6f7adb7314161502892f2f2d7a577d62fd2ea21a9b72d688e7f4f12ff3c26432edad22541afe263dbb-common_voice_en_530541 from training. Duration: 24.672 2023-04-17 08:10:52,957 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79212.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:11:03,400 INFO [train.py:893] (1/4) Epoch 30, batch 950, loss[loss=0.1352, simple_loss=0.1997, pruned_loss=0.03534, over 13453.00 frames. ], tot_loss[loss=0.1675, simple_loss=0.2298, pruned_loss=0.05257, over 2644786.39 frames. ], batch size: 106, lr: 4.34e-03, grad_scale: 32.0 2023-04-17 08:11:08,156 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.773e+02 2.620e+02 3.020e+02 3.590e+02 5.764e+02, threshold=6.040e+02, percent-clipped=0.0 2023-04-17 08:11:23,158 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.7618, 4.0743, 3.0407, 2.7544, 2.8088, 2.4577, 4.1147, 2.3104], device='cuda:1'), covar=tensor([0.1929, 0.0356, 0.1441, 0.2456, 0.1011, 0.3629, 0.0320, 0.4571], device='cuda:1'), in_proj_covar=tensor([0.0322, 0.0302, 0.0340, 0.0359, 0.0274, 0.0345, 0.0232, 0.0381], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 08:11:36,631 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79260.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:11:49,413 INFO [train.py:893] (1/4) Epoch 30, batch 1000, loss[loss=0.1573, simple_loss=0.2244, pruned_loss=0.04508, over 13376.00 frames. ], tot_loss[loss=0.1655, simple_loss=0.2273, pruned_loss=0.05183, over 2650025.91 frames. ], batch size: 113, lr: 4.34e-03, grad_scale: 32.0 2023-04-17 08:11:50,797 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.42 vs. limit=5.0 2023-04-17 08:11:59,383 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79285.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 08:12:32,590 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79321.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:12:34,799 INFO [train.py:893] (1/4) Epoch 30, batch 1050, loss[loss=0.1885, simple_loss=0.2484, pruned_loss=0.06429, over 13271.00 frames. ], tot_loss[loss=0.1643, simple_loss=0.2263, pruned_loss=0.05111, over 2653174.67 frames. ], batch size: 124, lr: 4.34e-03, grad_scale: 32.0 2023-04-17 08:12:40,141 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.832e+02 2.591e+02 3.017e+02 3.726e+02 6.684e+02, threshold=6.035e+02, percent-clipped=2.0 2023-04-17 08:12:43,730 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79333.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:12:46,986 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.2432, 4.7863, 4.6667, 4.7697, 4.5974, 4.6192, 5.2033, 4.7223], device='cuda:1'), covar=tensor([0.0626, 0.1217, 0.2149, 0.2256, 0.0911, 0.1504, 0.0741, 0.1117], device='cuda:1'), in_proj_covar=tensor([0.0365, 0.0413, 0.0509, 0.0511, 0.0329, 0.0383, 0.0471, 0.0387], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 08:12:55,047 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79346.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 08:13:02,506 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.1127, 4.9631, 5.1895, 4.9643, 5.4492, 4.8733, 5.4576, 5.4303], device='cuda:1'), covar=tensor([0.0510, 0.0582, 0.0693, 0.0723, 0.0583, 0.0951, 0.0462, 0.0490], device='cuda:1'), in_proj_covar=tensor([0.0322, 0.0323, 0.0336, 0.0248, 0.0480, 0.0375, 0.0314, 0.0332], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0005, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-04-17 08:13:08,825 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.2576, 2.6435, 2.2523, 4.1392, 4.5940, 3.4053, 4.4716, 4.2833], device='cuda:1'), covar=tensor([0.0093, 0.0923, 0.1015, 0.0100, 0.0072, 0.0464, 0.0077, 0.0084], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0095, 0.0103, 0.0088, 0.0076, 0.0085, 0.0063, 0.0086], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 08:13:12,098 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79364.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:13:20,291 INFO [train.py:893] (1/4) Epoch 30, batch 1100, loss[loss=0.1726, simple_loss=0.232, pruned_loss=0.05662, over 13434.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2269, pruned_loss=0.0507, over 2653989.93 frames. ], batch size: 65, lr: 4.34e-03, grad_scale: 32.0 2023-04-17 08:13:21,465 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79375.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 08:13:39,862 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79394.