2023-02-08 23:42:53,785 INFO [train.py:973] (0/4) Training started 2023-02-08 23:42:53,792 INFO [train.py:983] (0/4) Device: cuda:0 2023-02-08 23:42:53,850 INFO [train.py:992] (0/4) {'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': '1.23.3', 'k2-build-type': 'Debug', 'k2-with-cuda': True, 'k2-git-sha1': '3b81ac9686aee539d447bb2085b2cdfc131c7c91', 'k2-git-date': 'Thu Jan 26 20:40:25 2023', 'lhotse-version': '1.9.0.dev+git.97bf4b0.dirty', 'torch-version': '1.10.0+cu102', 'torch-cuda-available': True, 'torch-cuda-version': '10.2', 'python-version': '3.8', 'icefall-git-branch': 'surt', 'icefall-git-sha1': 'b3d0d34-dirty', 'icefall-git-date': 'Sat Feb 4 14:53:48 2023', 'icefall-path': '/exp/draj/mini_scale_2022/icefall', 'k2-path': '/exp/draj/mini_scale_2022/k2/k2/python/k2/__init__.py', 'lhotse-path': '/exp/draj/mini_scale_2022/lhotse/lhotse/__init__.py', 'hostname': 'r8n07', 'IP address': '10.1.8.7'}, 'world_size': 4, 'master_port': 12354, 'tensorboard': True, 'num_epochs': 30, 'start_epoch': 28, 'start_batch': 0, 'exp_dir': PosixPath('pruned_transducer_stateless7_streaming/exp/v1'), 'bpe_model': 'data/lang_bpe_500/bpe.model', 'base_lr': 0.05, 'lr_batches': 5000, 'lr_epochs': 3.5, 'context_size': 2, 'prune_range': 5, 'lm_scale': 0.25, 'am_scale': 0.0, 'simple_loss_scale': 0.5, 'seed': 42, 'print_diagnostics': False, 'inf_check': False, 'save_every_n': 2000, 'keep_last_k': 10, 'average_period': 200, 'use_fp16': True, 'num_encoder_layers': '2,2,2,2,2', 'feedforward_dims': '768,768,768,768,768', 'nhead': '8,8,8,8,8', 'encoder_dims': '256,256,256,256,256', 'attention_dims': '192,192,192,192,192', 'encoder_unmasked_dims': '192,192,192,192,192', 'zipformer_downsampling_factors': '1,2,4,8,2', 'cnn_module_kernels': '31,31,31,31,31', 'decoder_dim': 512, 'joiner_dim': 512, 'short_chunk_size': 50, 'num_left_chunks': 4, 'decode_chunk_len': 32, 'full_libri': True, 'manifest_dir': PosixPath('data/manifests'), 'max_duration': 500, '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} 2023-02-08 23:42:53,850 INFO [train.py:994] (0/4) About to create model 2023-02-08 23:42:54,160 INFO [zipformer.py:402] (0/4) 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. 2023-02-08 23:42:54,172 INFO [train.py:998] (0/4) Number of model parameters: 20697573 2023-02-08 23:42:54,388 INFO [checkpoint.py:112] (0/4) Loading checkpoint from pruned_transducer_stateless7_streaming/exp/v1/epoch-27.pt 2023-02-08 23:42:58,649 INFO [checkpoint.py:131] (0/4) Loading averaged model 2023-02-08 23:43:03,619 INFO [train.py:1013] (0/4) Using DDP 2023-02-08 23:43:03,869 INFO [train.py:1030] (0/4) Loading optimizer state dict 2023-02-08 23:43:04,236 INFO [train.py:1038] (0/4) Loading scheduler state dict 2023-02-08 23:43:04,237 INFO [asr_datamodule.py:420] (0/4) About to get the shuffled train-clean-100, train-clean-360 and train-other-500 cuts 2023-02-08 23:43:04,302 INFO [asr_datamodule.py:224] (0/4) Enable MUSAN 2023-02-08 23:43:04,302 INFO [asr_datamodule.py:225] (0/4) About to get Musan cuts 2023-02-08 23:43:05,967 INFO [asr_datamodule.py:249] (0/4) Enable SpecAugment 2023-02-08 23:43:05,967 INFO [asr_datamodule.py:250] (0/4) Time warp factor: 80 2023-02-08 23:43:05,967 INFO [asr_datamodule.py:260] (0/4) Num frame mask: 10 2023-02-08 23:43:05,967 INFO [asr_datamodule.py:273] (0/4) About to create train dataset 2023-02-08 23:43:05,967 INFO [asr_datamodule.py:300] (0/4) Using DynamicBucketingSampler. 2023-02-08 23:43:06,271 WARNING [train.py:1067] (0/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-02-08 23:43:08,591 INFO [asr_datamodule.py:316] (0/4) About to create train dataloader 2023-02-08 23:43:08,591 INFO [asr_datamodule.py:430] (0/4) About to get dev-clean cuts 2023-02-08 23:43:08,593 INFO [asr_datamodule.py:437] (0/4) About to get dev-other cuts 2023-02-08 23:43:08,594 INFO [asr_datamodule.py:347] (0/4) About to create dev dataset 2023-02-08 23:43:08,942 INFO [asr_datamodule.py:364] (0/4) About to create dev dataloader 2023-02-08 23:43:08,942 INFO [train.py:1122] (0/4) Loading grad scaler state dict 2023-02-08 23:43:20,390 WARNING [train.py:1067] (0/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-02-08 23:43:25,772 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.86 vs. limit=2.0 2023-02-08 23:43:26,064 INFO [train.py:901] (0/4) Epoch 28, batch 0, loss[loss=0.296, simple_loss=0.3346, pruned_loss=0.1286, over 7975.00 frames. ], tot_loss[loss=0.296, simple_loss=0.3346, pruned_loss=0.1286, over 7975.00 frames. ], batch size: 21, lr: 2.71e-03, grad_scale: 8.0 2023-02-08 23:43:26,064 INFO [train.py:926] (0/4) Computing validation loss 2023-02-08 23:43:38,193 INFO [train.py:935] (0/4) Epoch 28, validation: loss=0.1714, simple_loss=0.2712, pruned_loss=0.03579, over 944034.00 frames. 2023-02-08 23:43:38,195 INFO [train.py:936] (0/4) Maximum memory allocated so far is 5970MB 2023-02-08 23:43:48,590 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218250.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:43:55,323 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.44 vs. limit=2.0 2023-02-08 23:43:59,124 WARNING [train.py:1067] (0/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-02-08 23:43:59,192 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=218260.0, num_to_drop=1, layers_to_drop={0} 2023-02-08 23:44:12,579 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.6264, 2.3861, 3.0217, 2.5372, 3.0846, 2.5499, 2.5257, 2.2612], device='cuda:0'), covar=tensor([0.4481, 0.4732, 0.2180, 0.3421, 0.2352, 0.3004, 0.1662, 0.5073], device='cuda:0'), in_proj_covar=tensor([0.0956, 0.1016, 0.0823, 0.0986, 0.1019, 0.0922, 0.0763, 0.0846], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-08 23:44:26,826 INFO [train.py:901] (0/4) Epoch 28, batch 50, loss[loss=0.1862, simple_loss=0.2835, pruned_loss=0.04448, over 8464.00 frames. ], tot_loss[loss=0.2055, simple_loss=0.2893, pruned_loss=0.06078, over 364890.37 frames. ], batch size: 27, lr: 2.71e-03, grad_scale: 8.0 2023-02-08 23:44:44,822 WARNING [train.py:1067] (0/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-02-08 23:44:48,222 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.737e+02 2.503e+02 3.099e+02 3.838e+02 3.677e+03, threshold=6.198e+02, percent-clipped=7.0 2023-02-08 23:45:09,768 INFO [train.py:901] (0/4) Epoch 28, batch 100, loss[loss=0.1674, simple_loss=0.2468, pruned_loss=0.04395, over 7561.00 frames. ], tot_loss[loss=0.203, simple_loss=0.2865, pruned_loss=0.05978, over 646121.59 frames. ], batch size: 18, lr: 2.71e-03, grad_scale: 8.0 2023-02-08 23:45:12,260 WARNING [train.py:1067] (0/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-02-08 23:45:42,220 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218375.0, num_to_drop=1, layers_to_drop={0} 2023-02-08 23:45:52,962 INFO [train.py:901] (0/4) Epoch 28, batch 150, loss[loss=0.2359, simple_loss=0.3088, pruned_loss=0.08148, over 8034.00 frames. ], tot_loss[loss=0.2034, simple_loss=0.2867, pruned_loss=0.06005, over 858215.43 frames. ], batch size: 22, lr: 2.71e-03, grad_scale: 8.0 2023-02-08 23:46:01,161 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218397.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:46:12,821 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.615e+02 2.274e+02 2.796e+02 3.416e+02 5.816e+02, threshold=5.591e+02, percent-clipped=0.0 2023-02-08 23:46:19,646 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218422.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:46:32,308 INFO [train.py:901] (0/4) Epoch 28, batch 200, loss[loss=0.1937, simple_loss=0.2924, pruned_loss=0.04749, over 8349.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2845, pruned_loss=0.05843, over 1025052.38 frames. ], batch size: 24, lr: 2.71e-03, grad_scale: 8.0 2023-02-08 23:46:50,630 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=218462.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:47:10,726 INFO [train.py:901] (0/4) Epoch 28, batch 250, loss[loss=0.1891, simple_loss=0.2563, pruned_loss=0.06098, over 7428.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2837, pruned_loss=0.05733, over 1160062.83 frames. ], batch size: 17, lr: 2.71e-03, grad_scale: 16.0 2023-02-08 23:47:23,075 WARNING [train.py:1067] (0/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-02-08 23:47:31,292 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.545e+02 2.405e+02 2.917e+02 3.543e+02 7.929e+02, threshold=5.833e+02, percent-clipped=6.0 2023-02-08 23:47:33,433 WARNING [train.py:1067] (0/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-02-08 23:47:41,366 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218527.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:47:48,877 INFO [train.py:901] (0/4) Epoch 28, batch 300, loss[loss=0.197, simple_loss=0.2891, pruned_loss=0.05243, over 8493.00 frames. ], tot_loss[loss=0.2008, simple_loss=0.2843, pruned_loss=0.05866, over 1260762.45 frames. ], batch size: 49, lr: 2.71e-03, grad_scale: 16.0 2023-02-08 23:47:53,404 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=218544.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:48:14,321 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=218572.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:48:18,159 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218577.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:48:25,708 INFO [train.py:901] (0/4) Epoch 28, batch 350, loss[loss=0.2105, simple_loss=0.3012, pruned_loss=0.05992, over 8585.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2822, pruned_loss=0.05781, over 1336986.70 frames. ], batch size: 34, lr: 2.71e-03, grad_scale: 16.0 2023-02-08 23:48:28,663 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218592.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:48:29,481 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.6227, 1.5054, 2.0516, 1.7558, 1.7864, 1.6938, 1.4986, 0.8300], device='cuda:0'), covar=tensor([0.7343, 0.5809, 0.2471, 0.4114, 0.3473, 0.4639, 0.3089, 0.5412], device='cuda:0'), in_proj_covar=tensor([0.0959, 0.1017, 0.0823, 0.0986, 0.1018, 0.0922, 0.0763, 0.0846], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-08 23:48:43,881 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.664e+02 2.330e+02 2.853e+02 3.797e+02 9.826e+02, threshold=5.707e+02, percent-clipped=4.0 2023-02-08 23:48:59,299 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218631.0, num_to_drop=1, layers_to_drop={1} 2023-02-08 23:49:04,811 INFO [train.py:901] (0/4) Epoch 28, batch 400, loss[loss=0.1772, simple_loss=0.2641, pruned_loss=0.04513, over 7942.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2825, pruned_loss=0.05786, over 1396074.49 frames. ], batch size: 20, lr: 2.71e-03, grad_scale: 16.0 2023-02-08 23:49:16,388 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.6811, 2.0789, 3.2086, 1.5722, 2.3738, 2.1141, 1.8590, 2.5064], device='cuda:0'), covar=tensor([0.1923, 0.2714, 0.0874, 0.4743, 0.2005, 0.3306, 0.2403, 0.2150], device='cuda:0'), in_proj_covar=tensor([0.0542, 0.0637, 0.0566, 0.0672, 0.0660, 0.0615, 0.0568, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-08 23:49:17,857 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218656.0, num_to_drop=1, layers_to_drop={0} 2023-02-08 23:49:19,981 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218659.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:49:40,570 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218687.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:49:41,135 INFO [train.py:901] (0/4) Epoch 28, batch 450, loss[loss=0.1908, simple_loss=0.2847, pruned_loss=0.04842, over 8508.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.283, pruned_loss=0.05778, over 1451021.87 frames. ], batch size: 28, lr: 2.71e-03, grad_scale: 16.0 2023-02-08 23:49:59,791 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.572e+02 2.362e+02 2.836e+02 3.643e+02 9.062e+02, threshold=5.672e+02, percent-clipped=2.0 2023-02-08 23:50:18,544 INFO [train.py:901] (0/4) Epoch 28, batch 500, loss[loss=0.2342, simple_loss=0.3033, pruned_loss=0.08257, over 7133.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2831, pruned_loss=0.05748, over 1486358.08 frames. ], batch size: 74, lr: 2.71e-03, grad_scale: 16.0 2023-02-08 23:50:57,136 INFO [train.py:901] (0/4) Epoch 28, batch 550, loss[loss=0.1882, simple_loss=0.2682, pruned_loss=0.05414, over 8236.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2819, pruned_loss=0.05717, over 1509607.17 frames. ], batch size: 22, lr: 2.71e-03, grad_scale: 8.0 2023-02-08 23:51:05,270 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.4702, 2.3310, 2.9924, 2.4696, 3.0249, 2.5198, 2.4219, 1.7875], device='cuda:0'), covar=tensor([0.5789, 0.5441, 0.2227, 0.4242, 0.2618, 0.3391, 0.1913, 0.5993], device='cuda:0'), in_proj_covar=tensor([0.0963, 0.1021, 0.0829, 0.0991, 0.1024, 0.0928, 0.0768, 0.0849], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-08 23:51:16,042 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.820e+02 2.392e+02 2.925e+02 3.560e+02 1.211e+03, threshold=5.850e+02, percent-clipped=4.0 2023-02-08 23:51:29,379 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.5935, 2.4140, 1.8025, 2.3168, 2.2072, 1.5912, 2.0444, 2.1015], device='cuda:0'), covar=tensor([0.1451, 0.0447, 0.1272, 0.0650, 0.0731, 0.1688, 0.1132, 0.0958], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0244, 0.0342, 0.0315, 0.0305, 0.0349, 0.0352, 0.0325], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-08 23:51:30,166 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218833.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:51:33,463 INFO [train.py:901] (0/4) Epoch 28, batch 600, loss[loss=0.214, simple_loss=0.2959, pruned_loss=0.06604, over 8336.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.281, pruned_loss=0.05668, over 1533606.42 frames. ], batch size: 26, lr: 2.71e-03, grad_scale: 8.0 2023-02-08 23:51:33,882 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=2.63 vs. limit=5.0 2023-02-08 23:51:53,211 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218858.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:51:56,582 WARNING [train.py:1067] (0/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-02-08 23:52:04,139 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=218871.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:52:06,129 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=3.89 vs. limit=5.0 2023-02-08 23:52:10,236 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.5477, 1.7941, 1.8550, 1.2147, 1.9219, 1.3951, 0.4486, 1.7726], device='cuda:0'), covar=tensor([0.0614, 0.0435, 0.0358, 0.0587, 0.0470, 0.0983, 0.0938, 0.0308], device='cuda:0'), in_proj_covar=tensor([0.0474, 0.0412, 0.0366, 0.0458, 0.0395, 0.0553, 0.0402, 0.0442], device='cuda:0'), out_proj_covar=tensor([1.2539e-04, 1.0696e-04, 9.5266e-05, 1.1974e-04, 1.0344e-04, 1.5414e-04, 1.0732e-04, 1.1585e-04], device='cuda:0') 2023-02-08 23:52:18,546 INFO [train.py:901] (0/4) Epoch 28, batch 650, loss[loss=0.1839, simple_loss=0.2706, pruned_loss=0.04864, over 8017.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.2803, pruned_loss=0.05598, over 1552200.91 frames. ], batch size: 22, lr: 2.71e-03, grad_scale: 8.0 2023-02-08 23:52:40,037 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.733e+02 2.221e+02 2.637e+02 3.403e+02 7.509e+02, threshold=5.274e+02, percent-clipped=1.0 2023-02-08 23:52:41,074 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218915.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:52:54,633 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([5.6832, 5.8679, 5.1021, 2.3930, 5.1107, 5.4242, 5.3913, 5.2674], device='cuda:0'), covar=tensor([0.0507, 0.0355, 0.0864, 0.4357, 0.0752, 0.0790, 0.1115, 0.0622], device='cuda:0'), in_proj_covar=tensor([0.0540, 0.0461, 0.0445, 0.0555, 0.0444, 0.0463, 0.0439, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-08 23:52:55,955 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=218936.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:52:57,374 INFO [train.py:901] (0/4) Epoch 28, batch 700, loss[loss=0.1864, simple_loss=0.2611, pruned_loss=0.05586, over 7975.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2808, pruned_loss=0.05619, over 1566190.66 frames. ], batch size: 21, lr: 2.71e-03, grad_scale: 8.0 2023-02-08 23:52:59,057 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218940.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:53:01,212 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218943.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:53:18,874 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218968.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:53:31,053 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218983.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:53:33,195 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218986.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:53:34,536 INFO [train.py:901] (0/4) Epoch 28, batch 750, loss[loss=0.1806, simple_loss=0.2642, pruned_loss=0.04848, over 7916.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2813, pruned_loss=0.05646, over 1572632.23 frames. ], batch size: 20, lr: 2.71e-03, grad_scale: 8.0 2023-02-08 23:53:46,831 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219002.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:53:55,131 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.438e+02 2.280e+02 2.810e+02 3.388e+02 7.203e+02, threshold=5.620e+02, percent-clipped=6.0 2023-02-08 23:53:55,168 WARNING [train.py:1067] (0/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-02-08 23:54:04,609 WARNING [train.py:1067] (0/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-02-08 23:54:12,582 INFO [train.py:901] (0/4) Epoch 28, batch 800, loss[loss=0.1721, simple_loss=0.2583, pruned_loss=0.04292, over 7789.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.282, pruned_loss=0.05683, over 1580976.92 frames. ], batch size: 19, lr: 2.71e-03, grad_scale: 8.0 2023-02-08 23:54:12,699 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([3.2624, 3.1842, 2.9474, 1.5694, 2.8709, 2.9205, 2.8565, 2.8156], device='cuda:0'), covar=tensor([0.1242, 0.0909, 0.1444, 0.4479, 0.1322, 0.1274, 0.1789, 0.1076], device='cuda:0'), in_proj_covar=tensor([0.0541, 0.0460, 0.0445, 0.0555, 0.0444, 0.0464, 0.0438, 0.0405], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-08 23:54:13,423 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219039.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:54:22,041 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219051.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:54:49,147 INFO [train.py:901] (0/4) Epoch 28, batch 850, loss[loss=0.2083, simple_loss=0.2938, pruned_loss=0.06135, over 8740.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2805, pruned_loss=0.05644, over 1585766.61 frames. ], batch size: 30, lr: 2.71e-03, grad_scale: 8.0 2023-02-08 23:54:50,763 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219090.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:55:10,259 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.529e+02 2.432e+02 3.183e+02 3.929e+02 8.024e+02, threshold=6.365e+02, percent-clipped=6.0 2023-02-08 23:55:11,264 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.8674, 1.6176, 1.9257, 1.7205, 1.8721, 1.9158, 1.7581, 0.8200], device='cuda:0'), covar=tensor([0.6272, 0.4921, 0.2099, 0.3569, 0.2518, 0.3441, 0.2198, 0.5030], device='cuda:0'), in_proj_covar=tensor([0.0965, 0.1023, 0.0830, 0.0991, 0.1022, 0.0928, 0.0769, 0.0848], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-08 23:55:15,822 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 2023-02-08 23:55:27,553 INFO [train.py:901] (0/4) Epoch 28, batch 900, loss[loss=0.1909, simple_loss=0.2914, pruned_loss=0.04519, over 8231.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2805, pruned_loss=0.05616, over 1593382.53 frames. ], batch size: 22, lr: 2.71e-03, grad_scale: 8.0 2023-02-08 23:56:03,866 INFO [train.py:901] (0/4) Epoch 28, batch 950, loss[loss=0.2058, simple_loss=0.2939, pruned_loss=0.05886, over 8689.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2817, pruned_loss=0.05684, over 1602282.37 frames. ], batch size: 34, lr: 2.71e-03, grad_scale: 8.0 2023-02-08 23:56:15,323 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([6.0877, 1.9938, 6.2693, 2.4705, 5.6672, 5.2764, 5.8188, 5.6754], device='cuda:0'), covar=tensor([0.0483, 0.4403, 0.0374, 0.3639, 0.0978, 0.0884, 0.0474, 0.0545], device='cuda:0'), in_proj_covar=tensor([0.0676, 0.0659, 0.0729, 0.0654, 0.0739, 0.0631, 0.0637, 0.0711], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-08 23:56:22,894 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.715e+02 2.524e+02 3.053e+02 4.249e+02 9.516e+02, threshold=6.106e+02, percent-clipped=7.0 2023-02-08 23:56:29,698 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219221.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:56:34,861 WARNING [train.py:1067] (0/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-02-08 23:56:40,765 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([5.7189, 1.8562, 5.9275, 2.0158, 5.2637, 4.9490, 5.4503, 5.3133], device='cuda:0'), covar=tensor([0.0624, 0.4981, 0.0367, 0.4498, 0.1078, 0.0978, 0.0561, 0.0600], device='cuda:0'), in_proj_covar=tensor([0.0676, 0.0659, 0.0727, 0.0653, 0.0739, 0.0631, 0.0636, 0.0711], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-08 23:56:43,597 INFO [train.py:901] (0/4) Epoch 28, batch 1000, loss[loss=0.2239, simple_loss=0.3036, pruned_loss=0.07208, over 8489.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2807, pruned_loss=0.05631, over 1603294.50 frames. ], batch size: 49, lr: 2.71e-03, grad_scale: 8.0 2023-02-08 23:56:46,632 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219242.