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:13:46,146 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.7246, 4.0269, 2.8041, 3.7010, 3.8442, 2.5937, 3.3375, 2.8953], device='cuda:1'), covar=tensor([0.0376, 0.0380, 0.0936, 0.0370, 0.0322, 0.1149, 0.0687, 0.0929], device='cuda:1'), in_proj_covar=tensor([0.0170, 0.0188, 0.0178, 0.0240, 0.0145, 0.0163, 0.0166, 0.0163], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 08:14:07,256 INFO [train.py:893] (1/4) Epoch 30, batch 1150, loss[loss=0.1675, simple_loss=0.2342, pruned_loss=0.05039, over 13068.00 frames. ], tot_loss[loss=0.1629, simple_loss=0.2261, pruned_loss=0.04984, over 2653006.31 frames. ], batch size: 142, lr: 4.34e-03, grad_scale: 32.0 2023-04-17 08:14:08,364 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79425.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:14:11,338 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.798e+02 2.546e+02 3.063e+02 3.476e+02 8.677e+02, threshold=6.127e+02, percent-clipped=1.0 2023-04-17 08:14:17,397 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79436.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 08:14:44,492 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9425, 2.9109, 2.6108, 1.7320, 1.8859, 2.4341, 2.5071, 3.0291], device='cuda:1'), covar=tensor([0.1042, 0.0305, 0.0521, 0.1636, 0.0365, 0.0495, 0.0781, 0.0228], device='cuda:1'), in_proj_covar=tensor([0.0181, 0.0159, 0.0134, 0.0216, 0.0119, 0.0174, 0.0185, 0.0144], device='cuda:1'), out_proj_covar=tensor([1.3363e-04, 1.1866e-04, 1.0256e-04, 1.5974e-04, 8.5367e-05, 1.3160e-04, 1.3938e-04, 1.0575e-04], device='cuda:1') 2023-04-17 08:14:51,400 INFO [train.py:893] (1/4) Epoch 30, batch 1200, loss[loss=0.1719, simple_loss=0.2372, pruned_loss=0.05334, over 13392.00 frames. ], tot_loss[loss=0.1626, simple_loss=0.2263, pruned_loss=0.04951, over 2659704.50 frames. ], batch size: 113, lr: 4.34e-03, grad_scale: 32.0 2023-04-17 08:15:06,194 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0544, 4.4727, 4.2881, 4.3006, 4.3098, 4.1186, 4.5246, 4.5830], device='cuda:1'), covar=tensor([0.0249, 0.0230, 0.0232, 0.0317, 0.0289, 0.0298, 0.0273, 0.0239], device='cuda:1'), in_proj_covar=tensor([0.0149, 0.0211, 0.0179, 0.0192, 0.0165, 0.0213, 0.0144, 0.0197], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-04-17 08:15:19,167 WARNING [train.py:1054] (1/4) Exclude cut with ID 48528338442073e08f7886e133be1f7f773562a5c0257c0f3c33e19edf5de29dcefe46b9a0e497820295177ae021e53c9518185178f98612c169580ff3a07084-common_voice_en_19393114 from training. Duration: 20.016 2023-04-17 08:15:29,782 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6291, 3.4527, 2.6750, 3.0698, 2.8731, 2.0360, 3.4870, 2.0764], device='cuda:1'), covar=tensor([0.0755, 0.0759, 0.0615, 0.0517, 0.0745, 0.1997, 0.1041, 0.1313], device='cuda:1'), in_proj_covar=tensor([0.0150, 0.0154, 0.0142, 0.0124, 0.0155, 0.0195, 0.0194, 0.0150], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 08:15:31,929 WARNING [train.py:1054] (1/4) Exclude cut with ID c79cca7d4c2127ae9a701febbb70a6a67975417cf081587b35693f0d90245c00a68de59daa3af7798bfd7df858f7e1b43acc3ca7a6f2419bdfa14ae128150d0b-common_voice_en_170142 from training. Duration: 20.16 2023-04-17 08:15:37,640 INFO [train.py:893] (1/4) Epoch 30, batch 1250, loss[loss=0.171, simple_loss=0.2353, pruned_loss=0.05332, over 13463.00 frames. ], tot_loss[loss=0.1633, simple_loss=0.2268, pruned_loss=0.04988, over 2658164.87 frames. ], batch size: 100, lr: 4.34e-03, grad_scale: 32.0 2023-04-17 08:15:41,733 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.762e+02 2.699e+02 3.086e+02 3.611e+02 7.494e+02, threshold=6.172e+02, percent-clipped=2.0 2023-04-17 08:15:43,244 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 2023-04-17 08:16:23,291 INFO [train.py:893] (1/4) Epoch 30, batch 1300, loss[loss=0.183, simple_loss=0.2403, pruned_loss=0.06288, over 13542.00 frames. ], tot_loss[loss=0.164, simple_loss=0.2276, pruned_loss=0.05023, over 2657414.79 frames. ], batch size: 87, lr: 4.33e-03, grad_scale: 32.0 2023-04-17 08:16:34,056 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79585.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:16:39,808 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8535, 3.8423, 2.7287, 3.5049, 3.8352, 2.5282, 3.5179, 2.5525], device='cuda:1'), covar=tensor([0.0320, 0.0224, 0.1126, 0.0390, 0.0303, 0.1304, 0.0571, 0.1305], device='cuda:1'), in_proj_covar=tensor([0.0170, 0.0188, 0.0179, 0.0239, 0.0145, 0.0163, 0.0166, 0.0163], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 08:17:07,047 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79621.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:17:09,206 INFO [train.py:893] (1/4) Epoch 30, batch 1350, loss[loss=0.1641, simple_loss=0.2284, pruned_loss=0.0499, over 13227.00 frames. ], tot_loss[loss=0.1652, simple_loss=0.2288, pruned_loss=0.05081, over 2656620.91 frames. ], batch size: 132, lr: 4.33e-03, grad_scale: 32.0 2023-04-17 08:17:13,984 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 2.034e+02 2.559e+02 2.996e+02 3.614e+02 6.449e+02, threshold=5.992e+02, percent-clipped=1.0 2023-04-17 08:17:25,402 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79641.0, num_to_drop=1, layers_to_drop={1} 2023-04-17 08:17:30,261 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79646.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:17:39,150 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79657.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:17:42,507 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.3658, 3.0356, 2.8802, 3.4305, 2.7381, 3.4277, 3.4782, 2.9454], device='cuda:1'), covar=tensor([0.0077, 0.0198, 0.0203, 0.0171, 0.0198, 0.0116, 0.0133, 0.0162], device='cuda:1'), in_proj_covar=tensor([0.0117, 0.0131, 0.0135, 0.0139, 0.0146, 0.0125, 0.0117, 0.0122], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 08:17:48,937 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.9352, 4.2669, 3.1274, 2.8280, 2.9346, 2.5616, 4.3585, 2.3677], device='cuda:1'), covar=tensor([0.2390, 0.0450, 0.1727, 0.2904, 0.1145, 0.3841, 0.0318, 0.5221], device='cuda:1'), in_proj_covar=tensor([0.0324, 0.0303, 0.0340, 0.0361, 0.0274, 0.0347, 0.0232, 0.0384], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 08:17:50,403 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79669.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:17:55,581 INFO [train.py:893] (1/4) Epoch 30, batch 1400, loss[loss=0.1477, simple_loss=0.2119, pruned_loss=0.04168, over 13544.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2283, pruned_loss=0.05073, over 2650972.45 frames. ], batch size: 98, lr: 4.33e-03, grad_scale: 32.0 2023-04-17 08:18:08,739 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79689.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:18:34,657 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79718.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:18:36,867 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79720.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:18:40,082 INFO [train.py:893] (1/4) Epoch 30, batch 1450, loss[loss=0.1886, simple_loss=0.2524, pruned_loss=0.06236, over 13489.00 frames. ], tot_loss[loss=0.165, simple_loss=0.2282, pruned_loss=0.05087, over 2655774.03 frames. ], batch size: 93, lr: 4.33e-03, grad_scale: 32.0 2023-04-17 08:18:44,693 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.924e+02 2.494e+02 2.889e+02 3.354e+02 4.257e+02, threshold=5.778e+02, percent-clipped=0.0 2023-04-17 08:18:46,652 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79731.0, num_to_drop=1, layers_to_drop={2} 2023-04-17 08:19:14,836 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=79762.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:19:25,898 INFO [train.py:893] (1/4) Epoch 30, batch 1500, loss[loss=0.1503, simple_loss=0.2199, pruned_loss=0.04032, over 13346.00 frames. ], tot_loss[loss=0.1645, simple_loss=0.2282, pruned_loss=0.05042, over 2659284.67 frames. ], batch size: 118, lr: 4.33e-03, grad_scale: 32.0 2023-04-17 08:19:27,114 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6821, 3.4552, 4.2615, 2.9262, 2.8759, 2.8719, 4.5191, 4.6575], device='cuda:1'), covar=tensor([0.1180, 0.1883, 0.0358, 0.1958, 0.1560, 0.1755, 0.0291, 0.0276], device='cuda:1'), in_proj_covar=tensor([0.0261, 0.0288, 0.0209, 0.0233, 0.0228, 0.0191, 0.0226, 0.0203], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 08:19:52,538 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1262, 2.3942, 1.9954, 4.0247, 4.4386, 3.2014, 4.4014, 4.2027], device='cuda:1'), covar=tensor([0.0100, 0.1071, 0.1235, 0.0103, 0.0078, 0.0571, 0.0072, 0.0086], device='cuda:1'), in_proj_covar=tensor([0.0083, 0.0095, 0.0104, 0.0089, 0.0076, 0.0086, 0.0063, 0.0086], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 08:20:00,958 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-04-17 08:20:11,386 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=79823.