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:57:04,474 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219267.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:57:11,557 WARNING [train.py:1067] (0/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-02-08 23:57:19,442 INFO [train.py:901] (0/4) Epoch 28, batch 1050, loss[loss=0.2185, simple_loss=0.2972, pruned_loss=0.06986, over 8325.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.282, pruned_loss=0.0571, over 1608197.17 frames. ], batch size: 25, lr: 2.71e-03, grad_scale: 8.0 2023-02-08 23:57:23,688 WARNING [train.py:1067] (0/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-02-08 23:57:33,335 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219307.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:57:38,279 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.554e+02 2.456e+02 2.957e+02 3.788e+02 8.190e+02, threshold=5.915e+02, percent-clipped=1.0 2023-02-08 23:57:47,814 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=219327.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:57:51,441 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219332.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:57:52,299 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-02-08 23:57:56,812 INFO [train.py:901] (0/4) Epoch 28, batch 1100, loss[loss=0.2467, simple_loss=0.3354, pruned_loss=0.07904, over 8436.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2818, pruned_loss=0.05708, over 1609352.41 frames. ], batch size: 49, lr: 2.71e-03, grad_scale: 8.0 2023-02-08 23:58:03,515 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=219346.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:58:13,599 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219360.0, num_to_drop=1, layers_to_drop={1} 2023-02-08 23:58:30,131 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=219383.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:58:33,468 INFO [train.py:901] (0/4) Epoch 28, batch 1150, loss[loss=0.2096, simple_loss=0.2874, pruned_loss=0.06595, over 8328.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2819, pruned_loss=0.05695, over 1612926.11 frames. ], batch size: 26, lr: 2.71e-03, grad_scale: 8.0 2023-02-08 23:58:37,159 WARNING [train.py:1067] (0/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-02-08 23:58:52,525 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.530e+02 2.386e+02 3.071e+02 3.782e+02 1.293e+03, threshold=6.141e+02, percent-clipped=2.0 2023-02-08 23:59:07,136 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=219434.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:59:10,013 INFO [train.py:901] (0/4) Epoch 28, batch 1200, loss[loss=0.1971, simple_loss=0.2766, pruned_loss=0.05879, over 8428.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2807, pruned_loss=0.05633, over 1611297.89 frames. ], batch size: 27, lr: 2.71e-03, grad_scale: 8.0 2023-02-08 23:59:13,064 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219442.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:59:27,998 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219461.0, num_to_drop=0, layers_to_drop=set() 2023-02-08 23:59:47,599 INFO [train.py:901] (0/4) Epoch 28, batch 1250, loss[loss=0.1942, simple_loss=0.2664, pruned_loss=0.06101, over 7924.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2828, pruned_loss=0.05762, over 1613696.01 frames. ], batch size: 20, lr: 2.71e-03, grad_scale: 8.0 2023-02-08 23:59:48,524 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.7254, 1.5520, 1.8877, 1.5681, 1.0701, 1.5234, 2.2716, 1.9866], device='cuda:0'), covar=tensor([0.0481, 0.1301, 0.1677, 0.1492, 0.0633, 0.1562, 0.0636, 0.0634], device='cuda:0'), in_proj_covar=tensor([0.0099, 0.0153, 0.0188, 0.0161, 0.0101, 0.0163, 0.0112, 0.0146], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0') 2023-02-08 23:59:55,038 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219498.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:00:02,357 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219508.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:00:06,606 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.598e+02 2.357e+02 2.809e+02 3.466e+02 7.121e+02, threshold=5.618e+02, percent-clipped=3.0 2023-02-09 00:00:23,412 INFO [train.py:901] (0/4) Epoch 28, batch 1300, loss[loss=0.2167, simple_loss=0.2952, pruned_loss=0.06905, over 7713.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2829, pruned_loss=0.05792, over 1612069.41 frames. ], batch size: 18, lr: 2.71e-03, grad_scale: 8.0 2023-02-09 00:00:31,539 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219549.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:00:42,923 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=219565.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:01:02,349 INFO [train.py:901] (0/4) Epoch 28, batch 1350, loss[loss=0.1643, simple_loss=0.2549, pruned_loss=0.03688, over 8026.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2816, pruned_loss=0.05758, over 1610847.70 frames. ], batch size: 22, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:01:22,008 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.513e+02 2.365e+02 2.856e+02 3.377e+02 7.819e+02, threshold=5.713e+02, percent-clipped=4.0 2023-02-09 00:01:39,687 INFO [train.py:901] (0/4) Epoch 28, batch 1400, loss[loss=0.2366, simple_loss=0.3182, pruned_loss=0.07753, over 8508.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2819, pruned_loss=0.05719, over 1614188.86 frames. ], batch size: 28, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:02:09,818 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219680.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:02:15,338 INFO [train.py:901] (0/4) Epoch 28, batch 1450, loss[loss=0.1723, simple_loss=0.259, pruned_loss=0.04278, over 8258.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2812, pruned_loss=0.05699, over 1614928.54 frames. ], batch size: 22, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:02:23,496 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219698.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:02:25,407 WARNING [train.py:1067] (0/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-02-09 00:02:28,358 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=219704.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 00:02:36,731 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.725e+02 2.298e+02 2.874e+02 3.536e+02 7.746e+02, threshold=5.748e+02, percent-clipped=3.0 2023-02-09 00:02:39,224 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219717.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:02:43,538 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219723.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:02:54,217 INFO [train.py:901] (0/4) Epoch 28, batch 1500, loss[loss=0.1433, simple_loss=0.2275, pruned_loss=0.02951, over 7701.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2803, pruned_loss=0.05658, over 1611657.45 frames. ], batch size: 18, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:02:57,287 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219742.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:03:05,694 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219754.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:03:23,849 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219779.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:03:25,258 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219781.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:03:29,874 INFO [train.py:901] (0/4) Epoch 28, batch 1550, loss[loss=0.2023, simple_loss=0.2875, pruned_loss=0.05851, over 8195.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2813, pruned_loss=0.05718, over 1613628.49 frames. ], batch size: 23, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:03:42,636 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219805.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:03:49,438 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.606e+02 2.435e+02 2.945e+02 3.827e+02 6.900e+02, threshold=5.889e+02, percent-clipped=4.0 2023-02-09 00:03:54,629 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219819.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 00:03:55,294 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.8715, 1.6249, 1.9362, 1.6347, 1.1190, 1.6063, 2.3021, 1.9932], device='cuda:0'), covar=tensor([0.0461, 0.1280, 0.1639, 0.1434, 0.0601, 0.1507, 0.0600, 0.0629], device='cuda:0'), in_proj_covar=tensor([0.0098, 0.0152, 0.0188, 0.0161, 0.0101, 0.0162, 0.0112, 0.0145], device='cuda:0'), out_proj_covar=tensor([0.0006, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0') 2023-02-09 00:04:03,141 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219830.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:04:08,594 INFO [train.py:901] (0/4) Epoch 28, batch 1600, loss[loss=0.1969, simple_loss=0.2904, pruned_loss=0.05171, over 8205.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2815, pruned_loss=0.05672, over 1616144.04 frames. ], batch size: 23, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:04:18,700 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=219852.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:04:35,344 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219875.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:04:44,473 INFO [train.py:901] (0/4) Epoch 28, batch 1650, loss[loss=0.2121, simple_loss=0.2987, pruned_loss=0.06278, over 7967.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2817, pruned_loss=0.05673, over 1616758.70 frames. ], batch size: 21, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:05:02,695 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.357e+02 2.482e+02 2.898e+02 3.443e+02 5.647e+02, threshold=5.797e+02, percent-clipped=0.0 2023-02-09 00:05:20,276 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219936.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:05:21,472 INFO [train.py:901] (0/4) Epoch 28, batch 1700, loss[loss=0.2434, simple_loss=0.3174, pruned_loss=0.08471, over 8256.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2822, pruned_loss=0.0571, over 1617528.50 frames. ], batch size: 24, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:05:39,132 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219961.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:05:43,293 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219967.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:05:57,790 INFO [train.py:901] (0/4) Epoch 28, batch 1750, loss[loss=0.1933, simple_loss=0.2809, pruned_loss=0.05286, over 8468.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2824, pruned_loss=0.05773, over 1613424.06 frames. ], batch size: 29, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:06:06,483 INFO [checkpoint.py:75] (0/4) Saving checkpoint to pruned_transducer_stateless7_streaming/exp/v1/checkpoint-220000.pt 2023-02-09 00:06:17,589 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.415e+02 2.339e+02 2.848e+02 3.606e+02 1.047e+03, threshold=5.695e+02, percent-clipped=4.0 2023-02-09 00:06:34,442 INFO [train.py:901] (0/4) Epoch 28, batch 1800, loss[loss=0.1724, simple_loss=0.251, pruned_loss=0.04689, over 7441.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2818, pruned_loss=0.05751, over 1610054.88 frames. ], batch size: 17, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:07:02,974 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220075.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 00:07:11,952 INFO [train.py:901] (0/4) Epoch 28, batch 1850, loss[loss=0.1872, simple_loss=0.2839, pruned_loss=0.04523, over 8510.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2815, pruned_loss=0.05729, over 1609669.63 frames. ], batch size: 28, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:07:20,533 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220100.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 00:07:23,916 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220105.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:07:30,314 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.818e+02 2.328e+02 2.682e+02 3.608e+02 8.535e+02, threshold=5.364e+02, percent-clipped=7.0 2023-02-09 00:07:31,755 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220116.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:07:38,071 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=220125.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:07:47,183 INFO [train.py:901] (0/4) Epoch 28, batch 1900, loss[loss=0.1691, simple_loss=0.2561, pruned_loss=0.0411, over 8082.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2826, pruned_loss=0.05763, over 1607992.86 frames. ], batch size: 21, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:08:19,381 WARNING [train.py:1067] (0/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-02-09 00:08:25,774 INFO [train.py:901] (0/4) Epoch 28, batch 1950, loss[loss=0.1956, simple_loss=0.2808, pruned_loss=0.05524, over 8659.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2827, pruned_loss=0.05764, over 1615282.15 frames. ], batch size: 34, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:08:33,000 WARNING [train.py:1067] (0/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-02-09 00:08:44,726 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.518e+02 2.461e+02 2.916e+02 3.869e+02 7.609e+02, threshold=5.833e+02, percent-clipped=8.0 2023-02-09 00:08:48,264 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=220219.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:08:51,076 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220223.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:08:53,655 WARNING [train.py:1067] (0/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-02-09 00:09:01,371 INFO [train.py:901] (0/4) Epoch 28, batch 2000, loss[loss=0.185, simple_loss=0.2778, pruned_loss=0.04611, over 8659.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2814, pruned_loss=0.05697, over 1611478.66 frames. ], batch size: 34, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:09:02,868 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220240.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:09:07,805 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220247.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:09:08,589 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220248.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:09:29,136 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220276.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:09:37,448 INFO [train.py:901] (0/4) Epoch 28, batch 2050, loss[loss=0.1587, simple_loss=0.2234, pruned_loss=0.04699, over 7215.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2817, pruned_loss=0.05738, over 1609546.39 frames. ], batch size: 16, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:09:58,202 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.689e+02 2.398e+02 2.757e+02 3.324e+02 6.340e+02, threshold=5.514e+02, percent-clipped=2.0 2023-02-09 00:10:04,008 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.0528, 1.8484, 2.2540, 2.0131, 2.2987, 2.1768, 1.9984, 1.1996], device='cuda:0'), covar=tensor([0.6445, 0.5493, 0.2402, 0.4102, 0.2644, 0.3520, 0.2208, 0.5832], device='cuda:0'), in_proj_covar=tensor([0.0970, 0.1029, 0.0835, 0.0997, 0.1028, 0.0934, 0.0773, 0.0853], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 00:10:12,780 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220334.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:10:15,446 INFO [train.py:901] (0/4) Epoch 28, batch 2100, loss[loss=0.1915, simple_loss=0.2794, pruned_loss=0.05175, over 8484.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2819, pruned_loss=0.05719, over 1613570.54 frames. ], batch size: 27, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:10:20,361 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-02-09 00:10:42,395 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0 2023-02-09 00:10:51,259 INFO [train.py:901] (0/4) Epoch 28, batch 2150, loss[loss=0.2516, simple_loss=0.321, pruned_loss=0.09108, over 7158.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2813, pruned_loss=0.05681, over 1619032.83 frames. ], batch size: 71, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:11:11,486 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.499e+02 2.504e+02 2.973e+02 4.041e+02 1.001e+03, threshold=5.945e+02, percent-clipped=8.0 2023-02-09 00:11:28,330 INFO [train.py:901] (0/4) Epoch 28, batch 2200, loss[loss=0.2231, simple_loss=0.2951, pruned_loss=0.07548, over 8244.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2818, pruned_loss=0.05697, over 1620314.54 frames. ], batch size: 22, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:11:36,284 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=220449.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:11:44,024 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=220460.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:12:03,388 INFO [train.py:901] (0/4) Epoch 28, batch 2250, loss[loss=0.1985, simple_loss=0.2958, pruned_loss=0.05054, over 8105.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2812, pruned_loss=0.05632, over 1615471.78 frames. ], batch size: 23, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:12:09,263 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220496.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:12:15,028 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=3.66 vs. limit=5.0 2023-02-09 00:12:22,282 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.608e+02 2.331e+02 2.835e+02 3.325e+02 7.200e+02, threshold=5.671e+02, percent-clipped=3.0 2023-02-09 00:12:27,588 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220521.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:12:41,523 INFO [train.py:901] (0/4) Epoch 28, batch 2300, loss[loss=0.1959, simple_loss=0.289, pruned_loss=0.05134, over 8365.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2816, pruned_loss=0.05639, over 1621164.04 frames. ], batch size: 49, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:12:44,383 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([3.8158, 1.3745, 3.9786, 1.3943, 3.5300, 3.2704, 3.6238, 3.5143], device='cuda:0'), covar=tensor([0.0698, 0.4875, 0.0669, 0.4688, 0.1230, 0.1120, 0.0713, 0.0803], device='cuda:0'), in_proj_covar=tensor([0.0682, 0.0670, 0.0739, 0.0664, 0.0752, 0.0639, 0.0646, 0.0722], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 00:12:51,192 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.0895, 1.7666, 3.9761, 1.9364, 2.6087, 4.4770, 4.6323, 3.9311], device='cuda:0'), covar=tensor([0.1308, 0.1961, 0.0364, 0.2015, 0.1246, 0.0211, 0.0448, 0.0537], device='cuda:0'), in_proj_covar=tensor([0.0308, 0.0327, 0.0293, 0.0321, 0.0323, 0.0277, 0.0441, 0.0308], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 00:12:59,743 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220564.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:13:06,023 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220573.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:13:07,464 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220575.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:13:16,600 INFO [train.py:901] (0/4) Epoch 28, batch 2350, loss[loss=0.1815, simple_loss=0.2773, pruned_loss=0.04291, over 8253.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2809, pruned_loss=0.05583, over 1621554.49 frames. ], batch size: 24, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:13:18,225 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220590.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:13:18,798 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=220591.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:13:35,686 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.370e+02 2.329e+02 2.956e+02 3.826e+02 8.837e+02, threshold=5.912e+02, percent-clipped=4.0 2023-02-09 00:13:36,675 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220615.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:13:40,250 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=220620.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:13:53,608 INFO [train.py:901] (0/4) Epoch 28, batch 2400, loss[loss=0.1998, simple_loss=0.2873, pruned_loss=0.05617, over 8190.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.2801, pruned_loss=0.05545, over 1619166.02 frames. ], batch size: 23, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:14:16,673 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220669.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:14:29,716 INFO [train.py:901] (0/4) Epoch 28, batch 2450, loss[loss=0.2272, simple_loss=0.3243, pruned_loss=0.06504, over 8667.00 frames. ], tot_loss[loss=0.1953, simple_loss=0.2797, pruned_loss=0.05549, over 1617024.14 frames. ], batch size: 39, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:14:42,734 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220706.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:14:48,792 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.600e+02 2.507e+02 3.309e+02 3.917e+02 8.053e+02, threshold=6.618e+02, percent-clipped=4.0 2023-02-09 00:15:03,106 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220735.