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:20:11,889 INFO [train.py:893] (1/4) Epoch 30, batch 1550, loss[loss=0.176, simple_loss=0.2372, pruned_loss=0.05736, over 13425.00 frames. ], tot_loss[loss=0.1642, simple_loss=0.2278, pruned_loss=0.0503, over 2652435.91 frames. ], batch size: 95, lr: 4.33e-03, grad_scale: 32.0 2023-04-17 08:20:16,715 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.723e+02 2.643e+02 2.905e+02 3.656e+02 6.306e+02, threshold=5.811e+02, percent-clipped=2.0 2023-04-17 08:20:21,793 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.0878, 2.7484, 2.6502, 3.0935, 2.5043, 3.1897, 3.1364, 2.6649], device='cuda:1'), covar=tensor([0.0101, 0.0223, 0.0180, 0.0195, 0.0217, 0.0131, 0.0162, 0.0171], device='cuda:1'), in_proj_covar=tensor([0.0119, 0.0133, 0.0136, 0.0141, 0.0147, 0.0126, 0.0119, 0.0123], device='cuda:1'), out_proj_covar=tensor([0.0001, 0.0001, 0.0001, 0.0001, 0.0002, 0.0001, 0.0001, 0.0001], device='cuda:1') 2023-04-17 08:20:57,763 INFO [train.py:893] (1/4) Epoch 30, batch 1600, loss[loss=0.1828, simple_loss=0.2448, pruned_loss=0.06041, over 13579.00 frames. ], tot_loss[loss=0.1644, simple_loss=0.2284, pruned_loss=0.05018, over 2654979.63 frames. ], batch size: 89, lr: 4.33e-03, grad_scale: 32.0 2023-04-17 08:21:22,792 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.9712, 4.4967, 4.3588, 4.4371, 4.3420, 4.3143, 4.9433, 4.5407], device='cuda:1'), covar=tensor([0.0782, 0.1249, 0.2185, 0.2523, 0.1108, 0.1834, 0.0941, 0.1338], device='cuda:1'), in_proj_covar=tensor([0.0364, 0.0411, 0.0506, 0.0508, 0.0327, 0.0382, 0.0471, 0.0386], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 08:21:43,801 INFO [train.py:893] (1/4) Epoch 30, batch 1650, loss[loss=0.1684, simple_loss=0.2319, pruned_loss=0.05244, over 13329.00 frames. ], tot_loss[loss=0.1638, simple_loss=0.2285, pruned_loss=0.04956, over 2660516.16 frames. ], batch size: 118, lr: 4.32e-03, grad_scale: 32.0 2023-04-17 08:21:48,480 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.54 vs. limit=5.0 2023-04-17 08:21:48,674 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.912e+02 2.404e+02 2.671e+02 3.103e+02 7.046e+02, threshold=5.343e+02, percent-clipped=2.0 2023-04-17 08:21:59,456 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=79941.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:21:59,552 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79941.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 08:22:30,246 INFO [train.py:893] (1/4) Epoch 30, batch 1700, loss[loss=0.1319, simple_loss=0.2037, pruned_loss=0.03004, over 13535.00 frames. ], tot_loss[loss=0.1636, simple_loss=0.2286, pruned_loss=0.04924, over 2657584.69 frames. ], batch size: 72, lr: 4.32e-03, grad_scale: 32.0 2023-04-17 08:22:43,778 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=79989.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 08:22:43,841 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=79989.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:22:47,843 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0096, 3.7814, 3.9802, 2.2181, 4.2821, 4.0563, 4.0289, 4.2506], device='cuda:1'), covar=tensor([0.0267, 0.0161, 0.0142, 0.1235, 0.0130, 0.0256, 0.0140, 0.0121], device='cuda:1'), in_proj_covar=tensor([0.0103, 0.0063, 0.0092, 0.0110, 0.0109, 0.0122, 0.0091, 0.0087], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 08:23:09,723 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80013.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:23:16,314 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80020.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:23:20,048 INFO [train.py:893] (1/4) Epoch 30, batch 1750, loss[loss=0.1561, simple_loss=0.2243, pruned_loss=0.04393, over 13425.00 frames. ], tot_loss[loss=0.1624, simple_loss=0.2274, pruned_loss=0.04869, over 2660652.67 frames. ], batch size: 95, lr: 4.32e-03, grad_scale: 32.0 2023-04-17 08:23:24,293 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.625e+02 2.541e+02 2.943e+02 3.454e+02 5.416e+02, threshold=5.886e+02, percent-clipped=2.0 2023-04-17 08:23:26,154 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80031.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 08:23:32,288 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80037.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:23:43,903 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.1809, 3.1679, 2.9198, 2.1556, 2.4047, 2.8456, 2.8642, 3.3978], device='cuda:1'), covar=tensor([0.0976, 0.0368, 0.0558, 0.1341, 0.0456, 0.0533, 0.0627, 0.0244], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0161, 0.0136, 0.0217, 0.0121, 0.0176, 0.0187, 0.0144], device='cuda:1'), out_proj_covar=tensor([1.3512e-04, 1.1970e-04, 1.0393e-04, 1.6057e-04, 8.6242e-05, 1.3314e-04, 1.4084e-04, 1.0613e-04], device='cuda:1') 2023-04-17 08:24:00,578 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80068.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:24:05,335 INFO [train.py:893] (1/4) Epoch 30, batch 1800, loss[loss=0.1857, simple_loss=0.2499, pruned_loss=0.0607, over 13529.00 frames. ], tot_loss[loss=0.1619, simple_loss=0.2268, pruned_loss=0.04856, over 2659568.35 frames. ], batch size: 98, lr: 4.32e-03, grad_scale: 32.0 2023-04-17 08:24:10,281 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80079.0, num_to_drop=1, layers_to_drop={0} 2023-04-17 08:24:45,275 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80118.