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:15:05,107 INFO [train.py:901] (0/4) Epoch 28, batch 2500, loss[loss=0.162, simple_loss=0.2483, pruned_loss=0.03781, over 7975.00 frames. ], tot_loss[loss=0.196, simple_loss=0.2799, pruned_loss=0.05605, over 1617653.02 frames. ], batch size: 21, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:15:15,106 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.6774, 2.7947, 2.5845, 4.2547, 1.7169, 2.0548, 2.8229, 2.9358], device='cuda:0'), covar=tensor([0.0611, 0.0728, 0.0693, 0.0162, 0.0983, 0.1195, 0.0744, 0.0767], device='cuda:0'), in_proj_covar=tensor([0.0229, 0.0193, 0.0242, 0.0211, 0.0202, 0.0245, 0.0249, 0.0204], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], device='cuda:0') 2023-02-09 00:15:42,749 INFO [train.py:901] (0/4) Epoch 28, batch 2550, loss[loss=0.1972, simple_loss=0.2678, pruned_loss=0.06333, over 7805.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2812, pruned_loss=0.05655, over 1623033.75 frames. ], batch size: 20, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:16:02,766 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.451e+02 2.505e+02 3.011e+02 3.782e+02 1.017e+03, threshold=6.023e+02, percent-clipped=3.0 2023-02-09 00:16:06,730 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220820.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:16:14,507 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220831.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:16:19,228 INFO [train.py:901] (0/4) Epoch 28, batch 2600, loss[loss=0.173, simple_loss=0.2595, pruned_loss=0.04324, over 8294.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2808, pruned_loss=0.05616, over 1617615.46 frames. ], batch size: 23, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:16:20,651 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220840.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:16:24,305 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220845.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:16:32,242 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220856.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:16:57,370 INFO [train.py:901] (0/4) Epoch 28, batch 2650, loss[loss=0.1876, simple_loss=0.2769, pruned_loss=0.04914, over 8436.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2809, pruned_loss=0.05633, over 1618781.93 frames. ], batch size: 49, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:17:12,897 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.1226, 3.6129, 2.3302, 2.9803, 2.8377, 2.3000, 3.0700, 3.0147], device='cuda:0'), covar=tensor([0.1789, 0.0469, 0.1173, 0.0782, 0.0817, 0.1373, 0.1033, 0.1230], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0242, 0.0339, 0.0311, 0.0301, 0.0345, 0.0346, 0.0321], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 00:17:16,288 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.526e+02 2.381e+02 2.801e+02 3.642e+02 5.464e+02, threshold=5.602e+02, percent-clipped=0.0 2023-02-09 00:17:17,797 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=220917.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:17:21,283 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220922.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:17:32,916 INFO [train.py:901] (0/4) Epoch 28, batch 2700, loss[loss=0.2024, simple_loss=0.2907, pruned_loss=0.05702, over 8543.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2807, pruned_loss=0.05646, over 1618232.27 frames. ], batch size: 49, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:17:50,491 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220962.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:18:09,034 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220987.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:18:09,531 INFO [train.py:901] (0/4) Epoch 28, batch 2750, loss[loss=0.2087, simple_loss=0.2902, pruned_loss=0.06362, over 8472.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2814, pruned_loss=0.05671, over 1619039.57 frames. ], batch size: 25, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:18:11,791 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220991.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:18:29,681 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=221013.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:18:31,038 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.597e+02 2.419e+02 2.908e+02 3.517e+02 7.342e+02, threshold=5.816e+02, percent-clipped=5.0 2023-02-09 00:18:31,980 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221016.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:18:43,270 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221032.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:18:47,247 INFO [train.py:901] (0/4) Epoch 28, batch 2800, loss[loss=0.2039, simple_loss=0.2906, pruned_loss=0.05864, over 8573.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2822, pruned_loss=0.0571, over 1618897.49 frames. ], batch size: 31, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:18:57,821 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.9179, 1.7399, 2.5532, 1.6717, 1.4627, 2.4605, 0.4679, 1.5924], device='cuda:0'), covar=tensor([0.1438, 0.1232, 0.0331, 0.1035, 0.2047, 0.0435, 0.1924, 0.1215], device='cuda:0'), in_proj_covar=tensor([0.0201, 0.0205, 0.0136, 0.0224, 0.0277, 0.0146, 0.0174, 0.0199], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 00:19:22,679 INFO [train.py:901] (0/4) Epoch 28, batch 2850, loss[loss=0.1787, simple_loss=0.2794, pruned_loss=0.03906, over 8245.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2822, pruned_loss=0.05665, over 1619132.58 frames. ], batch size: 24, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:19:43,237 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.362e+02 2.366e+02 2.856e+02 3.627e+02 6.501e+02, threshold=5.713e+02, percent-clipped=2.0 2023-02-09 00:19:53,977 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221128.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:20:00,673 INFO [train.py:901] (0/4) Epoch 28, batch 2900, loss[loss=0.1982, simple_loss=0.286, pruned_loss=0.05518, over 8647.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2813, pruned_loss=0.05675, over 1616715.40 frames. ], batch size: 34, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:20:04,302 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.5009, 2.0651, 3.9150, 1.5069, 2.7408, 2.0627, 1.6690, 2.7888], device='cuda:0'), covar=tensor([0.2345, 0.3123, 0.0971, 0.5157, 0.2265, 0.3673, 0.2804, 0.2654], device='cuda:0'), in_proj_covar=tensor([0.0541, 0.0639, 0.0566, 0.0673, 0.0663, 0.0612, 0.0566, 0.0647], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 00:20:32,275 WARNING [train.py:1067] (0/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-02-09 00:20:33,728 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=221184.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:20:36,504 INFO [train.py:901] (0/4) Epoch 28, batch 2950, loss[loss=0.1739, simple_loss=0.2656, pruned_loss=0.04112, over 8106.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2814, pruned_loss=0.05696, over 1614921.43 frames. ], batch size: 23, lr: 2.70e-03, grad_scale: 8.0 2023-02-09 00:20:37,474 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.4501, 1.8103, 3.2131, 1.3541, 2.5218, 1.9583, 1.5299, 2.5262], device='cuda:0'), covar=tensor([0.2254, 0.3100, 0.0853, 0.5411, 0.1924, 0.3599, 0.2959, 0.2232], device='cuda:0'), in_proj_covar=tensor([0.0541, 0.0640, 0.0567, 0.0674, 0.0663, 0.0612, 0.0566, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 00:20:55,458 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.514e+02 2.299e+02 2.993e+02 3.879e+02 1.208e+03, threshold=5.985e+02, percent-clipped=10.0 2023-02-09 00:21:01,652 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=4.29 vs. limit=5.0 2023-02-09 00:21:13,554 INFO [train.py:901] (0/4) Epoch 28, batch 3000, loss[loss=0.228, simple_loss=0.3153, pruned_loss=0.07031, over 8500.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2806, pruned_loss=0.05643, over 1613096.98 frames. ], batch size: 26, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:21:13,555 INFO [train.py:926] (0/4) Computing validation loss 2023-02-09 00:21:31,976 INFO [train.py:935] (0/4) Epoch 28, validation: loss=0.1712, simple_loss=0.2708, pruned_loss=0.03578, over 944034.00 frames. 2023-02-09 00:21:31,978 INFO [train.py:936] (0/4) Maximum memory allocated so far is 6461MB 2023-02-09 00:21:43,325 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([4.1626, 1.4157, 4.3623, 1.4793, 3.8622, 3.6158, 3.9692, 3.8573], device='cuda:0'), covar=tensor([0.0649, 0.4843, 0.0548, 0.4486, 0.1157, 0.1036, 0.0591, 0.0668], device='cuda:0'), in_proj_covar=tensor([0.0674, 0.0660, 0.0725, 0.0653, 0.0739, 0.0630, 0.0637, 0.0709], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 00:21:54,456 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=221266.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:22:10,149 INFO [train.py:901] (0/4) Epoch 28, batch 3050, loss[loss=0.2069, simple_loss=0.2864, pruned_loss=0.06366, over 8656.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2808, pruned_loss=0.05704, over 1613487.55 frames. ], batch size: 34, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:22:10,377 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221288.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:22:11,515 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.84 vs. limit=2.0 2023-02-09 00:22:18,089 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221299.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:22:24,685 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.7662, 2.2128, 3.1447, 1.6448, 2.5510, 2.1037, 1.9599, 2.5354], device='cuda:0'), covar=tensor([0.1980, 0.2519, 0.0981, 0.4606, 0.1990, 0.3227, 0.2439, 0.2411], device='cuda:0'), in_proj_covar=tensor([0.0538, 0.0635, 0.0563, 0.0669, 0.0658, 0.0607, 0.0563, 0.0643], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 00:22:28,234 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221313.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:22:29,365 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.568e+02 2.361e+02 2.830e+02 3.600e+02 1.199e+03, threshold=5.660e+02, percent-clipped=4.0 2023-02-09 00:22:45,363 INFO [train.py:901] (0/4) Epoch 28, batch 3100, loss[loss=0.192, simple_loss=0.2824, pruned_loss=0.05083, over 7931.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2811, pruned_loss=0.05669, over 1620113.01 frames. ], batch size: 20, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:23:18,469 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221381.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:23:20,627 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221384.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:23:23,256 INFO [train.py:901] (0/4) Epoch 28, batch 3150, loss[loss=0.1696, simple_loss=0.2558, pruned_loss=0.04175, over 7799.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2802, pruned_loss=0.05608, over 1614450.94 frames. ], batch size: 19, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:23:31,509 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.3646, 1.6571, 4.3639, 2.0338, 2.5727, 4.9578, 5.1040, 4.3416], device='cuda:0'), covar=tensor([0.1137, 0.1964, 0.0311, 0.2052, 0.1275, 0.0208, 0.0325, 0.0577], device='cuda:0'), in_proj_covar=tensor([0.0308, 0.0329, 0.0294, 0.0323, 0.0326, 0.0278, 0.0443, 0.0309], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 00:23:34,619 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.8240, 2.4468, 1.8630, 2.3329, 2.1932, 1.6652, 2.1581, 2.3412], device='cuda:0'), covar=tensor([0.1515, 0.0569, 0.1398, 0.0634, 0.0813, 0.1807, 0.1037, 0.0986], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0246, 0.0344, 0.0315, 0.0305, 0.0350, 0.0351, 0.0325], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 00:23:38,999 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221409.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:23:43,027 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.782e+02 2.344e+02 3.031e+02 3.872e+02 9.124e+02, threshold=6.062e+02, percent-clipped=5.0 2023-02-09 00:24:00,282 INFO [train.py:901] (0/4) Epoch 28, batch 3200, loss[loss=0.1766, simple_loss=0.2594, pruned_loss=0.04691, over 8088.00 frames. ], tot_loss[loss=0.1951, simple_loss=0.2787, pruned_loss=0.05574, over 1611026.98 frames. ], batch size: 21, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:24:21,671 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.9736, 1.7573, 2.6323, 1.5759, 2.2426, 2.9209, 2.8914, 2.6292], device='cuda:0'), covar=tensor([0.0982, 0.1491, 0.0680, 0.1806, 0.1915, 0.0275, 0.0795, 0.0486], device='cuda:0'), in_proj_covar=tensor([0.0308, 0.0329, 0.0294, 0.0323, 0.0325, 0.0278, 0.0442, 0.0309], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 00:24:35,521 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.1301, 1.6255, 3.5677, 1.5989, 2.5533, 3.9475, 4.0311, 3.3768], device='cuda:0'), covar=tensor([0.1162, 0.1814, 0.0277, 0.2056, 0.0990, 0.0199, 0.0467, 0.0537], device='cuda:0'), in_proj_covar=tensor([0.0307, 0.0328, 0.0294, 0.0323, 0.0325, 0.0278, 0.0441, 0.0309], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 00:24:36,745 INFO [train.py:901] (0/4) Epoch 28, batch 3250, loss[loss=0.1931, simple_loss=0.2829, pruned_loss=0.05168, over 8521.00 frames. ], tot_loss[loss=0.1954, simple_loss=0.2792, pruned_loss=0.05584, over 1614488.84 frames. ], batch size: 28, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:24:56,686 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.554e+02 2.359e+02 2.800e+02 3.771e+02 8.910e+02, threshold=5.600e+02, percent-clipped=3.0 2023-02-09 00:25:04,051 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221525.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:25:12,970 INFO [train.py:901] (0/4) Epoch 28, batch 3300, loss[loss=0.1993, simple_loss=0.2878, pruned_loss=0.05537, over 8526.00 frames. ], tot_loss[loss=0.196, simple_loss=0.2792, pruned_loss=0.05637, over 1611675.03 frames. ], batch size: 49, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:25:25,047 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221555.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:25:42,956 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221580.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:25:48,453 INFO [train.py:901] (0/4) Epoch 28, batch 3350, loss[loss=0.2024, simple_loss=0.291, pruned_loss=0.05689, over 8326.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2798, pruned_loss=0.05687, over 1606553.50 frames. ], batch size: 25, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:26:09,950 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.399e+02 2.531e+02 3.062e+02 3.663e+02 8.444e+02, threshold=6.124e+02, percent-clipped=3.0 2023-02-09 00:26:21,012 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.6479, 1.8784, 2.0107, 1.3941, 2.1073, 1.5492, 0.5913, 1.9292], device='cuda:0'), covar=tensor([0.0663, 0.0442, 0.0340, 0.0658, 0.0452, 0.1010, 0.1080, 0.0340], device='cuda:0'), in_proj_covar=tensor([0.0475, 0.0412, 0.0365, 0.0461, 0.0396, 0.0552, 0.0404, 0.0444], device='cuda:0'), out_proj_covar=tensor([1.2578e-04, 1.0687e-04, 9.5010e-05, 1.2038e-04, 1.0348e-04, 1.5392e-04, 1.0800e-04, 1.1642e-04], device='cuda:0') 2023-02-09 00:26:26,097 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221637.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:26:26,622 INFO [train.py:901] (0/4) Epoch 28, batch 3400, loss[loss=0.1699, simple_loss=0.2533, pruned_loss=0.04326, over 7816.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2801, pruned_loss=0.05684, over 1608273.32 frames. ], batch size: 20, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:26:43,893 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221662.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:27:02,344 INFO [train.py:901] (0/4) Epoch 28, batch 3450, loss[loss=0.2083, simple_loss=0.2927, pruned_loss=0.06191, over 8349.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.2788, pruned_loss=0.05605, over 1604426.71 frames. ], batch size: 49, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:27:21,424 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.552e+02 2.306e+02 2.763e+02 3.583e+02 8.756e+02, threshold=5.526e+02, percent-clipped=3.0 2023-02-09 00:27:39,503 INFO [train.py:901] (0/4) Epoch 28, batch 3500, loss[loss=0.1632, simple_loss=0.2348, pruned_loss=0.04582, over 7692.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.2797, pruned_loss=0.05628, over 1607155.77 frames. ], batch size: 18, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:27:56,086 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.27 vs. limit=2.0 2023-02-09 00:28:03,519 WARNING [train.py:1067] (0/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-02-09 00:28:07,252 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221776.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:28:15,815 INFO [train.py:901] (0/4) Epoch 28, batch 3550, loss[loss=0.2211, simple_loss=0.3205, pruned_loss=0.06085, over 8492.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.2794, pruned_loss=0.05619, over 1602427.12 frames. ], batch size: 26, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:28:35,058 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 2023-02-09 00:28:35,275 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.790e+02 2.405e+02 2.949e+02 3.672e+02 8.337e+02, threshold=5.897e+02, percent-clipped=3.0 2023-02-09 00:28:52,589 INFO [train.py:901] (0/4) Epoch 28, batch 3600, loss[loss=0.2102, simple_loss=0.2922, pruned_loss=0.06413, over 8501.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2813, pruned_loss=0.05711, over 1606910.89 frames. ], batch size: 26, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:29:08,876 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 2023-02-09 00:29:09,928 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([5.7645, 5.9094, 5.0535, 2.5310, 5.1744, 5.6339, 5.3676, 5.4777], device='cuda:0'), covar=tensor([0.0509, 0.0367, 0.0927, 0.4739, 0.0755, 0.0737, 0.1032, 0.0583], device='cuda:0'), in_proj_covar=tensor([0.0542, 0.0457, 0.0448, 0.0560, 0.0444, 0.0465, 0.0441, 0.0406], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 00:29:15,424 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=221869.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:29:28,482 INFO [train.py:901] (0/4) Epoch 28, batch 3650, loss[loss=0.2051, simple_loss=0.2806, pruned_loss=0.06479, over 7966.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2824, pruned_loss=0.05791, over 1608775.83 frames. ], batch size: 21, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:29:42,354 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221908.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:29:47,071 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.635e+02 2.399e+02 3.022e+02 3.885e+02 8.966e+02, threshold=6.044e+02, percent-clipped=2.0 2023-02-09 00:30:02,964 INFO [train.py:901] (0/4) Epoch 28, batch 3700, loss[loss=0.198, simple_loss=0.2701, pruned_loss=0.06296, over 7422.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2823, pruned_loss=0.05788, over 1611396.39 frames. ], batch size: 17, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:30:05,071 WARNING [train.py:1067] (0/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-02-09 00:30:38,665 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221984.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:30:41,397 INFO [train.py:901] (0/4) Epoch 28, batch 3750, loss[loss=0.1759, simple_loss=0.2616, pruned_loss=0.04511, over 7542.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2828, pruned_loss=0.0578, over 1612069.50 frames. ], batch size: 18, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:30:49,687 INFO [checkpoint.py:75] (0/4) Saving checkpoint to pruned_transducer_stateless7_streaming/exp/v1/checkpoint-222000.pt 2023-02-09 00:31:01,421 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.685e+02 2.662e+02 3.142e+02 4.083e+02 1.270e+03, threshold=6.284e+02, percent-clipped=8.0 2023-02-09 00:31:04,457 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.9512, 1.5667, 3.4561, 1.5127, 2.4054, 3.8660, 3.9736, 3.2908], device='cuda:0'), covar=tensor([0.1266, 0.1938, 0.0360, 0.2311, 0.1124, 0.0230, 0.0524, 0.0539], device='cuda:0'), in_proj_covar=tensor([0.0310, 0.0330, 0.0295, 0.0325, 0.0326, 0.0279, 0.0445, 0.0311], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 00:31:10,203 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222027.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:31:17,966 INFO [train.py:901] (0/4) Epoch 28, batch 3800, loss[loss=0.2042, simple_loss=0.2831, pruned_loss=0.06262, over 7255.00 frames. ], tot_loss[loss=0.2, simple_loss=0.2833, pruned_loss=0.05839, over 1610243.42 frames. ], batch size: 16, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:31:55,281 INFO [train.py:901] (0/4) Epoch 28, batch 3850, loss[loss=0.1866, simple_loss=0.2596, pruned_loss=0.0568, over 8093.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2837, pruned_loss=0.05878, over 1606633.22 frames. ], batch size: 21, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:32:11,760 WARNING [train.py:1067] (0/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-02-09 00:32:13,842 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.497e+02 2.208e+02 2.768e+02 3.453e+02 7.901e+02, threshold=5.537e+02, percent-clipped=1.0 2023-02-09 00:32:15,426 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([4.2507, 4.1669, 3.8616, 1.9038, 3.7820, 3.8502, 3.7439, 3.6906], device='cuda:0'), covar=tensor([0.0726, 0.0561, 0.1116, 0.4858, 0.0906, 0.0938, 0.1351, 0.0790], device='cuda:0'), in_proj_covar=tensor([0.0545, 0.0459, 0.0448, 0.0560, 0.0444, 0.0467, 0.0442, 0.0408], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 00:32:17,497 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=222120.