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:24:50,705 INFO [train.py:893] (1/4) Epoch 30, batch 1850, loss[loss=0.1523, simple_loss=0.2121, pruned_loss=0.04622, over 13373.00 frames. ], tot_loss[loss=0.1611, simple_loss=0.2259, pruned_loss=0.04818, over 2662319.60 frames. ], batch size: 67, lr: 4.32e-03, grad_scale: 32.0 2023-04-17 08:24:52,420 WARNING [train.py:1054] (1/4) Exclude cut with ID d2dfd1e3b98dcf1e35fa846f29120a93095efe5069904116be6b42831927a53f115bfc3446ded9aa455d24bd9d4fc7289e2c122d3e44590ac9a27e0b7f43c89e-common_voice_en_517302 from training. Duration: 21.72 2023-04-17 08:24:55,472 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.690e+02 2.423e+02 2.778e+02 3.234e+02 9.030e+02, threshold=5.555e+02, percent-clipped=2.0 2023-04-17 08:25:30,191 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.8608, 3.9119, 2.7879, 3.4898, 3.8437, 2.5805, 3.6263, 2.5854], device='cuda:1'), covar=tensor([0.0321, 0.0276, 0.0980, 0.0346, 0.0250, 0.1237, 0.0472, 0.1239], device='cuda:1'), in_proj_covar=tensor([0.0171, 0.0189, 0.0180, 0.0242, 0.0147, 0.0164, 0.0167, 0.0164], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 08:25:36,144 INFO [train.py:893] (1/4) Epoch 30, batch 1900, loss[loss=0.1534, simple_loss=0.2153, pruned_loss=0.04574, over 13529.00 frames. ], tot_loss[loss=0.161, simple_loss=0.2255, pruned_loss=0.04821, over 2657156.25 frames. ], batch size: 83, lr: 4.32e-03, grad_scale: 32.0 2023-04-17 08:26:06,710 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([2.6455, 3.1281, 2.5964, 2.7817, 2.8167, 2.0164, 3.1845, 2.1070], device='cuda:1'), covar=tensor([0.0556, 0.0683, 0.0543, 0.0464, 0.0549, 0.1647, 0.0814, 0.1104], device='cuda:1'), in_proj_covar=tensor([0.0149, 0.0154, 0.0141, 0.0123, 0.0154, 0.0194, 0.0193, 0.0148], device='cuda:1'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002], device='cuda:1') 2023-04-17 08:26:22,615 INFO [train.py:893] (1/4) Epoch 30, batch 1950, loss[loss=0.1562, simple_loss=0.216, pruned_loss=0.04822, over 13523.00 frames. ], tot_loss[loss=0.1612, simple_loss=0.2255, pruned_loss=0.0484, over 2657433.24 frames. ], batch size: 91, lr: 4.32e-03, grad_scale: 32.0 2023-04-17 08:26:26,823 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.736e+02 2.552e+02 2.954e+02 3.319e+02 5.622e+02, threshold=5.908e+02, percent-clipped=2.0 2023-04-17 08:26:37,577 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80241.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:26:43,432 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6554, 3.4185, 4.0968, 3.0272, 2.9020, 2.9354, 4.3949, 4.5457], device='cuda:1'), covar=tensor([0.1245, 0.1961, 0.0477, 0.1878, 0.1590, 0.1571, 0.0350, 0.0347], device='cuda:1'), in_proj_covar=tensor([0.0261, 0.0290, 0.0210, 0.0232, 0.0228, 0.0192, 0.0227, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 08:27:07,357 INFO [train.py:893] (1/4) Epoch 30, batch 2000, loss[loss=0.1836, simple_loss=0.2508, pruned_loss=0.05816, over 13441.00 frames. ], tot_loss[loss=0.1635, simple_loss=0.2279, pruned_loss=0.04951, over 2660117.61 frames. ], batch size: 106, lr: 4.32e-03, grad_scale: 32.0 2023-04-17 08:27:10,209 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.19 vs. limit=5.0 2023-04-17 08:27:13,768 WARNING [train.py:1054] (1/4) Exclude cut with ID 66b27ad6bd4f9cb9a012d8e5cb8391b2c5494a180128e0910c3fa9b3566bbf033a3ef43178d36693ba7c41a5e1373d8fa4699e5c91228afc6a64c29f3e583b7a-common_voice_en_32639323 from training. Duration: 0.9 2023-04-17 08:27:16,530 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80283.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:27:22,106 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80289.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:27:43,954 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80313.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:27:53,419 INFO [train.py:893] (1/4) Epoch 30, batch 2050, loss[loss=0.1553, simple_loss=0.219, pruned_loss=0.04578, over 13484.00 frames. ], tot_loss[loss=0.1649, simple_loss=0.2292, pruned_loss=0.05031, over 2659423.73 frames. ], batch size: 70, lr: 4.31e-03, grad_scale: 32.0 2023-04-17 08:27:56,471 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-04-17 08:27:58,268 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.934e+02 2.617e+02 2.996e+02 3.704e+02 6.259e+02, threshold=5.993e+02, percent-clipped=2.0 2023-04-17 08:28:12,263 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80344.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:28:13,955 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-04-17 08:28:27,569 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80361.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:28:28,544 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4632, 2.7482, 2.4279, 4.4034, 4.9418, 3.5698, 4.8475, 4.5725], device='cuda:1'), covar=tensor([0.0115, 0.0922, 0.0991, 0.0108, 0.0079, 0.0477, 0.0072, 0.0080], device='cuda:1'), in_proj_covar=tensor([0.0083, 0.0095, 0.0103, 0.0089, 0.0076, 0.0085, 0.0063, 0.0086], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 08:28:36,689 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9754, 2.9361, 2.6136, 1.9536, 1.9176, 2.5968, 2.7330, 3.1332], device='cuda:1'), covar=tensor([0.1087, 0.0401, 0.0597, 0.1576, 0.0420, 0.0510, 0.0674, 0.0245], device='cuda:1'), in_proj_covar=tensor([0.0183, 0.0161, 0.0136, 0.0218, 0.0121, 0.0177, 0.0188, 0.0143], device='cuda:1'), out_proj_covar=tensor([1.3518e-04, 1.1960e-04, 1.0389e-04, 1.6093e-04, 8.6906e-05, 1.3363e-04, 1.4143e-04, 1.0511e-04], device='cuda:1') 2023-04-17 08:28:39,453 INFO [train.py:893] (1/4) Epoch 30, batch 2100, loss[loss=0.1634, simple_loss=0.2337, pruned_loss=0.04654, over 13527.00 frames. ], tot_loss[loss=0.1637, simple_loss=0.2281, pruned_loss=0.04965, over 2663347.32 frames. ], batch size: 98, lr: 4.31e-03, grad_scale: 32.0 2023-04-17 08:28:45,375 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.1933, 2.3674, 2.2208, 4.0378, 4.5565, 3.3806, 4.4943, 4.2977], device='cuda:1'), covar=tensor([0.0102, 0.1108, 0.1062, 0.0111, 0.0067, 0.0498, 0.0078, 0.0080], device='cuda:1'), in_proj_covar=tensor([0.0082, 0.0095, 0.0103, 0.0089, 0.0076, 0.0085, 0.0063, 0.0085], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003], device='cuda:1') 2023-04-17 08:28:46,408 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=3.53 vs. limit=5.0 2023-04-17 08:28:56,806 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80393.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:29:11,818 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.6014, 3.3126, 4.1019, 2.9561, 2.8152, 2.8873, 4.3199, 4.4961], device='cuda:1'), covar=tensor([0.1186, 0.2050, 0.0344, 0.1816, 0.1563, 0.1623, 0.0254, 0.0178], device='cuda:1'), in_proj_covar=tensor([0.0261, 0.0290, 0.0209, 0.0232, 0.0228, 0.0192, 0.0227, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 08:29:19,836 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80418.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:29:25,184 INFO [train.py:893] (1/4) Epoch 30, batch 2150, loss[loss=0.1574, simple_loss=0.2278, pruned_loss=0.04355, over 13539.00 frames. ], tot_loss[loss=0.1633, simple_loss=0.2281, pruned_loss=0.04925, over 2662173.83 frames. ], batch size: 98, lr: 4.31e-03, grad_scale: 32.0 2023-04-17 08:29:29,310 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.734e+02 2.403e+02 2.859e+02 3.405e+02 5.659e+02, threshold=5.718e+02, percent-clipped=0.0 2023-04-17 08:29:31,220 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9262, 2.9239, 2.5377, 1.9392, 1.9755, 2.5359, 2.5925, 3.0451], device='cuda:1'), covar=tensor([0.1077, 0.0359, 0.0600, 0.1522, 0.0388, 0.0645, 0.0853, 0.0218], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0161, 0.0135, 0.0218, 0.0121, 0.0177, 0.0189, 0.0143], device='cuda:1'), out_proj_covar=tensor([1.3522e-04, 1.2000e-04, 1.0370e-04, 1.6094e-04, 8.6580e-05, 1.3351e-04, 1.4170e-04, 1.0508e-04], device='cuda:1') 2023-04-17 08:29:52,265 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80454.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:29:59,497 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80463.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:30:02,975 INFO [zipformer.py:625] (1/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=80466.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:30:10,033 INFO [train.py:893] (1/4) Epoch 30, batch 2200, loss[loss=0.163, simple_loss=0.2246, pruned_loss=0.05068, over 11594.00 frames. ], tot_loss[loss=0.1631, simple_loss=0.2278, pruned_loss=0.04918, over 2659154.01 frames. ], batch size: 157, lr: 4.31e-03, grad_scale: 32.0 2023-04-17 08:30:27,510 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80494.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:30:53,963 INFO [train.py:893] (1/4) Epoch 30, batch 2250, loss[loss=0.1673, simple_loss=0.2315, pruned_loss=0.05158, over 13533.00 frames. ], tot_loss[loss=0.1616, simple_loss=0.226, pruned_loss=0.0486, over 2659655.42 frames. ], batch size: 91, lr: 4.31e-03, grad_scale: 32.0 2023-04-17 08:30:54,285 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80524.