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:32:30,078 INFO [train.py:901] (0/4) Epoch 28, batch 3900, loss[loss=0.2084, simple_loss=0.2936, pruned_loss=0.06166, over 8469.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2837, pruned_loss=0.05883, over 1606805.16 frames. ], batch size: 27, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:32:30,479 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 2023-02-09 00:33:06,448 INFO [train.py:901] (0/4) Epoch 28, batch 3950, loss[loss=0.1643, simple_loss=0.2481, pruned_loss=0.04025, over 8134.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2836, pruned_loss=0.05851, over 1609648.82 frames. ], batch size: 22, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:33:17,880 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222203.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:33:20,808 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.6373, 2.7079, 1.9992, 2.3767, 2.1495, 1.8191, 2.1315, 2.2742], device='cuda:0'), covar=tensor([0.1554, 0.0421, 0.1148, 0.0642, 0.0809, 0.1510, 0.1127, 0.1039], device='cuda:0'), in_proj_covar=tensor([0.0358, 0.0245, 0.0342, 0.0314, 0.0304, 0.0347, 0.0350, 0.0323], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 00:33:26,088 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.641e+02 2.338e+02 2.821e+02 3.606e+02 1.107e+03, threshold=5.643e+02, percent-clipped=4.0 2023-02-09 00:33:31,796 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222223.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:33:35,387 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.3160, 1.2365, 1.5758, 1.0207, 1.0607, 1.5776, 0.6428, 1.2180], device='cuda:0'), covar=tensor([0.1326, 0.1083, 0.0439, 0.0884, 0.1916, 0.0424, 0.1555, 0.1118], device='cuda:0'), in_proj_covar=tensor([0.0202, 0.0207, 0.0137, 0.0225, 0.0279, 0.0147, 0.0176, 0.0200], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 00:33:37,688 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-02-09 00:33:40,240 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222235.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:33:42,146 INFO [train.py:901] (0/4) Epoch 28, batch 4000, loss[loss=0.2204, simple_loss=0.308, pruned_loss=0.06642, over 8477.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2844, pruned_loss=0.05915, over 1610894.82 frames. ], batch size: 26, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:33:43,704 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222240.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:33:51,741 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=222252.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:34:01,389 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222265.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:34:17,514 INFO [train.py:901] (0/4) Epoch 28, batch 4050, loss[loss=0.2177, simple_loss=0.3042, pruned_loss=0.06559, over 8290.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2853, pruned_loss=0.05916, over 1615082.72 frames. ], batch size: 23, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:34:38,324 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.744e+02 2.414e+02 3.092e+02 4.009e+02 1.246e+03, threshold=6.184e+02, percent-clipped=7.0 2023-02-09 00:34:54,244 INFO [train.py:901] (0/4) Epoch 28, batch 4100, loss[loss=0.179, simple_loss=0.2568, pruned_loss=0.05063, over 7545.00 frames. ], tot_loss[loss=0.2013, simple_loss=0.2845, pruned_loss=0.05904, over 1610454.72 frames. ], batch size: 18, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:35:14,884 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222367.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:35:17,634 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=222371.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:35:29,384 INFO [train.py:901] (0/4) Epoch 28, batch 4150, loss[loss=0.2012, simple_loss=0.2887, pruned_loss=0.05686, over 8028.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2832, pruned_loss=0.05808, over 1610359.00 frames. ], batch size: 22, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:35:49,105 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.589e+02 2.330e+02 2.692e+02 3.176e+02 6.436e+02, threshold=5.384e+02, percent-clipped=1.0 2023-02-09 00:36:07,146 INFO [train.py:901] (0/4) Epoch 28, batch 4200, loss[loss=0.2236, simple_loss=0.3048, pruned_loss=0.07119, over 8492.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2826, pruned_loss=0.05753, over 1610192.25 frames. ], batch size: 29, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:36:14,730 WARNING [train.py:1067] (0/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-02-09 00:36:37,845 WARNING [train.py:1067] (0/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-02-09 00:36:41,351 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222486.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:36:42,590 INFO [train.py:901] (0/4) Epoch 28, batch 4250, loss[loss=0.1868, simple_loss=0.274, pruned_loss=0.04982, over 7967.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2824, pruned_loss=0.0574, over 1605790.74 frames. ], batch size: 21, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:36:44,947 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222491.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:36:53,806 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 2023-02-09 00:37:00,853 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.734e+02 2.539e+02 3.193e+02 4.198e+02 8.289e+02, threshold=6.386e+02, percent-clipped=5.0 2023-02-09 00:37:01,767 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222516.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:37:18,009 INFO [train.py:901] (0/4) Epoch 28, batch 4300, loss[loss=0.2172, simple_loss=0.3056, pruned_loss=0.06443, over 8348.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2832, pruned_loss=0.05796, over 1607590.58 frames. ], batch size: 26, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:37:20,160 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222541.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:37:25,005 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=222547.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:37:39,459 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=222567.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:37:54,080 INFO [train.py:901] (0/4) Epoch 28, batch 4350, loss[loss=0.1986, simple_loss=0.2783, pruned_loss=0.05947, over 8072.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2825, pruned_loss=0.05765, over 1610175.12 frames. ], batch size: 21, lr: 2.69e-03, grad_scale: 8.0 2023-02-09 00:37:54,230 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222588.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:38:11,695 WARNING [train.py:1067] (0/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-02-09 00:38:13,114 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.579e+02 2.501e+02 2.979e+02 3.614e+02 7.360e+02, threshold=5.959e+02, percent-clipped=2.0 2023-02-09 00:38:18,814 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222623.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:38:29,011 INFO [train.py:901] (0/4) Epoch 28, batch 4400, loss[loss=0.2005, simple_loss=0.2796, pruned_loss=0.06067, over 8601.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2818, pruned_loss=0.05735, over 1609770.23 frames. ], batch size: 39, lr: 2.69e-03, grad_scale: 16.0 2023-02-09 00:38:36,668 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222648.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:38:40,087 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.0886, 2.2916, 3.1219, 1.9720, 2.6934, 2.3639, 2.1569, 2.6087], device='cuda:0'), covar=tensor([0.1427, 0.2311, 0.0753, 0.3569, 0.1407, 0.2523, 0.1891, 0.2005], device='cuda:0'), in_proj_covar=tensor([0.0540, 0.0640, 0.0568, 0.0675, 0.0664, 0.0613, 0.0567, 0.0647], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 00:38:47,084 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222662.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:38:54,372 WARNING [train.py:1067] (0/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-02-09 00:39:01,980 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222682.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:39:04,790 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.1300, 1.4884, 1.7517, 1.4515, 0.9658, 1.5601, 1.7238, 1.5928], device='cuda:0'), covar=tensor([0.0507, 0.1205, 0.1623, 0.1442, 0.0583, 0.1414, 0.0686, 0.0662], device='cuda:0'), in_proj_covar=tensor([0.0099, 0.0153, 0.0190, 0.0162, 0.0101, 0.0163, 0.0113, 0.0146], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0') 2023-02-09 00:39:05,979 INFO [train.py:901] (0/4) Epoch 28, batch 4450, loss[loss=0.1872, simple_loss=0.2705, pruned_loss=0.05193, over 7248.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2835, pruned_loss=0.0581, over 1613068.89 frames. ], batch size: 16, lr: 2.69e-03, grad_scale: 16.0 2023-02-09 00:39:24,970 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.632e+02 2.353e+02 2.798e+02 3.446e+02 6.111e+02, threshold=5.597e+02, percent-clipped=1.0 2023-02-09 00:39:38,914 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=3.93 vs. limit=5.0 2023-02-09 00:39:41,229 INFO [train.py:901] (0/4) Epoch 28, batch 4500, loss[loss=0.1849, simple_loss=0.2776, pruned_loss=0.04609, over 8197.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2824, pruned_loss=0.05756, over 1606037.32 frames. ], batch size: 23, lr: 2.69e-03, grad_scale: 16.0 2023-02-09 00:39:44,237 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222742.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:39:45,367 WARNING [train.py:1067] (0/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-02-09 00:39:58,788 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.64 vs. limit=2.0 2023-02-09 00:40:02,744 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222767.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:40:18,336 INFO [train.py:901] (0/4) Epoch 28, batch 4550, loss[loss=0.1499, simple_loss=0.2305, pruned_loss=0.03463, over 7217.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2823, pruned_loss=0.05733, over 1605768.31 frames. ], batch size: 16, lr: 2.69e-03, grad_scale: 16.0 2023-02-09 00:40:26,128 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.4168, 2.7160, 2.2108, 3.6696, 1.6491, 2.0618, 2.5615, 2.7453], device='cuda:0'), covar=tensor([0.0706, 0.0731, 0.0824, 0.0344, 0.1090, 0.1183, 0.0788, 0.0745], device='cuda:0'), in_proj_covar=tensor([0.0232, 0.0195, 0.0245, 0.0214, 0.0205, 0.0248, 0.0251, 0.0205], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], device='cuda:0') 2023-02-09 00:40:37,183 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.362e+02 2.324e+02 2.721e+02 3.677e+02 6.861e+02, threshold=5.442e+02, percent-clipped=4.0 2023-02-09 00:40:53,695 INFO [train.py:901] (0/4) Epoch 28, batch 4600, loss[loss=0.1726, simple_loss=0.2564, pruned_loss=0.04444, over 7807.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2809, pruned_loss=0.0568, over 1606321.03 frames. ], batch size: 20, lr: 2.69e-03, grad_scale: 16.0 2023-02-09 00:41:11,253 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.9197, 1.6382, 2.4832, 1.9894, 2.2561, 1.8889, 1.7476, 1.2160], device='cuda:0'), covar=tensor([0.7987, 0.6572, 0.2500, 0.4451, 0.3347, 0.5029, 0.2977, 0.5864], device='cuda:0'), in_proj_covar=tensor([0.0965, 0.1029, 0.0833, 0.0996, 0.1023, 0.0934, 0.0772, 0.0852], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 00:41:27,920 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=222885.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:41:29,998 INFO [train.py:901] (0/4) Epoch 28, batch 4650, loss[loss=0.169, simple_loss=0.2487, pruned_loss=0.04462, over 7801.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2815, pruned_loss=0.05737, over 1601179.40 frames. ], batch size: 20, lr: 2.68e-03, grad_scale: 16.0 2023-02-09 00:41:50,685 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.622e+02 2.423e+02 3.099e+02 3.500e+02 7.849e+02, threshold=6.198e+02, percent-clipped=6.0 2023-02-09 00:41:53,084 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222918.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:42:02,693 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=222932.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:42:06,739 INFO [train.py:901] (0/4) Epoch 28, batch 4700, loss[loss=0.2068, simple_loss=0.2764, pruned_loss=0.06866, over 7546.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2812, pruned_loss=0.05734, over 1605184.67 frames. ], batch size: 18, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:42:06,963 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222938.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:42:10,436 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222943.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:42:20,644 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222958.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:42:24,246 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222963.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:42:40,836 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222987.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:42:41,331 INFO [train.py:901] (0/4) Epoch 28, batch 4750, loss[loss=0.2459, simple_loss=0.3148, pruned_loss=0.08847, over 6866.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2814, pruned_loss=0.05755, over 1605715.76 frames. ], batch size: 71, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:42:50,012 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223000.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:42:51,379 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223002.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:42:54,709 WARNING [train.py:1067] (0/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-02-09 00:42:58,165 WARNING [train.py:1067] (0/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-02-09 00:43:02,878 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.833e+02 2.412e+02 2.807e+02 3.833e+02 7.869e+02, threshold=5.613e+02, percent-clipped=5.0 2023-02-09 00:43:18,664 INFO [train.py:901] (0/4) Epoch 28, batch 4800, loss[loss=0.1958, simple_loss=0.2858, pruned_loss=0.05288, over 8468.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2815, pruned_loss=0.05766, over 1607974.01 frames. ], batch size: 29, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:43:25,194 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223047.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:43:36,303 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.1303, 1.5493, 1.8562, 1.4022, 0.9916, 1.5782, 1.8549, 1.6843], device='cuda:0'), covar=tensor([0.0538, 0.1224, 0.1577, 0.1537, 0.0597, 0.1463, 0.0675, 0.0647], device='cuda:0'), in_proj_covar=tensor([0.0099, 0.0153, 0.0190, 0.0161, 0.0101, 0.0163, 0.0113, 0.0146], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0') 2023-02-09 00:43:48,814 WARNING [train.py:1067] (0/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-02-09 00:43:49,662 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223082.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:43:53,652 INFO [train.py:901] (0/4) Epoch 28, batch 4850, loss[loss=0.1629, simple_loss=0.2474, pruned_loss=0.03918, over 7922.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2817, pruned_loss=0.05746, over 1611843.73 frames. ], batch size: 20, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:44:13,747 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.627e+02 2.508e+02 3.332e+02 4.408e+02 9.671e+02, threshold=6.663e+02, percent-clipped=7.0 2023-02-09 00:44:31,161 INFO [train.py:901] (0/4) Epoch 28, batch 4900, loss[loss=0.1736, simple_loss=0.2483, pruned_loss=0.04942, over 7419.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2818, pruned_loss=0.0572, over 1608959.24 frames. ], batch size: 17, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:44:57,024 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-02-09 00:45:07,000 INFO [train.py:901] (0/4) Epoch 28, batch 4950, loss[loss=0.2179, simple_loss=0.3011, pruned_loss=0.06741, over 8027.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2823, pruned_loss=0.05702, over 1612808.58 frames. ], batch size: 22, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:45:09,246 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223191.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 00:45:26,410 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.723e+02 2.334e+02 2.712e+02 3.560e+02 9.309e+02, threshold=5.424e+02, percent-clipped=3.0 2023-02-09 00:45:42,287 INFO [train.py:901] (0/4) Epoch 28, batch 5000, loss[loss=0.2226, simple_loss=0.307, pruned_loss=0.0691, over 8132.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2809, pruned_loss=0.05638, over 1609447.49 frames. ], batch size: 22, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:45:56,107 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223256.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:46:14,348 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223281.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:46:18,927 INFO [train.py:901] (0/4) Epoch 28, batch 5050, loss[loss=0.1782, simple_loss=0.2579, pruned_loss=0.04919, over 7416.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2812, pruned_loss=0.05676, over 1611356.68 frames. ], batch size: 17, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:46:28,800 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=223302.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:46:29,640 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223303.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:46:32,879 WARNING [train.py:1067] (0/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-02-09 00:46:38,379 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.592e+02 2.274e+02 2.931e+02 3.573e+02 6.090e+02, threshold=5.862e+02, percent-clipped=1.0 2023-02-09 00:46:42,968 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.59 vs. limit=2.0 2023-02-09 00:46:46,945 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223328.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:46:48,957 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=223331.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:46:53,557 INFO [train.py:901] (0/4) Epoch 28, batch 5100, loss[loss=0.2065, simple_loss=0.2991, pruned_loss=0.05697, over 8249.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2807, pruned_loss=0.05635, over 1615566.31 frames. ], batch size: 24, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:46:59,472 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=223346.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:47:31,193 INFO [train.py:901] (0/4) Epoch 28, batch 5150, loss[loss=0.2021, simple_loss=0.2903, pruned_loss=0.05694, over 8449.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2814, pruned_loss=0.05675, over 1618656.53 frames. ], batch size: 27, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:47:50,499 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.755e+02 2.349e+02 2.964e+02 3.516e+02 1.122e+03, threshold=5.928e+02, percent-clipped=3.0 2023-02-09 00:47:51,405 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223417.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:47:57,694 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=223426.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:47:59,104 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([4.5509, 4.5448, 4.1312, 2.2812, 4.0106, 4.2136, 4.1434, 3.9839], device='cuda:0'), covar=tensor([0.0705, 0.0466, 0.0966, 0.4431, 0.0870, 0.0990, 0.1244, 0.0723], device='cuda:0'), in_proj_covar=tensor([0.0540, 0.0454, 0.0449, 0.0554, 0.0443, 0.0463, 0.0441, 0.0407], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 00:48:05,974 INFO [train.py:901] (0/4) Epoch 28, batch 5200, loss[loss=0.2003, simple_loss=0.2952, pruned_loss=0.05276, over 8248.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2823, pruned_loss=0.05766, over 1617835.73 frames. ], batch size: 22, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:48:11,634 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223446.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:48:22,132 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223461.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:48:27,856 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.6321, 2.4981, 3.1709, 2.6014, 3.1135, 2.7440, 2.6066, 2.0509], device='cuda:0'), covar=tensor([0.5867, 0.5269, 0.2179, 0.4308, 0.2904, 0.3153, 0.1881, 0.5961], device='cuda:0'), in_proj_covar=tensor([0.0965, 0.1031, 0.0836, 0.0999, 0.1026, 0.0936, 0.0773, 0.0855], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 00:48:31,235 WARNING [train.py:1067] (0/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-02-09 00:48:44,077 INFO [train.py:901] (0/4) Epoch 28, batch 5250, loss[loss=0.2291, simple_loss=0.315, pruned_loss=0.07161, over 8475.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2835, pruned_loss=0.05849, over 1619757.39 frames. ], batch size: 49, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:49:03,773 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.671e+02 2.237e+02 2.837e+02 3.561e+02 7.405e+02, threshold=5.674e+02, percent-clipped=6.0 2023-02-09 00:49:17,141 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=223535.