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:30:58,861 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.751e+02 2.489e+02 2.862e+02 3.477e+02 6.119e+02, threshold=5.724e+02, percent-clipped=2.0 2023-04-17 08:31:23,210 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80555.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:31:41,411 INFO [train.py:893] (1/4) Epoch 30, batch 2300, loss[loss=0.179, simple_loss=0.2376, pruned_loss=0.06024, over 13190.00 frames. ], tot_loss[loss=0.1607, simple_loss=0.2251, pruned_loss=0.04813, over 2662720.49 frames. ], batch size: 132, lr: 4.31e-03, grad_scale: 32.0 2023-04-17 08:32:25,295 INFO [train.py:893] (1/4) Epoch 30, batch 2350, loss[loss=0.1618, simple_loss=0.2213, pruned_loss=0.05117, over 11744.00 frames. ], tot_loss[loss=0.1607, simple_loss=0.2251, pruned_loss=0.04813, over 2663056.22 frames. ], batch size: 157, lr: 4.31e-03, grad_scale: 32.0 2023-04-17 08:32:30,854 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.921e+02 2.384e+02 2.743e+02 3.266e+02 4.484e+02, threshold=5.485e+02, percent-clipped=0.0 2023-04-17 08:32:36,014 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.3509, 4.7895, 4.6033, 4.6282, 4.6505, 4.4035, 4.9085, 4.8792], device='cuda:1'), covar=tensor([0.0220, 0.0202, 0.0204, 0.0331, 0.0273, 0.0253, 0.0215, 0.0203], device='cuda:1'), in_proj_covar=tensor([0.0151, 0.0212, 0.0181, 0.0195, 0.0165, 0.0215, 0.0145, 0.0198], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0004, 0.0003, 0.0004], device='cuda:1') 2023-04-17 08:32:40,113 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80639.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:32:48,949 WARNING [train.py:1054] (1/4) Exclude cut with ID 7963691c43c8cc498c58f117527522bf772c76c38530570bc55ef04834f67fb7a9227bd0fa1f13e64e8de1cde6594f3501e172ab86559697c08726cac26f4c6f-common_voice_en_21692090 from training. Duration: 20.16 2023-04-17 08:32:51,095 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.10 vs. limit=5.0 2023-04-17 08:33:12,432 INFO [train.py:893] (1/4) Epoch 30, batch 2400, loss[loss=0.1565, simple_loss=0.2208, pruned_loss=0.04612, over 13521.00 frames. ], tot_loss[loss=0.1598, simple_loss=0.2242, pruned_loss=0.04776, over 2661256.29 frames. ], batch size: 85, lr: 4.30e-03, grad_scale: 32.0 2023-04-17 08:33:30,504 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([5.2300, 4.7123, 4.6288, 4.7487, 4.5535, 4.5737, 5.1650, 4.7390], device='cuda:1'), covar=tensor([0.0593, 0.1115, 0.2003, 0.2191, 0.0981, 0.1587, 0.0793, 0.1166], device='cuda:1'), in_proj_covar=tensor([0.0362, 0.0410, 0.0499, 0.0501, 0.0324, 0.0377, 0.0470, 0.0383], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 08:33:55,197 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 2023-04-17 08:33:57,816 INFO [train.py:893] (1/4) Epoch 30, batch 2450, loss[loss=0.1722, simple_loss=0.2391, pruned_loss=0.05272, over 13468.00 frames. ], tot_loss[loss=0.1603, simple_loss=0.2244, pruned_loss=0.04805, over 2659056.19 frames. ], batch size: 100, lr: 4.30e-03, grad_scale: 32.0 2023-04-17 08:34:01,993 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.530e+02 2.276e+02 2.683e+02 3.157e+02 8.436e+02, threshold=5.367e+02, percent-clipped=3.0 2023-04-17 08:34:03,098 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.9367, 2.8305, 2.3964, 1.8102, 1.7146, 2.3248, 2.4418, 2.9426], device='cuda:1'), covar=tensor([0.0996, 0.0329, 0.0631, 0.1605, 0.0301, 0.0606, 0.0860, 0.0248], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0161, 0.0135, 0.0220, 0.0121, 0.0177, 0.0190, 0.0144], device='cuda:1'), out_proj_covar=tensor([1.3584e-04, 1.2018e-04, 1.0384e-04, 1.6240e-04, 8.6961e-05, 1.3335e-04, 1.4287e-04, 1.0572e-04], device='cuda:1') 2023-04-17 08:34:09,721 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80736.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:34:20,283 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80749.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:34:32,947 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.7804, 2.6463, 2.2914, 1.5737, 1.6058, 2.1364, 2.2666, 2.7399], device='cuda:1'), covar=tensor([0.0908, 0.0345, 0.0474, 0.1535, 0.0190, 0.0585, 0.0766, 0.0210], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0161, 0.0135, 0.0220, 0.0122, 0.0177, 0.0190, 0.0144], device='cuda:1'), out_proj_covar=tensor([1.3568e-04, 1.2019e-04, 1.0379e-04, 1.6222e-04, 8.7299e-05, 1.3326e-04, 1.4255e-04, 1.0560e-04], device='cuda:1') 2023-04-17 08:34:41,014 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.0216, 3.7811, 3.9403, 2.2862, 4.2530, 4.0380, 3.9936, 4.2757], device='cuda:1'), covar=tensor([0.0245, 0.0158, 0.0145, 0.1181, 0.0123, 0.0263, 0.0142, 0.0101], device='cuda:1'), in_proj_covar=tensor([0.0102, 0.0062, 0.0090, 0.0107, 0.0107, 0.0121, 0.0090, 0.0086], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0002, 0.0003, 0.0004, 0.0004, 0.0004, 0.0003, 0.0003], device='cuda:1') 2023-04-17 08:34:42,784 INFO [zipformer.py:625] (1/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=80773.