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 00:49:19,101 INFO [train.py:901] (0/4) Epoch 28, batch 5300, loss[loss=0.21, simple_loss=0.294, pruned_loss=0.06301, over 8108.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2834, pruned_loss=0.05855, over 1620881.39 frames. ], batch size: 23, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:49:21,379 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223541.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:49:55,534 INFO [train.py:901] (0/4) Epoch 28, batch 5350, loss[loss=0.2048, simple_loss=0.2893, pruned_loss=0.06008, over 8100.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2835, pruned_loss=0.05876, over 1618661.53 frames. ], batch size: 23, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:50:01,806 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223596.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:50:15,668 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.554e+02 2.359e+02 2.840e+02 3.657e+02 7.209e+02, threshold=5.681e+02, percent-clipped=3.0 2023-02-09 00:50:30,881 INFO [train.py:901] (0/4) Epoch 28, batch 5400, loss[loss=0.2009, simple_loss=0.2788, pruned_loss=0.0615, over 7984.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2838, pruned_loss=0.05871, over 1620250.11 frames. ], batch size: 21, lr: 2.68e-03, grad_scale: 4.0 2023-02-09 00:50:39,719 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223650.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 00:50:55,924 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223673.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:51:06,378 INFO [train.py:901] (0/4) Epoch 28, batch 5450, loss[loss=0.1988, simple_loss=0.2845, pruned_loss=0.05658, over 8312.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2829, pruned_loss=0.05778, over 1621939.29 frames. ], batch size: 25, lr: 2.68e-03, grad_scale: 4.0 2023-02-09 00:51:13,820 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223698.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:51:16,779 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223702.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:51:28,424 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.718e+02 2.405e+02 2.886e+02 3.694e+02 6.837e+02, threshold=5.773e+02, percent-clipped=3.0 2023-02-09 00:51:28,696 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223717.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:51:31,421 WARNING [train.py:1067] (0/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-02-09 00:51:36,553 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223727.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:51:44,550 INFO [train.py:901] (0/4) Epoch 28, batch 5500, loss[loss=0.192, simple_loss=0.2596, pruned_loss=0.0622, over 7794.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2822, pruned_loss=0.05751, over 1622449.36 frames. ], batch size: 19, lr: 2.68e-03, grad_scale: 4.0 2023-02-09 00:51:47,335 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223742.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:52:18,976 INFO [train.py:901] (0/4) Epoch 28, batch 5550, loss[loss=0.1742, simple_loss=0.2529, pruned_loss=0.04779, over 7421.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.283, pruned_loss=0.05815, over 1612300.98 frames. ], batch size: 17, lr: 2.68e-03, grad_scale: 4.0 2023-02-09 00:52:25,548 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223797.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:52:31,280 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.5968, 1.6202, 1.8008, 1.5473, 0.9526, 1.5382, 1.9773, 1.9972], device='cuda:0'), covar=tensor([0.0510, 0.1215, 0.1688, 0.1523, 0.0639, 0.1498, 0.0706, 0.0613], device='cuda:0'), in_proj_covar=tensor([0.0100, 0.0154, 0.0191, 0.0162, 0.0102, 0.0164, 0.0113, 0.0147], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0') 2023-02-09 00:52:39,667 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.770e+02 2.462e+02 3.010e+02 3.574e+02 1.274e+03, threshold=6.020e+02, percent-clipped=3.0 2023-02-09 00:52:43,470 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223822.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:52:55,575 INFO [train.py:901] (0/4) Epoch 28, batch 5600, loss[loss=0.2162, simple_loss=0.2882, pruned_loss=0.07208, over 7528.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2815, pruned_loss=0.05739, over 1612808.11 frames. ], batch size: 18, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:52:57,099 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.6402, 1.9195, 2.8794, 1.5098, 2.2032, 2.0176, 1.7194, 2.1842], device='cuda:0'), covar=tensor([0.2089, 0.2655, 0.1017, 0.4833, 0.1924, 0.3365, 0.2444, 0.2260], device='cuda:0'), in_proj_covar=tensor([0.0540, 0.0636, 0.0565, 0.0670, 0.0662, 0.0611, 0.0563, 0.0645], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 00:53:31,702 INFO [train.py:901] (0/4) Epoch 28, batch 5650, loss[loss=0.1864, simple_loss=0.2776, pruned_loss=0.04762, over 8468.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2817, pruned_loss=0.05737, over 1613076.75 frames. ], batch size: 25, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:53:41,215 WARNING [train.py:1067] (0/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-02-09 00:53:44,200 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223906.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 00:53:51,307 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.465e+02 2.313e+02 2.789e+02 3.752e+02 1.102e+03, threshold=5.578e+02, percent-clipped=3.0 2023-02-09 00:53:52,429 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=2.79 vs. limit=5.0 2023-02-09 00:54:01,611 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223931.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 00:54:06,150 INFO [train.py:901] (0/4) Epoch 28, batch 5700, loss[loss=0.2215, simple_loss=0.3144, pruned_loss=0.06434, over 8500.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2831, pruned_loss=0.05786, over 1614888.93 frames. ], batch size: 26, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:54:07,515 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=223940.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:54:43,236 INFO [train.py:901] (0/4) Epoch 28, batch 5750, loss[loss=0.1958, simple_loss=0.2881, pruned_loss=0.05172, over 8526.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2824, pruned_loss=0.05717, over 1614870.42 frames. ], batch size: 28, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:54:48,705 WARNING [train.py:1067] (0/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-02-09 00:54:51,646 INFO [checkpoint.py:75] (0/4) Saving checkpoint to pruned_transducer_stateless7_streaming/exp/v1/checkpoint-224000.pt 2023-02-09 00:55:04,150 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.665e+02 2.263e+02 2.713e+02 3.241e+02 8.661e+02, threshold=5.425e+02, percent-clipped=3.0 2023-02-09 00:55:18,775 INFO [train.py:901] (0/4) Epoch 28, batch 5800, loss[loss=0.1758, simple_loss=0.2593, pruned_loss=0.04609, over 7804.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2827, pruned_loss=0.05738, over 1615326.94 frames. ], batch size: 19, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:55:30,600 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=224055.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:55:55,798 INFO [train.py:901] (0/4) Epoch 28, batch 5850, loss[loss=0.1995, simple_loss=0.2888, pruned_loss=0.05509, over 8455.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2822, pruned_loss=0.05697, over 1613611.79 frames. ], batch size: 27, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:56:15,668 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.578e+02 2.538e+02 3.148e+02 4.118e+02 7.183e+02, threshold=6.296e+02, percent-clipped=12.0 2023-02-09 00:56:30,207 INFO [train.py:901] (0/4) Epoch 28, batch 5900, loss[loss=0.2252, simple_loss=0.2879, pruned_loss=0.08129, over 7782.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2824, pruned_loss=0.0568, over 1617531.29 frames. ], batch size: 19, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:57:06,196 INFO [train.py:901] (0/4) Epoch 28, batch 5950, loss[loss=0.2349, simple_loss=0.3047, pruned_loss=0.08262, over 7030.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2832, pruned_loss=0.05717, over 1619420.47 frames. ], batch size: 71, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:57:16,703 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.63 vs. limit=2.0 2023-02-09 00:57:18,144 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 2023-02-09 00:57:19,333 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.5794, 1.4633, 1.8288, 1.2343, 1.2518, 1.8056, 0.2428, 1.1796], device='cuda:0'), covar=tensor([0.1459, 0.1122, 0.0382, 0.0771, 0.2259, 0.0436, 0.1763, 0.1147], device='cuda:0'), in_proj_covar=tensor([0.0202, 0.0207, 0.0138, 0.0224, 0.0279, 0.0148, 0.0175, 0.0200], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 00:57:28,304 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.818e+02 2.486e+02 3.110e+02 3.888e+02 7.674e+02, threshold=6.220e+02, percent-clipped=4.0 2023-02-09 00:57:28,889 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 2023-02-09 00:57:37,738 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=224230.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:57:43,112 INFO [train.py:901] (0/4) Epoch 28, batch 6000, loss[loss=0.2005, simple_loss=0.2864, pruned_loss=0.05724, over 8256.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2829, pruned_loss=0.05784, over 1614447.06 frames. ], batch size: 24, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:57:43,113 INFO [train.py:926] (0/4) Computing validation loss 2023-02-09 00:57:56,803 INFO [train.py:935] (0/4) Epoch 28, validation: loss=0.1714, simple_loss=0.2708, pruned_loss=0.03603, over 944034.00 frames. 2023-02-09 00:57:56,804 INFO [train.py:936] (0/4) Maximum memory allocated so far is 6461MB 2023-02-09 00:58:05,912 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=2.91 vs. limit=5.0 2023-02-09 00:58:33,320 INFO [train.py:901] (0/4) Epoch 28, batch 6050, loss[loss=0.1902, simple_loss=0.2655, pruned_loss=0.05746, over 7811.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2831, pruned_loss=0.05803, over 1613756.31 frames. ], batch size: 20, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:58:38,452 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.0678, 1.4641, 1.7030, 1.3440, 0.9198, 1.4432, 1.8238, 1.7271], device='cuda:0'), covar=tensor([0.0578, 0.1272, 0.1731, 0.1556, 0.0616, 0.1530, 0.0736, 0.0607], device='cuda:0'), in_proj_covar=tensor([0.0099, 0.0153, 0.0190, 0.0161, 0.0101, 0.0163, 0.0113, 0.0146], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0') 2023-02-09 00:58:49,962 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=224311.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:58:54,048 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.623e+02 2.462e+02 3.109e+02 3.867e+02 1.260e+03, threshold=6.217e+02, percent-clipped=5.0 2023-02-09 00:59:08,880 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=224336.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 00:59:10,076 INFO [train.py:901] (0/4) Epoch 28, batch 6100, loss[loss=0.2167, simple_loss=0.2988, pruned_loss=0.06733, over 8635.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2843, pruned_loss=0.05899, over 1614512.72 frames. ], batch size: 34, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:59:26,306 WARNING [train.py:1067] (0/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-02-09 00:59:46,740 INFO [train.py:901] (0/4) Epoch 28, batch 6150, loss[loss=0.1938, simple_loss=0.2919, pruned_loss=0.04787, over 8302.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2833, pruned_loss=0.058, over 1613450.07 frames. ], batch size: 25, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 00:59:48,616 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=3.20 vs. limit=5.0 2023-02-09 00:59:54,613 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.6321, 1.9381, 2.0185, 1.4154, 2.1901, 1.4671, 0.7790, 1.9347], device='cuda:0'), covar=tensor([0.0795, 0.0460, 0.0347, 0.0739, 0.0530, 0.1108, 0.1014, 0.0451], device='cuda:0'), in_proj_covar=tensor([0.0471, 0.0409, 0.0363, 0.0458, 0.0393, 0.0550, 0.0402, 0.0440], device='cuda:0'), out_proj_covar=tensor([1.2457e-04, 1.0580e-04, 9.4517e-05, 1.1960e-04, 1.0283e-04, 1.5344e-04, 1.0723e-04, 1.1521e-04], device='cuda:0') 2023-02-09 01:00:06,959 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.535e+02 2.362e+02 2.823e+02 3.455e+02 8.158e+02, threshold=5.645e+02, percent-clipped=2.0 2023-02-09 01:00:21,356 INFO [train.py:901] (0/4) Epoch 28, batch 6200, loss[loss=0.163, simple_loss=0.2362, pruned_loss=0.04487, over 7554.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2825, pruned_loss=0.05763, over 1609299.28 frames. ], batch size: 18, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 01:00:58,230 INFO [train.py:901] (0/4) Epoch 28, batch 6250, loss[loss=0.2008, simple_loss=0.268, pruned_loss=0.06681, over 7690.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2816, pruned_loss=0.05709, over 1609660.09 frames. ], batch size: 18, lr: 2.68e-03, grad_scale: 8.0 2023-02-09 01:00:59,404 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=3.64 vs. limit=5.0 2023-02-09 01:01:18,582 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.516e+02 2.593e+02 3.043e+02 4.250e+02 9.084e+02, threshold=6.087e+02, percent-clipped=11.0 2023-02-09 01:01:33,266 INFO [train.py:901] (0/4) Epoch 28, batch 6300, loss[loss=0.2024, simple_loss=0.2855, pruned_loss=0.05963, over 8238.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2819, pruned_loss=0.05735, over 1606921.67 frames. ], batch size: 22, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:01:50,186 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.1874, 1.9654, 2.3491, 2.0481, 2.2690, 2.2513, 2.1395, 1.2472], device='cuda:0'), covar=tensor([0.5665, 0.4698, 0.2163, 0.3920, 0.2546, 0.3311, 0.1964, 0.5177], device='cuda:0'), in_proj_covar=tensor([0.0971, 0.1034, 0.0841, 0.1006, 0.1029, 0.0940, 0.0776, 0.0857], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 01:02:00,131 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=224574.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:02:04,365 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=224580.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:02:04,427 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.0468, 1.5952, 1.4030, 1.5269, 1.3399, 1.2642, 1.2585, 1.2987], device='cuda:0'), covar=tensor([0.1218, 0.0575, 0.1389, 0.0656, 0.0807, 0.1646, 0.0996, 0.0910], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0245, 0.0342, 0.0315, 0.0303, 0.0348, 0.0352, 0.0325], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 01:02:10,454 INFO [train.py:901] (0/4) Epoch 28, batch 6350, loss[loss=0.2095, simple_loss=0.3048, pruned_loss=0.05715, over 8357.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2832, pruned_loss=0.05817, over 1611737.79 frames. ], batch size: 26, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:02:30,929 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.559e+02 2.315e+02 2.720e+02 3.259e+02 6.733e+02, threshold=5.440e+02, percent-clipped=2.0 2023-02-09 01:02:45,886 INFO [train.py:901] (0/4) Epoch 28, batch 6400, loss[loss=0.2031, simple_loss=0.2815, pruned_loss=0.06234, over 7802.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2828, pruned_loss=0.0576, over 1613843.66 frames. ], batch size: 19, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:03:21,511 INFO [train.py:901] (0/4) Epoch 28, batch 6450, loss[loss=0.2208, simple_loss=0.3042, pruned_loss=0.06868, over 8545.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.282, pruned_loss=0.05737, over 1614994.14 frames. ], batch size: 31, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:03:22,404 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=224689.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:03:39,013 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.2485, 2.0321, 2.5615, 2.1747, 2.5083, 2.3316, 2.1828, 1.3518], device='cuda:0'), covar=tensor([0.5633, 0.4948, 0.2015, 0.3914, 0.2493, 0.3392, 0.1991, 0.5555], device='cuda:0'), in_proj_covar=tensor([0.0966, 0.1030, 0.0837, 0.1000, 0.1025, 0.0935, 0.0773, 0.0852], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 01:03:43,003 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.578e+02 2.303e+02 2.784e+02 3.485e+02 7.082e+02, threshold=5.567e+02, percent-clipped=7.0 2023-02-09 01:03:57,598 INFO [train.py:901] (0/4) Epoch 28, batch 6500, loss[loss=0.2041, simple_loss=0.2934, pruned_loss=0.05739, over 8467.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2823, pruned_loss=0.0574, over 1616698.59 frames. ], batch size: 25, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:04:02,118 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=224744.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:04:11,957 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=224758.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:04:32,262 INFO [train.py:901] (0/4) Epoch 28, batch 6550, loss[loss=0.2297, simple_loss=0.3035, pruned_loss=0.07791, over 8304.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.284, pruned_loss=0.05857, over 1616443.94 frames. ], batch size: 25, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:04:47,796 WARNING [train.py:1067] (0/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-02-09 01:04:53,997 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.746e+02 2.489e+02 3.184e+02 3.768e+02 7.222e+02, threshold=6.368e+02, percent-clipped=1.0 2023-02-09 01:05:08,028 WARNING [train.py:1067] (0/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-02-09 01:05:09,352 INFO [train.py:901] (0/4) Epoch 28, batch 6600, loss[loss=0.2287, simple_loss=0.3122, pruned_loss=0.07264, over 8353.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2844, pruned_loss=0.05894, over 1618167.46 frames. ], batch size: 26, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:05:26,402 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=2.74 vs. limit=5.0 2023-02-09 01:05:44,398 INFO [train.py:901] (0/4) Epoch 28, batch 6650, loss[loss=0.202, simple_loss=0.2892, pruned_loss=0.0574, over 8468.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2823, pruned_loss=0.05769, over 1611817.62 frames. ], batch size: 25, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:05:50,128 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.8817, 2.6525, 2.0483, 2.4025, 2.3412, 1.8501, 2.2144, 2.4096], device='cuda:0'), covar=tensor([0.1430, 0.0416, 0.1179, 0.0660, 0.0723, 0.1412, 0.1021, 0.1028], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0247, 0.0346, 0.0317, 0.0304, 0.0350, 0.0354, 0.0327], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 01:06:04,780 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.705e+02 2.463e+02 2.971e+02 3.895e+02 9.422e+02, threshold=5.941e+02, percent-clipped=4.0 2023-02-09 01:06:10,943 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=224924.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:06:21,167 INFO [train.py:901] (0/4) Epoch 28, batch 6700, loss[loss=0.2165, simple_loss=0.3105, pruned_loss=0.06124, over 8524.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2828, pruned_loss=0.05776, over 1607211.10 frames. ], batch size: 28, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:06:22,611 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([4.1769, 4.1845, 3.7510, 2.0396, 3.6818, 3.8016, 3.7203, 3.6368], device='cuda:0'), covar=tensor([0.0756, 0.0539, 0.1030, 0.4327, 0.0962, 0.1039, 0.1274, 0.0874], device='cuda:0'), in_proj_covar=tensor([0.0550, 0.0462, 0.0456, 0.0566, 0.0448, 0.0471, 0.0448, 0.0415], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 01:06:26,245 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=224945.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:06:30,454 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.8926, 1.6114, 2.8635, 1.4053, 2.3342, 3.0604, 3.2527, 2.6365], device='cuda:0'), covar=tensor([0.1051, 0.1575, 0.0377, 0.2091, 0.0877, 0.0290, 0.0568, 0.0548], device='cuda:0'), in_proj_covar=tensor([0.0307, 0.0326, 0.0294, 0.0322, 0.0324, 0.0276, 0.0441, 0.0306], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 01:06:44,453 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=224970.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:06:57,025 INFO [train.py:901] (0/4) Epoch 28, batch 6750, loss[loss=0.1386, simple_loss=0.2222, pruned_loss=0.02747, over 7436.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2817, pruned_loss=0.05642, over 1611217.80 frames. ], batch size: 17, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:06:57,472 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.60 vs. limit=2.0 2023-02-09 01:06:58,667 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.7239, 2.5333, 1.9188, 2.2940, 2.2105, 1.6946, 2.1574, 2.2097], device='cuda:0'), covar=tensor([0.1551, 0.0467, 0.1176, 0.0721, 0.0780, 0.1602, 0.1089, 0.1144], device='cuda:0'), in_proj_covar=tensor([0.0362, 0.0247, 0.0345, 0.0317, 0.0304, 0.0350, 0.0354, 0.0327], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 01:07:01,390 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.4382, 2.6919, 3.0253, 1.7242, 3.3417, 2.1555, 1.5604, 2.4029], device='cuda:0'), covar=tensor([0.0941, 0.0418, 0.0400, 0.0994, 0.0551, 0.0874, 0.1172, 0.0613], device='cuda:0'), in_proj_covar=tensor([0.0473, 0.0409, 0.0364, 0.0459, 0.0396, 0.0552, 0.0403, 0.0442], device='cuda:0'), out_proj_covar=tensor([1.2520e-04, 1.0596e-04, 9.4811e-05, 1.2003e-04, 1.0343e-04, 1.5400e-04, 1.0766e-04, 1.1576e-04], device='cuda:0') 2023-02-09 01:07:17,011 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.474e+02 2.343e+02 2.979e+02 3.883e+02 6.136e+02, threshold=5.958e+02, percent-clipped=2.0 2023-02-09 01:07:28,014 WARNING [train.py:1067] (0/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-02-09 01:07:32,087 INFO [train.py:901] (0/4) Epoch 28, batch 6800, loss[loss=0.2307, simple_loss=0.3102, pruned_loss=0.07565, over 8366.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2824, pruned_loss=0.05687, over 1613586.86 frames. ], batch size: 24, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:07:32,960 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225039.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:07:57,284 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.6084, 1.6745, 1.9955, 1.6981, 1.0419, 1.7599, 2.1559, 1.9747], device='cuda:0'), covar=tensor([0.0513, 0.1235, 0.1581, 0.1427, 0.0627, 0.1406, 0.0685, 0.0620], device='cuda:0'), in_proj_covar=tensor([0.0099, 0.0153, 0.0189, 0.0161, 0.0102, 0.0163, 0.0113, 0.0147], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0') 2023-02-09 01:08:08,384 INFO [train.py:901] (0/4) Epoch 28, batch 6850, loss[loss=0.1674, simple_loss=0.2613, pruned_loss=0.03672, over 8030.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2827, pruned_loss=0.05733, over 1607846.70 frames. ], batch size: 22, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:08:08,456 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=225088.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:08:18,039 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=225102.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:08:18,712 WARNING [train.py:1067] (0/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-02-09 01:08:28,499 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.612e+02 2.283e+02 2.996e+02 3.907e+02 8.918e+02, threshold=5.992e+02, percent-clipped=3.0 2023-02-09 01:08:31,362 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225121.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:08:42,882 INFO [train.py:901] (0/4) Epoch 28, batch 6900, loss[loss=0.1742, simple_loss=0.2595, pruned_loss=0.04441, over 7922.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2813, pruned_loss=0.05684, over 1606909.67 frames. ], batch size: 20, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:08:47,495 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0 2023-02-09 01:08:58,650 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225160.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:09:18,985 INFO [train.py:901] (0/4) Epoch 28, batch 6950, loss[loss=0.2179, simple_loss=0.2991, pruned_loss=0.06833, over 8582.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.281, pruned_loss=0.0573, over 1606123.48 frames. ], batch size: 39, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:09:30,334 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225203.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:09:30,870 WARNING [train.py:1067] (0/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-02-09 01:09:40,019 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.583e+02 2.456e+02 2.946e+02 3.977e+02 8.721e+02, threshold=5.892e+02, percent-clipped=6.0 2023-02-09 01:09:40,242 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225217.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:09:54,782 INFO [train.py:901] (0/4) Epoch 28, batch 7000, loss[loss=0.2124, simple_loss=0.2877, pruned_loss=0.06852, over 7641.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2811, pruned_loss=0.05744, over 1609044.07 frames. ], batch size: 19, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:10:31,206 INFO [train.py:901] (0/4) Epoch 28, batch 7050, loss[loss=0.2148, simple_loss=0.2995, pruned_loss=0.06504, over 8468.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2807, pruned_loss=0.05738, over 1610005.39 frames. ], batch size: 25, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:10:36,318 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225295.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:10:41,256 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([4.6841, 4.6875, 4.1896, 2.3639, 4.0843, 4.3287, 4.2530, 4.1532], device='cuda:0'), covar=tensor([0.0643, 0.0440, 0.0932, 0.4201, 0.0846, 0.0927, 0.1060, 0.0758], device='cuda:0'), in_proj_covar=tensor([0.0549, 0.0463, 0.0457, 0.0564, 0.0448, 0.0470, 0.0447, 0.0415], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 01:10:52,627 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.739e+02 2.402e+02 2.844e+02 3.449e+02 6.425e+02, threshold=5.688e+02, percent-clipped=2.0 2023-02-09 01:10:55,656 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225320.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:11:08,211 INFO [train.py:901] (0/4) Epoch 28, batch 7100, loss[loss=0.223, simple_loss=0.3143, pruned_loss=0.06583, over 8709.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.281, pruned_loss=0.0571, over 1613871.16 frames. ], batch size: 34, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:11:43,051 INFO [train.py:901] (0/4) Epoch 28, batch 7150, loss[loss=0.2217, simple_loss=0.3075, pruned_loss=0.06792, over 8342.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.281, pruned_loss=0.05712, over 1611726.18 frames. ], batch size: 26, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:12:05,413 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.501e+02 2.377e+02 2.906e+02 3.542e+02 6.036e+02, threshold=5.811e+02, percent-clipped=2.0 2023-02-09 01:12:21,605 INFO [train.py:901] (0/4) Epoch 28, batch 7200, loss[loss=0.1367, simple_loss=0.2272, pruned_loss=0.02312, over 8078.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2811, pruned_loss=0.05692, over 1614685.87 frames. ], batch size: 21, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:12:35,170 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225457.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:12:36,600 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225459.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:12:40,500 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=225465.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:12:44,932 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.5998, 2.0467, 3.1856, 1.5041, 2.4172, 2.0034, 1.7246, 2.4690], device='cuda:0'), covar=tensor([0.1981, 0.2728, 0.1087, 0.4797, 0.1952, 0.3429, 0.2542, 0.2353], device='cuda:0'), in_proj_covar=tensor([0.0538, 0.0638, 0.0563, 0.0673, 0.0663, 0.0614, 0.0563, 0.0647], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 01:12:46,198 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225473.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:12:51,117 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225480.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:12:53,940 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225484.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:12:56,555 INFO [train.py:901] (0/4) Epoch 28, batch 7250, loss[loss=0.1823, simple_loss=0.2499, pruned_loss=0.05732, over 7702.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2826, pruned_loss=0.05772, over 1616793.77 frames. ], batch size: 18, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:13:03,605 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225498.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:13:07,601 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=225504.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:13:16,363 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.440e+02 2.486e+02 3.022e+02 3.617e+02 8.325e+02, threshold=6.044e+02, percent-clipped=6.0 2023-02-09 01:13:21,951 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225523.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:13:32,772 INFO [train.py:901] (0/4) Epoch 28, batch 7300, loss[loss=0.1806, simple_loss=0.2635, pruned_loss=0.04879, over 7710.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2834, pruned_loss=0.0578, over 1620594.37 frames. ], batch size: 18, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:13:34,183 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225540.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:13:56,395 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-02-09 01:14:00,398 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225577.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 01:14:02,410 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225580.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:14:07,683 INFO [train.py:901] (0/4) Epoch 28, batch 7350, loss[loss=0.219, simple_loss=0.2936, pruned_loss=0.07218, over 8593.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2832, pruned_loss=0.05823, over 1621177.20 frames. ], batch size: 34, lr: 2.67e-03, grad_scale: 8.0 2023-02-09 01:14:24,556 WARNING [train.py:1067] (0/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-02-09 01:14:27,898 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.502e+02 2.380e+02 2.753e+02 3.463e+02 7.224e+02, threshold=5.506e+02, percent-clipped=3.0 2023-02-09 01:14:29,532 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225619.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:14:42,951 INFO [train.py:901] (0/4) Epoch 28, batch 7400, loss[loss=0.1818, simple_loss=0.2706, pruned_loss=0.04651, over 7800.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2832, pruned_loss=0.05846, over 1621525.63 frames. ], batch size: 20, lr: 2.67e-03, grad_scale: 16.0 2023-02-09 01:14:42,964 WARNING [train.py:1067] (0/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-02-09 01:15:18,728 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([3.2776, 1.3571, 3.3919, 1.1195, 3.0127, 2.8042, 3.0669, 2.9808], device='cuda:0'), covar=tensor([0.0832, 0.4317, 0.0821, 0.4472, 0.1326, 0.1144, 0.0799, 0.0952], device='cuda:0'), in_proj_covar=tensor([0.0683, 0.0664, 0.0735, 0.0660, 0.0747, 0.0636, 0.0645, 0.0717], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 01:15:19,317 INFO [train.py:901] (0/4) Epoch 28, batch 7450, loss[loss=0.2015, simple_loss=0.2876, pruned_loss=0.05774, over 8107.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2831, pruned_loss=0.05867, over 1619114.97 frames. ], batch size: 23, lr: 2.67e-03, grad_scale: 16.0 2023-02-09 01:15:23,699 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.7783, 1.8613, 2.1604, 1.8481, 0.9873, 1.9939, 2.2077, 2.4095], device='cuda:0'), covar=tensor([0.0467, 0.1139, 0.1533, 0.1308, 0.0581, 0.1288, 0.0633, 0.0517], device='cuda:0'), in_proj_covar=tensor([0.0099, 0.0153, 0.0189, 0.0160, 0.0101, 0.0162, 0.0113, 0.0146], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0') 2023-02-09 01:15:25,013 WARNING [train.py:1067] (0/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-02-09 01:15:40,024 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.685e+02 2.396e+02 3.007e+02 3.866e+02 7.466e+02, threshold=6.014e+02, percent-clipped=6.0 2023-02-09 01:15:54,483 INFO [train.py:901] (0/4) Epoch 28, batch 7500, loss[loss=0.2199, simple_loss=0.298, pruned_loss=0.07088, over 8330.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.283, pruned_loss=0.05813, over 1620068.30 frames. ], batch size: 26, lr: 2.67e-03, grad_scale: 16.0 2023-02-09 01:16:32,508 INFO [train.py:901] (0/4) Epoch 28, batch 7550, loss[loss=0.1925, simple_loss=0.2977, pruned_loss=0.04362, over 8301.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2839, pruned_loss=0.0587, over 1617038.25 frames. ], batch size: 23, lr: 2.67e-03, grad_scale: 16.0 2023-02-09 01:16:39,082 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.4354, 2.3442, 1.7879, 2.0663, 2.0678, 1.5216, 1.9837, 2.0210], device='cuda:0'), covar=tensor([0.1604, 0.0453, 0.1262, 0.0716, 0.0741, 0.1659, 0.1016, 0.1010], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0244, 0.0341, 0.0312, 0.0300, 0.0345, 0.0348, 0.0320], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 01:16:41,875 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=225801.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:16:52,725 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.729e+02 2.388e+02 3.127e+02 4.485e+02 1.321e+03, threshold=6.254e+02, percent-clipped=11.0 2023-02-09 01:16:57,627 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=225824.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:17:05,959 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225836.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:17:07,145 INFO [train.py:901] (0/4) Epoch 28, batch 7600, loss[loss=0.1683, simple_loss=0.2402, pruned_loss=0.04813, over 7802.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2835, pruned_loss=0.05834, over 1620414.21 frames. ], batch size: 19, lr: 2.67e-03, grad_scale: 16.0 2023-02-09 01:17:23,282 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225861.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:17:27,275 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=225867.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:17:34,161 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225875.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:17:37,629 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225880.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:17:40,886 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=225884.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:17:43,480 INFO [train.py:901] (0/4) Epoch 28, batch 7650, loss[loss=0.1984, simple_loss=0.2751, pruned_loss=0.06083, over 7811.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2827, pruned_loss=0.05836, over 1619007.11 frames. ], batch size: 20, lr: 2.67e-03, grad_scale: 16.0 2023-02-09 01:17:51,611 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225900.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:18:01,187 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225913.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:18:03,351 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225916.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:18:03,835 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.518e+02 2.353e+02 2.789e+02 3.444e+02 7.654e+02, threshold=5.579e+02, percent-clipped=1.0 2023-02-09 01:18:06,735 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=225921.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 01:18:18,649 INFO [train.py:901] (0/4) Epoch 28, batch 7700, loss[loss=0.1995, simple_loss=0.2877, pruned_loss=0.05567, over 7277.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.282, pruned_loss=0.05754, over 1618470.04 frames. ], batch size: 16, lr: 2.67e-03, grad_scale: 16.0 2023-02-09 01:18:19,539 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225939.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:18:31,345 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 2023-02-09 01:18:44,258 WARNING [train.py:1067] (0/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-02-09 01:18:44,513 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.4937, 2.4042, 3.0701, 2.5460, 3.0578, 2.5027, 2.4929, 2.0680], device='cuda:0'), covar=tensor([0.5765, 0.4993, 0.2206, 0.4159, 0.2672, 0.3410, 0.1828, 0.5536], device='cuda:0'), in_proj_covar=tensor([0.0964, 0.1031, 0.0836, 0.0999, 0.1025, 0.0936, 0.0772, 0.0855], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 01:18:49,254 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225982.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:18:53,148 INFO [train.py:901] (0/4) Epoch 28, batch 7750, loss[loss=0.2002, simple_loss=0.281, pruned_loss=0.05967, over 8685.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2818, pruned_loss=0.05718, over 1619903.93 frames. ], batch size: 49, lr: 2.67e-03, grad_scale: 16.0 2023-02-09 01:19:01,902 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225999.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:19:02,550 INFO [checkpoint.py:75] (0/4) Saving checkpoint to pruned_transducer_stateless7_streaming/exp/v1/checkpoint-226000.pt 2023-02-09 01:19:15,708 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.697e+02 2.439e+02 2.815e+02 3.514e+02 7.333e+02, threshold=5.630e+02, percent-clipped=1.0 2023-02-09 01:19:29,663 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226036.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 01:19:30,729 INFO [train.py:901] (0/4) Epoch 28, batch 7800, loss[loss=0.2396, simple_loss=0.3157, pruned_loss=0.0818, over 8125.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2812, pruned_loss=0.05694, over 1615171.85 frames. ], batch size: 22, lr: 2.67e-03, grad_scale: 16.0 2023-02-09 01:19:45,505 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226059.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:20:05,563 INFO [train.py:901] (0/4) Epoch 28, batch 7850, loss[loss=0.1676, simple_loss=0.2582, pruned_loss=0.03855, over 7963.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2795, pruned_loss=0.05646, over 1607744.62 frames. ], batch size: 21, lr: 2.67e-03, grad_scale: 16.0 2023-02-09 01:20:25,276 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.578e+02 2.405e+02 3.030e+02 3.828e+02 1.208e+03, threshold=6.060e+02, percent-clipped=4.0 2023-02-09 01:20:39,662 INFO [train.py:901] (0/4) Epoch 28, batch 7900, loss[loss=0.1904, simple_loss=0.2816, pruned_loss=0.04958, over 8473.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2799, pruned_loss=0.0563, over 1611916.71 frames. ], batch size: 25, lr: 2.67e-03, grad_scale: 16.0 2023-02-09 01:21:02,890 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226172.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:21:13,382 INFO [train.py:901] (0/4) Epoch 28, batch 7950, loss[loss=0.2032, simple_loss=0.2693, pruned_loss=0.06852, over 7650.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2813, pruned_loss=0.05682, over 1615593.82 frames. ], batch size: 19, lr: 2.67e-03, grad_scale: 16.0 2023-02-09 01:21:18,316 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226195.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:21:19,697 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226197.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:21:33,033 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.821e+02 2.474e+02 2.920e+02 3.612e+02 7.690e+02, threshold=5.839e+02, percent-clipped=4.0 2023-02-09 01:21:35,323 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226220.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:21:36,616 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.6945, 1.5298, 1.9924, 1.5056, 1.1324, 1.6367, 2.2157, 1.9904], device='cuda:0'), covar=tensor([0.0484, 0.1264, 0.1595, 0.1471, 0.0616, 0.1406, 0.0622, 0.0625], device='cuda:0'), in_proj_covar=tensor([0.0099, 0.0153, 0.0189, 0.0161, 0.0102, 0.0163, 0.0113, 0.0146], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0') 2023-02-09 01:21:37,935 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=226224.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:21:47,691 INFO [train.py:901] (0/4) Epoch 28, batch 8000, loss[loss=0.1667, simple_loss=0.2482, pruned_loss=0.04255, over 8098.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2819, pruned_loss=0.05696, over 1618603.81 frames. ], batch size: 21, lr: 2.66e-03, grad_scale: 16.0 2023-02-09 01:21:47,907 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226238.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:21:52,819 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.2729, 1.0326, 2.3812, 0.9525, 2.1051, 2.0315, 2.1874, 2.1365], device='cuda:0'), covar=tensor([0.0831, 0.3163, 0.1015, 0.3644, 0.1071, 0.0969, 0.0683, 0.0774], device='cuda:0'), in_proj_covar=tensor([0.0688, 0.0668, 0.0738, 0.0664, 0.0753, 0.0641, 0.0648, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 01:21:57,001 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226251.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:21:59,706 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226255.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:22:00,948 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=226257.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:22:05,129 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226263.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:22:17,266 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226280.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:22:22,557 INFO [train.py:901] (0/4) Epoch 28, batch 8050, loss[loss=0.1689, simple_loss=0.2442, pruned_loss=0.04685, over 7797.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2806, pruned_loss=0.05756, over 1596546.09 frames. ], batch size: 19, lr: 2.66e-03, grad_scale: 16.0 2023-02-09 01:22:25,375 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226292.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 01:22:30,122 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([4.3759, 1.5350, 4.5984, 1.7259, 4.0937, 3.8397, 4.1943, 4.0757], device='cuda:0'), covar=tensor([0.0630, 0.4538, 0.0540, 0.4313, 0.1092, 0.0967, 0.0546, 0.0679], device='cuda:0'), in_proj_covar=tensor([0.0685, 0.0665, 0.0735, 0.0661, 0.0749, 0.0638, 0.0644, 0.0718], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 01:22:37,474 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-02-09 01:22:42,674 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.612e+02 2.434e+02 3.095e+02 3.696e+02 6.520e+02, threshold=6.190e+02, percent-clipped=3.0 2023-02-09 01:22:42,894 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226317.