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:34:43,342 INFO [train.py:893] (1/4) Epoch 30, batch 2500, loss[loss=0.1729, simple_loss=0.2289, pruned_loss=0.05845, over 13520.00 frames. ], tot_loss[loss=0.1603, simple_loss=0.2244, pruned_loss=0.04809, over 2659438.12 frames. ], batch size: 85, lr: 4.30e-03, grad_scale: 32.0 2023-04-17 08:34:59,322 INFO [scaling.py:679] (1/4) Whitening: num_groups=8, num_channels=96, metric=1.57 vs. limit=2.0 2023-04-17 08:35:04,617 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80797.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:35:24,637 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80819.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:35:28,605 INFO [train.py:893] (1/4) Epoch 30, batch 2550, loss[loss=0.1572, simple_loss=0.2245, pruned_loss=0.045, over 13354.00 frames. ], tot_loss[loss=0.1609, simple_loss=0.225, pruned_loss=0.04842, over 2661855.19 frames. ], batch size: 73, lr: 4.30e-03, grad_scale: 32.0 2023-04-17 08:35:33,326 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.688e+02 2.409e+02 2.835e+02 3.365e+02 6.222e+02, threshold=5.670e+02, percent-clipped=1.0 2023-04-17 08:35:38,589 INFO [zipformer.py:625] (1/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=80834.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:35:45,177 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([1.8404, 2.5599, 2.1119, 1.5814, 1.5299, 2.0353, 2.2818, 2.6336], device='cuda:1'), covar=tensor([0.0976, 0.0373, 0.0788, 0.1679, 0.0218, 0.0665, 0.0869, 0.0265], device='cuda:1'), in_proj_covar=tensor([0.0184, 0.0162, 0.0136, 0.0220, 0.0122, 0.0177, 0.0190, 0.0144], device='cuda:1'), out_proj_covar=tensor([1.3590e-04, 1.2051e-04, 1.0408e-04, 1.6216e-04, 8.7412e-05, 1.3334e-04, 1.4232e-04, 1.0557e-04], device='cuda:1') 2023-04-17 08:35:52,113 INFO [zipformer.py:625] (1/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=80850.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:35:54,357 WARNING [train.py:1054] (1/4) Exclude cut with ID 3c1e147a7e233e57b4a58ba4484e77c5a4ea7f61b0e09b142dd61df911d700cb91f4f6d097065ac73a6cdf3eba8864c9aeea6825a69566bbeb12797de4ca20b9-common_voice_en_505097 from training. Duration: 30.264 2023-04-17 08:36:13,729 INFO [train.py:893] (1/4) Epoch 30, batch 2600, loss[loss=0.1607, simple_loss=0.2249, pruned_loss=0.0483, over 13349.00 frames. ], tot_loss[loss=0.1611, simple_loss=0.2249, pruned_loss=0.04867, over 2663314.52 frames. ], batch size: 67, lr: 4.30e-03, grad_scale: 32.0 2023-04-17 08:36:24,475 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([3.4998, 3.2461, 3.9212, 2.8773, 2.6528, 2.7863, 4.1957, 4.3461], device='cuda:1'), covar=tensor([0.1323, 0.2004, 0.0430, 0.1939, 0.1639, 0.1567, 0.0307, 0.0259], device='cuda:1'), in_proj_covar=tensor([0.0261, 0.0290, 0.0209, 0.0233, 0.0228, 0.0192, 0.0227, 0.0204], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 08:36:55,469 INFO [train.py:893] (1/4) Epoch 30, batch 2650, loss[loss=0.176, simple_loss=0.2383, pruned_loss=0.05684, over 13211.00 frames. ], tot_loss[loss=0.1617, simple_loss=0.2253, pruned_loss=0.0491, over 2657950.17 frames. ], batch size: 132, lr: 4.30e-03, grad_scale: 32.0 2023-04-17 08:36:59,342 INFO [optim.py:368] (1/4) Clipping_scale=2.0, grad-norm quartiles 1.862e+02 2.620e+02 3.122e+02 3.929e+02 7.636e+02, threshold=6.244e+02, percent-clipped=3.0 2023-04-17 08:37:07,033 INFO [zipformer.py:625] (1/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=80939.0, num_to_drop=0, layers_to_drop=set() 2023-04-17 08:37:10,217 INFO [zipformer.py:1454] (1/4) attn_weights_entropy = tensor([4.4199, 4.5576, 3.2454, 4.1810, 4.4417, 3.0398, 4.0563, 3.1500], device='cuda:1'), covar=tensor([0.0247, 0.0218, 0.0862, 0.0329, 0.0216, 0.1005, 0.0379, 0.1087], device='cuda:1'), in_proj_covar=tensor([0.0170, 0.0187, 0.0179, 0.0239, 0.0146, 0.0163, 0.0166, 0.0162], device='cuda:1'), out_proj_covar=tensor([0.0003, 0.0004, 0.0004, 0.0005, 0.0003, 0.0003, 0.0003, 0.0003], device='cuda:1') 2023-04-17 08:37:12,051 INFO [scaling.py:679] (1/4) Whitening: num_groups=1, num_channels=384, metric=4.94 vs. limit=5.0 2023-04-17 08:37:33,514 INFO [train.py:1151] (1/4) Done!