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 01:22:45,974 INFO [checkpoint.py:75] (0/4) Saving checkpoint to pruned_transducer_stateless7_streaming/exp/v1/epoch-28.pt 2023-02-09 01:22:57,715 WARNING [train.py:1067] (0/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-02-09 01:23:01,693 INFO [train.py:901] (0/4) Epoch 29, batch 0, loss[loss=0.2242, simple_loss=0.3075, pruned_loss=0.07048, over 8626.00 frames. ], tot_loss[loss=0.2242, simple_loss=0.3075, pruned_loss=0.07048, over 8626.00 frames. ], batch size: 34, lr: 2.62e-03, grad_scale: 16.0 2023-02-09 01:23:01,694 INFO [train.py:926] (0/4) Computing validation loss 2023-02-09 01:23:13,268 INFO [train.py:935] (0/4) Epoch 29, validation: loss=0.1705, simple_loss=0.2705, pruned_loss=0.03528, over 944034.00 frames. 2023-02-09 01:23:13,269 INFO [train.py:936] (0/4) Maximum memory allocated so far is 6461MB 2023-02-09 01:23:19,034 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.9453, 1.9733, 2.3098, 2.0591, 1.1589, 1.9153, 2.4261, 2.6898], device='cuda:0'), covar=tensor([0.0432, 0.1033, 0.1455, 0.1195, 0.0547, 0.1273, 0.0547, 0.0482], device='cuda:0'), in_proj_covar=tensor([0.0099, 0.0153, 0.0189, 0.0161, 0.0102, 0.0163, 0.0113, 0.0146], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0') 2023-02-09 01:23:26,241 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226339.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:23:26,283 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.7866, 1.7119, 2.2612, 1.4629, 1.4218, 2.2626, 0.4294, 1.4404], device='cuda:0'), covar=tensor([0.1493, 0.1191, 0.0377, 0.1049, 0.2274, 0.0386, 0.1855, 0.1213], device='cuda:0'), in_proj_covar=tensor([0.0202, 0.0206, 0.0138, 0.0224, 0.0278, 0.0147, 0.0173, 0.0200], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 01:23:29,665 WARNING [train.py:1067] (0/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-02-09 01:23:49,913 INFO [train.py:901] (0/4) Epoch 29, batch 50, loss[loss=0.1936, simple_loss=0.275, pruned_loss=0.05607, over 7805.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2798, pruned_loss=0.05705, over 366295.24 frames. ], batch size: 20, lr: 2.62e-03, grad_scale: 16.0 2023-02-09 01:23:50,840 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226372.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:24:06,035 WARNING [train.py:1067] (0/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-02-09 01:24:12,507 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=226403.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:24:22,930 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.399e+02 2.293e+02 2.936e+02 3.721e+02 6.222e+02, threshold=5.872e+02, percent-clipped=1.0 2023-02-09 01:24:25,771 INFO [train.py:901] (0/4) Epoch 29, batch 100, loss[loss=0.1716, simple_loss=0.2597, pruned_loss=0.04176, over 8194.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2821, pruned_loss=0.05763, over 644294.06 frames. ], batch size: 23, lr: 2.62e-03, grad_scale: 16.0 2023-02-09 01:24:30,610 WARNING [train.py:1067] (0/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-02-09 01:25:02,686 INFO [train.py:901] (0/4) Epoch 29, batch 150, loss[loss=0.1783, simple_loss=0.2628, pruned_loss=0.04695, over 7978.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2823, pruned_loss=0.05811, over 859065.55 frames. ], batch size: 21, lr: 2.62e-03, grad_scale: 16.0 2023-02-09 01:25:34,581 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.592e+02 2.438e+02 2.916e+02 4.111e+02 7.524e+02, threshold=5.832e+02, percent-clipped=2.0 2023-02-09 01:25:35,520 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226518.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:25:37,497 INFO [train.py:901] (0/4) Epoch 29, batch 200, loss[loss=0.1851, simple_loss=0.2753, pruned_loss=0.04747, over 8456.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2836, pruned_loss=0.05867, over 1026619.96 frames. ], batch size: 27, lr: 2.62e-03, grad_scale: 16.0 2023-02-09 01:25:51,888 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.6616, 2.6352, 1.9779, 2.3662, 2.2181, 1.7185, 2.1191, 2.2531], device='cuda:0'), covar=tensor([0.1536, 0.0451, 0.1310, 0.0673, 0.0884, 0.1637, 0.1112, 0.1050], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0245, 0.0342, 0.0313, 0.0301, 0.0348, 0.0349, 0.0322], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 01:26:13,674 INFO [train.py:901] (0/4) Epoch 29, batch 250, loss[loss=0.2013, simple_loss=0.2844, pruned_loss=0.05907, over 8466.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2831, pruned_loss=0.0583, over 1155270.58 frames. ], batch size: 29, lr: 2.62e-03, grad_scale: 16.0 2023-02-09 01:26:26,050 WARNING [train.py:1067] (0/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-02-09 01:26:30,591 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.7062, 2.1310, 3.1264, 1.5714, 2.3866, 2.1799, 1.7875, 2.4580], device='cuda:0'), covar=tensor([0.1950, 0.2700, 0.0972, 0.4613, 0.1928, 0.3280, 0.2506, 0.2210], device='cuda:0'), in_proj_covar=tensor([0.0542, 0.0640, 0.0564, 0.0674, 0.0666, 0.0616, 0.0566, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:0') 2023-02-09 01:26:31,144 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=226595.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:26:31,306 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226595.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:26:33,831 WARNING [train.py:1067] (0/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-02-09 01:26:42,336 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.6777, 2.5754, 1.8712, 2.3065, 2.1963, 1.5843, 2.1701, 2.2334], device='cuda:0'), covar=tensor([0.1525, 0.0418, 0.1212, 0.0682, 0.0828, 0.1666, 0.1007, 0.1028], device='cuda:0'), in_proj_covar=tensor([0.0355, 0.0244, 0.0341, 0.0312, 0.0300, 0.0347, 0.0348, 0.0320], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 01:26:44,587 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.2205, 2.0075, 2.4509, 2.0891, 2.4209, 2.2949, 2.1250, 1.2896], device='cuda:0'), covar=tensor([0.5572, 0.5012, 0.2225, 0.3966, 0.2524, 0.3276, 0.1943, 0.5578], device='cuda:0'), in_proj_covar=tensor([0.0966, 0.1031, 0.0838, 0.0999, 0.1026, 0.0936, 0.0773, 0.0852], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 01:26:46,394 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.753e+02 2.514e+02 3.003e+02 3.645e+02 8.891e+02, threshold=6.006e+02, percent-clipped=9.0 2023-02-09 01:26:48,787 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226620.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:26:49,258 INFO [train.py:901] (0/4) Epoch 29, batch 300, loss[loss=0.2184, simple_loss=0.2968, pruned_loss=0.06995, over 8504.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.284, pruned_loss=0.05842, over 1261071.08 frames. ], batch size: 28, lr: 2.62e-03, grad_scale: 16.0 2023-02-09 01:26:54,272 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226628.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:27:12,144 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226653.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:27:25,801 INFO [train.py:901] (0/4) Epoch 29, batch 350, loss[loss=0.2007, simple_loss=0.2814, pruned_loss=0.06002, over 7648.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2839, pruned_loss=0.05799, over 1343812.25 frames. ], batch size: 19, lr: 2.62e-03, grad_scale: 16.0 2023-02-09 01:27:32,882 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.36 vs. limit=2.0 2023-02-09 01:27:54,191 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226710.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:27:58,934 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.666e+02 2.400e+02 2.862e+02 3.557e+02 6.632e+02, threshold=5.725e+02, percent-clipped=2.0 2023-02-09 01:28:01,714 INFO [train.py:901] (0/4) Epoch 29, batch 400, loss[loss=0.2033, simple_loss=0.2942, pruned_loss=0.05622, over 8328.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.283, pruned_loss=0.05705, over 1405327.29 frames. ], batch size: 25, lr: 2.61e-03, grad_scale: 16.0 2023-02-09 01:28:14,639 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.0703, 1.6300, 1.4250, 1.5331, 1.3449, 1.3110, 1.3372, 1.3354], device='cuda:0'), covar=tensor([0.1295, 0.0537, 0.1464, 0.0694, 0.0848, 0.1710, 0.0985, 0.0897], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0244, 0.0342, 0.0313, 0.0301, 0.0347, 0.0349, 0.0321], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 01:28:37,572 INFO [train.py:901] (0/4) Epoch 29, batch 450, loss[loss=0.214, simple_loss=0.3032, pruned_loss=0.06238, over 8333.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2814, pruned_loss=0.05639, over 1448392.96 frames. ], batch size: 26, lr: 2.61e-03, grad_scale: 16.0 2023-02-09 01:28:39,871 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226774.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:28:49,335 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([3.2443, 1.4132, 3.3987, 1.1712, 3.0212, 2.8805, 3.1149, 3.0326], device='cuda:0'), covar=tensor([0.0873, 0.3946, 0.0820, 0.4186, 0.1423, 0.1070, 0.0773, 0.0916], device='cuda:0'), in_proj_covar=tensor([0.0683, 0.0664, 0.0736, 0.0658, 0.0746, 0.0633, 0.0644, 0.0715], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 01:28:58,215 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226799.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:29:11,134 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.602e+02 2.449e+02 2.964e+02 3.856e+02 9.700e+02, threshold=5.929e+02, percent-clipped=9.0 2023-02-09 01:29:13,796 INFO [train.py:901] (0/4) Epoch 29, batch 500, loss[loss=0.2723, simple_loss=0.3349, pruned_loss=0.1048, over 7011.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2805, pruned_loss=0.05666, over 1485104.72 frames. ], batch size: 72, lr: 2.61e-03, grad_scale: 16.0 2023-02-09 01:29:13,946 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([3.2326, 1.4471, 3.3871, 1.1588, 3.0283, 2.8408, 3.0815, 3.0260], device='cuda:0'), covar=tensor([0.0857, 0.4046, 0.0825, 0.4321, 0.1270, 0.1059, 0.0814, 0.0871], device='cuda:0'), in_proj_covar=tensor([0.0683, 0.0663, 0.0736, 0.0658, 0.0746, 0.0633, 0.0645, 0.0715], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 01:29:48,217 INFO [train.py:901] (0/4) Epoch 29, batch 550, loss[loss=0.1695, simple_loss=0.2529, pruned_loss=0.04308, over 7810.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.281, pruned_loss=0.05632, over 1516348.00 frames. ], batch size: 20, lr: 2.61e-03, grad_scale: 16.0 2023-02-09 01:30:21,909 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.374e+02 2.486e+02 3.156e+02 4.092e+02 1.034e+03, threshold=6.313e+02, percent-clipped=6.0 2023-02-09 01:30:24,630 INFO [train.py:901] (0/4) Epoch 29, batch 600, loss[loss=0.1869, simple_loss=0.2715, pruned_loss=0.05113, over 7790.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2821, pruned_loss=0.05686, over 1538741.34 frames. ], batch size: 19, lr: 2.61e-03, grad_scale: 16.0 2023-02-09 01:30:28,553 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=4.05 vs. limit=5.0 2023-02-09 01:30:35,120 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.3189, 1.5849, 4.2729, 2.1680, 2.7086, 4.7743, 4.8944, 4.1803], device='cuda:0'), covar=tensor([0.1152, 0.2008, 0.0333, 0.1815, 0.1135, 0.0230, 0.0595, 0.0592], device='cuda:0'), in_proj_covar=tensor([0.0310, 0.0327, 0.0295, 0.0325, 0.0327, 0.0278, 0.0444, 0.0309], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 01:30:38,681 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.3731, 2.3144, 3.0256, 2.5518, 3.1210, 2.5001, 2.3061, 2.0561], device='cuda:0'), covar=tensor([0.6014, 0.5550, 0.2296, 0.4432, 0.2778, 0.3512, 0.2024, 0.5895], device='cuda:0'), in_proj_covar=tensor([0.0965, 0.1033, 0.0838, 0.0999, 0.1026, 0.0936, 0.0772, 0.0852], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 01:30:43,085 WARNING [train.py:1067] (0/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-02-09 01:30:56,518 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226966.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:30:59,760 INFO [train.py:901] (0/4) Epoch 29, batch 650, loss[loss=0.1747, simple_loss=0.2479, pruned_loss=0.05069, over 7710.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2813, pruned_loss=0.0565, over 1559149.99 frames. ], batch size: 18, lr: 2.61e-03, grad_scale: 16.0 2023-02-09 01:31:13,854 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226991.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:31:25,436 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=227007.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:31:32,755 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.693e+02 2.425e+02 2.942e+02 3.859e+02 6.314e+02, threshold=5.885e+02, percent-clipped=1.0 2023-02-09 01:31:36,265 INFO [train.py:901] (0/4) Epoch 29, batch 700, loss[loss=0.1673, simple_loss=0.2623, pruned_loss=0.03615, over 8238.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2821, pruned_loss=0.05663, over 1572679.32 frames. ], batch size: 22, lr: 2.61e-03, grad_scale: 16.0 2023-02-09 01:31:56,849 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.6340, 1.7491, 1.5192, 2.3159, 0.9861, 1.4209, 1.6697, 1.8173], device='cuda:0'), covar=tensor([0.0900, 0.0833, 0.1134, 0.0385, 0.1107, 0.1418, 0.0814, 0.0773], device='cuda:0'), in_proj_covar=tensor([0.0231, 0.0196, 0.0245, 0.0214, 0.0204, 0.0249, 0.0251, 0.0206], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], device='cuda:0') 2023-02-09 01:32:12,700 INFO [train.py:901] (0/4) Epoch 29, batch 750, loss[loss=0.2002, simple_loss=0.2889, pruned_loss=0.05579, over 8103.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.282, pruned_loss=0.05663, over 1581478.93 frames. ], batch size: 23, lr: 2.61e-03, grad_scale: 16.0 2023-02-09 01:32:31,397 WARNING [train.py:1067] (0/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-02-09 01:32:40,245 WARNING [train.py:1067] (0/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-02-09 01:32:44,348 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.694e+02 2.547e+02 3.055e+02 4.023e+02 1.198e+03, threshold=6.109e+02, percent-clipped=3.0 2023-02-09 01:32:47,763 INFO [train.py:901] (0/4) Epoch 29, batch 800, loss[loss=0.1771, simple_loss=0.2738, pruned_loss=0.0402, over 8250.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.281, pruned_loss=0.05624, over 1590400.40 frames. ], batch size: 24, lr: 2.61e-03, grad_scale: 16.0 2023-02-09 01:33:24,585 INFO [train.py:901] (0/4) Epoch 29, batch 850, loss[loss=0.2077, simple_loss=0.293, pruned_loss=0.0612, over 8454.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.281, pruned_loss=0.05642, over 1595832.33 frames. ], batch size: 49, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:33:57,034 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.572e+02 2.463e+02 2.886e+02 3.949e+02 8.845e+02, threshold=5.773e+02, percent-clipped=3.0 2023-02-09 01:33:59,158 INFO [train.py:901] (0/4) Epoch 29, batch 900, loss[loss=0.1557, simple_loss=0.2491, pruned_loss=0.03114, over 7525.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.2802, pruned_loss=0.05596, over 1604060.00 frames. ], batch size: 18, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:34:35,617 INFO [train.py:901] (0/4) Epoch 29, batch 950, loss[loss=0.1853, simple_loss=0.2706, pruned_loss=0.04997, over 8027.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2812, pruned_loss=0.05658, over 1608748.34 frames. ], batch size: 22, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:35:04,862 WARNING [train.py:1067] (0/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-02-09 01:35:09,007 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.642e+02 2.581e+02 3.033e+02 3.905e+02 1.035e+03, threshold=6.066e+02, percent-clipped=4.0 2023-02-09 01:35:11,169 INFO [train.py:901] (0/4) Epoch 29, batch 1000, loss[loss=0.2136, simple_loss=0.2821, pruned_loss=0.07253, over 8239.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2814, pruned_loss=0.05664, over 1614332.78 frames. ], batch size: 22, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:35:32,311 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=227351.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:35:39,825 WARNING [train.py:1067] (0/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-02-09 01:35:47,253 INFO [train.py:901] (0/4) Epoch 29, batch 1050, loss[loss=0.2214, simple_loss=0.3149, pruned_loss=0.06394, over 8181.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2821, pruned_loss=0.05681, over 1617999.33 frames. ], batch size: 23, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:35:52,715 WARNING [train.py:1067] (0/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-02-09 01:36:22,054 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.565e+02 2.373e+02 3.020e+02 3.651e+02 1.051e+03, threshold=6.040e+02, percent-clipped=1.0 2023-02-09 01:36:23,043 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.8655, 1.6742, 2.4783, 1.6818, 1.4407, 2.3766, 0.4703, 1.6063], device='cuda:0'), covar=tensor([0.1566, 0.1420, 0.0368, 0.1038, 0.2380, 0.0474, 0.2041, 0.1296], device='cuda:0'), in_proj_covar=tensor([0.0202, 0.0207, 0.0137, 0.0224, 0.0278, 0.0147, 0.0173, 0.0199], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 01:36:24,330 INFO [train.py:901] (0/4) Epoch 29, batch 1100, loss[loss=0.1907, simple_loss=0.2655, pruned_loss=0.05795, over 7663.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2807, pruned_loss=0.05615, over 1616235.04 frames. ], batch size: 19, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:36:30,390 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=227429.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:36:52,864 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.9699, 2.1102, 2.2823, 2.0353, 1.1796, 2.0424, 2.3046, 2.6270], device='cuda:0'), covar=tensor([0.0421, 0.1094, 0.1570, 0.1287, 0.0573, 0.1307, 0.0614, 0.0504], device='cuda:0'), in_proj_covar=tensor([0.0100, 0.0154, 0.0191, 0.0163, 0.0102, 0.0164, 0.0114, 0.0147], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0') 2023-02-09 01:36:54,943 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.3139, 2.1695, 1.7918, 2.0512, 1.8562, 1.5364, 1.7885, 1.7845], device='cuda:0'), covar=tensor([0.1206, 0.0443, 0.1128, 0.0533, 0.0672, 0.1478, 0.0848, 0.0780], device='cuda:0'), in_proj_covar=tensor([0.0358, 0.0246, 0.0344, 0.0314, 0.0303, 0.0349, 0.0352, 0.0323], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 01:36:56,368 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=227466.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:36:59,550 INFO [train.py:901] (0/4) Epoch 29, batch 1150, loss[loss=0.1907, simple_loss=0.2761, pruned_loss=0.05263, over 8501.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2808, pruned_loss=0.05587, over 1617699.69 frames. ], batch size: 26, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:37:06,476 WARNING [train.py:1067] (0/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-02-09 01:37:34,267 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.484e+02 2.320e+02 2.698e+02 3.625e+02 7.425e+02, threshold=5.396e+02, percent-clipped=3.0 2023-02-09 01:37:36,428 INFO [train.py:901] (0/4) Epoch 29, batch 1200, loss[loss=0.1741, simple_loss=0.2557, pruned_loss=0.04626, over 7925.00 frames. ], tot_loss[loss=0.195, simple_loss=0.2796, pruned_loss=0.05522, over 1612576.80 frames. ], batch size: 20, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:37:57,387 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.5858, 2.0861, 3.2078, 1.4730, 2.5849, 2.0802, 1.6716, 2.5348], device='cuda:0'), covar=tensor([0.2032, 0.2620, 0.0985, 0.4832, 0.1890, 0.3492, 0.2566, 0.2298], device='cuda:0'), in_proj_covar=tensor([0.0540, 0.0640, 0.0565, 0.0672, 0.0667, 0.0618, 0.0566, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:0') 2023-02-09 01:38:11,533 INFO [train.py:901] (0/4) Epoch 29, batch 1250, loss[loss=0.2216, simple_loss=0.302, pruned_loss=0.07065, over 8575.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2803, pruned_loss=0.05606, over 1613472.91 frames. ], batch size: 31, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:38:46,455 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.451e+02 2.398e+02 2.828e+02 3.393e+02 7.704e+02, threshold=5.657e+02, percent-clipped=4.0 2023-02-09 01:38:48,672 INFO [train.py:901] (0/4) Epoch 29, batch 1300, loss[loss=0.1667, simple_loss=0.2556, pruned_loss=0.03889, over 8074.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.28, pruned_loss=0.05573, over 1618466.34 frames. ], batch size: 21, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:39:24,557 INFO [train.py:901] (0/4) Epoch 29, batch 1350, loss[loss=0.163, simple_loss=0.2449, pruned_loss=0.04053, over 7546.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2805, pruned_loss=0.05624, over 1620929.17 frames. ], batch size: 18, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:39:52,816 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([4.7221, 1.4707, 4.9020, 1.8136, 4.4577, 4.0955, 4.4297, 4.3435], device='cuda:0'), covar=tensor([0.0544, 0.4749, 0.0440, 0.4298, 0.0857, 0.0827, 0.0518, 0.0586], device='cuda:0'), in_proj_covar=tensor([0.0680, 0.0664, 0.0737, 0.0657, 0.0745, 0.0632, 0.0641, 0.0714], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 01:39:52,826 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=227711.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:39:58,209 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.335e+02 2.785e+02 3.650e+02 1.055e+03, threshold=5.570e+02, percent-clipped=4.0 2023-02-09 01:40:00,382 INFO [train.py:901] (0/4) Epoch 29, batch 1400, loss[loss=0.1891, simple_loss=0.2735, pruned_loss=0.05234, over 8477.00 frames. ], tot_loss[loss=0.1953, simple_loss=0.2793, pruned_loss=0.05566, over 1620420.49 frames. ], batch size: 27, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:40:01,320 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=227722.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:40:02,248 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=3.97 vs. limit=5.0 2023-02-09 01:40:11,989 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.2493, 3.6080, 2.5076, 3.1598, 2.9288, 2.1827, 2.9538, 3.1283], device='cuda:0'), covar=tensor([0.1778, 0.0461, 0.1106, 0.0664, 0.0834, 0.1480, 0.1098, 0.1180], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0247, 0.0344, 0.0315, 0.0304, 0.0348, 0.0352, 0.0324], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 01:40:20,617 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=227747.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:40:38,107 INFO [train.py:901] (0/4) Epoch 29, batch 1450, loss[loss=0.1845, simple_loss=0.2571, pruned_loss=0.05598, over 7242.00 frames. ], tot_loss[loss=0.1956, simple_loss=0.2792, pruned_loss=0.05601, over 1616089.67 frames. ], batch size: 16, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:40:39,473 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=227773.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:40:47,996 WARNING [train.py:1067] (0/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-02-09 01:41:11,905 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.751e+02 2.369e+02 2.795e+02 3.434e+02 1.018e+03, threshold=5.589e+02, percent-clipped=3.0 2023-02-09 01:41:14,107 INFO [train.py:901] (0/4) Epoch 29, batch 1500, loss[loss=0.1935, simple_loss=0.2836, pruned_loss=0.05174, over 8604.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2805, pruned_loss=0.05629, over 1619357.28 frames. ], batch size: 34, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:41:31,885 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=227845.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:41:48,977 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=2.64 vs. limit=5.0 2023-02-09 01:41:51,483 INFO [train.py:901] (0/4) Epoch 29, batch 1550, loss[loss=0.1772, simple_loss=0.2631, pruned_loss=0.04565, over 7975.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2811, pruned_loss=0.05674, over 1619696.12 frames. ], batch size: 21, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:42:04,423 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=227888.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:42:25,548 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.467e+02 2.452e+02 3.275e+02 4.864e+02 1.208e+03, threshold=6.551e+02, percent-clipped=17.0 2023-02-09 01:42:27,701 INFO [train.py:901] (0/4) Epoch 29, batch 1600, loss[loss=0.2186, simple_loss=0.3073, pruned_loss=0.06493, over 8609.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2805, pruned_loss=0.05636, over 1618512.46 frames. ], batch size: 31, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:43:04,959 INFO [train.py:901] (0/4) Epoch 29, batch 1650, loss[loss=0.1921, simple_loss=0.2789, pruned_loss=0.05262, over 8194.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2811, pruned_loss=0.05652, over 1617926.25 frames. ], batch size: 23, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:43:26,177 INFO [checkpoint.py:75] (0/4) Saving checkpoint to pruned_transducer_stateless7_streaming/exp/v1/checkpoint-228000.pt 2023-02-09 01:43:39,885 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.526e+02 2.401e+02 2.687e+02 3.396e+02 6.045e+02, threshold=5.374e+02, percent-clipped=0.0 2023-02-09 01:43:42,082 INFO [train.py:901] (0/4) Epoch 29, batch 1700, loss[loss=0.183, simple_loss=0.2663, pruned_loss=0.04983, over 8082.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2808, pruned_loss=0.05613, over 1617164.64 frames. ], batch size: 21, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:43:45,197 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228025.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:44:06,274 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=228055.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:44:18,095 INFO [train.py:901] (0/4) Epoch 29, batch 1750, loss[loss=0.1927, simple_loss=0.2675, pruned_loss=0.05901, over 7926.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.281, pruned_loss=0.0564, over 1613731.31 frames. ], batch size: 20, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:44:38,634 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228098.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:44:52,736 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.427e+02 2.425e+02 2.925e+02 3.463e+02 6.679e+02, threshold=5.849e+02, percent-clipped=2.0 2023-02-09 01:44:55,477 INFO [train.py:901] (0/4) Epoch 29, batch 1800, loss[loss=0.2156, simple_loss=0.2903, pruned_loss=0.07042, over 8575.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2813, pruned_loss=0.05668, over 1613733.06 frames. ], batch size: 49, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:45:11,249 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228144.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:45:12,124 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.80 vs. limit=2.0 2023-02-09 01:45:28,616 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228169.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:45:29,250 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228170.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:45:29,719 INFO [train.py:901] (0/4) Epoch 29, batch 1850, loss[loss=0.2177, simple_loss=0.2926, pruned_loss=0.0714, over 7925.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2827, pruned_loss=0.05758, over 1618168.49 frames. ], batch size: 20, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:45:42,838 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=228189.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:46:03,754 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.759e+02 2.509e+02 2.932e+02 3.424e+02 5.958e+02, threshold=5.864e+02, percent-clipped=1.0 2023-02-09 01:46:05,871 INFO [train.py:901] (0/4) Epoch 29, batch 1900, loss[loss=0.1533, simple_loss=0.2347, pruned_loss=0.03592, over 7770.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2822, pruned_loss=0.05723, over 1616225.78 frames. ], batch size: 19, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:46:29,148 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=2.43 vs. limit=5.0 2023-02-09 01:46:32,895 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.29 vs. limit=2.0 2023-02-09 01:46:38,538 WARNING [train.py:1067] (0/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-02-09 01:46:41,408 INFO [train.py:901] (0/4) Epoch 29, batch 1950, loss[loss=0.193, simple_loss=0.2885, pruned_loss=0.04875, over 8447.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2818, pruned_loss=0.05726, over 1615060.74 frames. ], batch size: 27, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:46:50,974 WARNING [train.py:1067] (0/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-02-09 01:47:05,321 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228304.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:47:10,124 WARNING [train.py:1067] (0/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-02-09 01:47:16,245 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.886e+02 2.422e+02 3.056e+02 3.762e+02 8.552e+02, threshold=6.111e+02, percent-clipped=4.0 2023-02-09 01:47:18,246 INFO [train.py:901] (0/4) Epoch 29, batch 2000, loss[loss=0.1997, simple_loss=0.2869, pruned_loss=0.05621, over 8524.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2815, pruned_loss=0.05725, over 1613990.64 frames. ], batch size: 26, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:47:36,281 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228347.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:47:39,115 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228351.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:47:44,568 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.2928, 2.5967, 2.9491, 1.5997, 3.2690, 1.8484, 1.4964, 2.1899], device='cuda:0'), covar=tensor([0.0861, 0.0431, 0.0293, 0.0891, 0.0535, 0.0949, 0.1181, 0.0642], device='cuda:0'), in_proj_covar=tensor([0.0476, 0.0414, 0.0369, 0.0460, 0.0397, 0.0552, 0.0403, 0.0443], device='cuda:0'), out_proj_covar=tensor([1.2611e-04, 1.0730e-04, 9.6024e-05, 1.2017e-04, 1.0384e-04, 1.5379e-04, 1.0767e-04, 1.1589e-04], device='cuda:0') 2023-02-09 01:47:52,204 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=228369.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:47:53,548 INFO [train.py:901] (0/4) Epoch 29, batch 2050, loss[loss=0.1871, simple_loss=0.2757, pruned_loss=0.04929, over 8460.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.281, pruned_loss=0.05676, over 1615077.39 frames. ], batch size: 29, lr: 2.61e-03, grad_scale: 8.0 2023-02-09 01:48:25,964 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.725e+02 2.333e+02 2.866e+02 3.600e+02 5.490e+02, threshold=5.733e+02, percent-clipped=0.0 2023-02-09 01:48:28,131 INFO [train.py:901] (0/4) Epoch 29, batch 2100, loss[loss=0.2378, simple_loss=0.3115, pruned_loss=0.08203, over 8479.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2817, pruned_loss=0.05689, over 1613202.16 frames. ], batch size: 39, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 01:48:32,356 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228426.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:48:43,922 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=228442.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:48:48,021 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228447.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 01:48:50,751 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228451.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:48:59,127 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.5728, 1.8847, 2.8606, 1.5386, 2.2596, 1.9846, 1.6822, 2.3051], device='cuda:0'), covar=tensor([0.2136, 0.2904, 0.1090, 0.4937, 0.1999, 0.3548, 0.2676, 0.2298], device='cuda:0'), in_proj_covar=tensor([0.0543, 0.0642, 0.0568, 0.0675, 0.0667, 0.0617, 0.0566, 0.0650], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:0') 2023-02-09 01:49:04,310 INFO [train.py:901] (0/4) Epoch 29, batch 2150, loss[loss=0.1867, simple_loss=0.2658, pruned_loss=0.05381, over 7813.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2819, pruned_loss=0.05705, over 1611736.15 frames. ], batch size: 20, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 01:49:05,173 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([4.5864, 1.5336, 4.7961, 1.8327, 4.2762, 4.0413, 4.3369, 4.2031], device='cuda:0'), covar=tensor([0.0557, 0.4658, 0.0474, 0.4107, 0.1070, 0.0888, 0.0529, 0.0659], device='cuda:0'), in_proj_covar=tensor([0.0686, 0.0669, 0.0744, 0.0664, 0.0750, 0.0639, 0.0647, 0.0721], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 01:49:14,153 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228484.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:49:37,708 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.627e+02 2.659e+02 3.270e+02 3.951e+02 1.171e+03, threshold=6.540e+02, percent-clipped=10.0 2023-02-09 01:49:39,913 INFO [train.py:901] (0/4) Epoch 29, batch 2200, loss[loss=0.1756, simple_loss=0.2739, pruned_loss=0.03863, over 8470.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2812, pruned_loss=0.05665, over 1611788.95 frames. ], batch size: 25, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 01:49:47,027 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([4.3508, 1.5309, 4.5631, 1.7009, 4.0246, 3.8058, 4.1351, 4.0057], device='cuda:0'), covar=tensor([0.0624, 0.4602, 0.0518, 0.4356, 0.1065, 0.0940, 0.0594, 0.0668], device='cuda:0'), in_proj_covar=tensor([0.0684, 0.0666, 0.0742, 0.0662, 0.0747, 0.0637, 0.0644, 0.0719], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 01:50:06,281 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228557.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:50:08,437 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228560.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:50:16,509 INFO [train.py:901] (0/4) Epoch 29, batch 2250, loss[loss=0.1664, simple_loss=0.2476, pruned_loss=0.04262, over 8053.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2802, pruned_loss=0.05624, over 1607187.19 frames. ], batch size: 20, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 01:50:26,603 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228585.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:50:44,222 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.0412, 1.7336, 4.3838, 1.7604, 3.0075, 4.9105, 5.2309, 3.9115], device='cuda:0'), covar=tensor([0.1679, 0.2424, 0.0378, 0.2653, 0.1054, 0.0282, 0.0476, 0.0911], device='cuda:0'), in_proj_covar=tensor([0.0311, 0.0327, 0.0297, 0.0325, 0.0329, 0.0279, 0.0447, 0.0312], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 01:50:50,343 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.529e+02 2.350e+02 2.877e+02 3.583e+02 6.549e+02, threshold=5.755e+02, percent-clipped=1.0 2023-02-09 01:50:52,571 INFO [train.py:901] (0/4) Epoch 29, batch 2300, loss[loss=0.1656, simple_loss=0.2449, pruned_loss=0.04313, over 7418.00 frames. ], tot_loss[loss=0.1956, simple_loss=0.2793, pruned_loss=0.05591, over 1607966.75 frames. ], batch size: 17, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 01:50:53,452 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.5404, 1.4211, 1.8357, 1.2778, 1.1819, 1.8064, 0.3342, 1.2266], device='cuda:0'), covar=tensor([0.1227, 0.1133, 0.0387, 0.0788, 0.2232, 0.0480, 0.1681, 0.1121], device='cuda:0'), in_proj_covar=tensor([0.0202, 0.0206, 0.0138, 0.0223, 0.0278, 0.0149, 0.0173, 0.0199], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 01:51:06,312 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.79 vs. limit=2.0 2023-02-09 01:51:22,789 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([5.2360, 2.1809, 5.3969, 2.3506, 4.8815, 4.5692, 4.9305, 4.8032], device='cuda:0'), covar=tensor([0.0576, 0.4258, 0.0439, 0.4040, 0.0991, 0.0991, 0.0528, 0.0627], device='cuda:0'), in_proj_covar=tensor([0.0686, 0.0669, 0.0746, 0.0666, 0.0751, 0.0640, 0.0647, 0.0722], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 01:51:28,773 INFO [train.py:901] (0/4) Epoch 29, batch 2350, loss[loss=0.2456, simple_loss=0.3225, pruned_loss=0.08436, over 8770.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.281, pruned_loss=0.05704, over 1610128.05 frames. ], batch size: 40, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 01:51:43,129 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=228691.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:51:45,871 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=228695.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:52:02,330 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.786e+02 2.434e+02 3.194e+02 3.873e+02 7.294e+02, threshold=6.388e+02, percent-clipped=4.0 2023-02-09 01:52:04,277 INFO [train.py:901] (0/4) Epoch 29, batch 2400, loss[loss=0.2158, simple_loss=0.2979, pruned_loss=0.06689, over 8237.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2808, pruned_loss=0.0573, over 1604690.68 frames. ], batch size: 24, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 01:52:17,436 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228740.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:52:34,952 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228765.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:52:39,605 INFO [train.py:901] (0/4) Epoch 29, batch 2450, loss[loss=0.1819, simple_loss=0.2671, pruned_loss=0.04833, over 8478.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2802, pruned_loss=0.05686, over 1607090.21 frames. ], batch size: 25, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 01:52:55,034 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=228791.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 01:53:05,501 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228806.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:53:08,258 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228810.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:53:10,301 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228813.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:53:13,500 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.470e+02 2.278e+02 2.630e+02 3.527e+02 5.802e+02, threshold=5.259e+02, percent-clipped=0.0 2023-02-09 01:53:16,190 INFO [train.py:901] (0/4) Epoch 29, batch 2500, loss[loss=0.1982, simple_loss=0.2844, pruned_loss=0.05602, over 8236.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.281, pruned_loss=0.05769, over 1607495.52 frames. ], batch size: 22, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 01:53:28,185 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228838.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:53:49,912 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-02-09 01:53:51,004 INFO [train.py:901] (0/4) Epoch 29, batch 2550, loss[loss=0.1909, simple_loss=0.2715, pruned_loss=0.05511, over 8144.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2804, pruned_loss=0.05715, over 1606500.94 frames. ], batch size: 22, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 01:54:17,030 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228906.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 01:54:25,706 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.542e+02 2.320e+02 2.761e+02 3.420e+02 6.403e+02, threshold=5.523e+02, percent-clipped=2.0 2023-02-09 01:54:27,396 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.6885, 2.1127, 3.1900, 1.5771, 2.5054, 2.1521, 1.7461, 2.5489], device='cuda:0'), covar=tensor([0.1850, 0.2622, 0.0816, 0.4543, 0.1754, 0.3054, 0.2376, 0.2143], device='cuda:0'), in_proj_covar=tensor([0.0546, 0.0644, 0.0569, 0.0679, 0.0670, 0.0619, 0.0568, 0.0652], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:0') 2023-02-09 01:54:27,845 INFO [train.py:901] (0/4) Epoch 29, batch 2600, loss[loss=0.2027, simple_loss=0.2865, pruned_loss=0.05941, over 8668.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2802, pruned_loss=0.05698, over 1610469.30 frames. ], batch size: 34, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 01:54:51,304 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228953.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:55:03,877 INFO [train.py:901] (0/4) Epoch 29, batch 2650, loss[loss=0.2036, simple_loss=0.2884, pruned_loss=0.05941, over 8597.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2808, pruned_loss=0.05717, over 1610205.78 frames. ], batch size: 40, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 01:55:27,205 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([4.5833, 1.4838, 4.8111, 1.8809, 4.3191, 3.9860, 4.3465, 4.2360], device='cuda:0'), covar=tensor([0.0536, 0.4667, 0.0457, 0.4062, 0.0903, 0.0863, 0.0474, 0.0598], device='cuda:0'), in_proj_covar=tensor([0.0681, 0.0663, 0.0741, 0.0662, 0.0743, 0.0635, 0.0640, 0.0715], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 01:55:37,363 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.671e+02 2.477e+02 2.922e+02 3.574e+02 8.790e+02, threshold=5.845e+02, percent-clipped=2.0 2023-02-09 01:55:39,871 INFO [train.py:901] (0/4) Epoch 29, batch 2700, loss[loss=0.197, simple_loss=0.2895, pruned_loss=0.05219, over 8336.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2803, pruned_loss=0.05663, over 1611910.41 frames. ], batch size: 26, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 01:56:11,113 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229062.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:56:14,606 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229066.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:56:17,759 INFO [train.py:901] (0/4) Epoch 29, batch 2750, loss[loss=0.2003, simple_loss=0.2967, pruned_loss=0.05193, over 8513.00 frames. ], tot_loss[loss=0.1947, simple_loss=0.2785, pruned_loss=0.05545, over 1608919.74 frames. ], batch size: 28, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 01:56:29,435 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229087.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:56:32,358 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229091.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:56:51,304 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.676e+02 2.406e+02 2.857e+02 3.824e+02 7.570e+02, threshold=5.715e+02, percent-clipped=1.0 2023-02-09 01:56:53,439 INFO [train.py:901] (0/4) Epoch 29, batch 2800, loss[loss=0.2345, simple_loss=0.3008, pruned_loss=0.08412, over 8459.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2804, pruned_loss=0.05641, over 1614778.50 frames. ], batch size: 29, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 01:56:54,332 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.8966, 1.3038, 1.5905, 1.2705, 0.9694, 1.4194, 1.7587, 1.7331], device='cuda:0'), covar=tensor([0.0601, 0.1745, 0.2306, 0.1908, 0.0678, 0.1994, 0.0765, 0.0698], device='cuda:0'), in_proj_covar=tensor([0.0100, 0.0154, 0.0190, 0.0162, 0.0102, 0.0163, 0.0114, 0.0148], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], device='cuda:0') 2023-02-09 01:57:24,927 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229162.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 01:57:31,026 INFO [train.py:901] (0/4) Epoch 29, batch 2850, loss[loss=0.1899, simple_loss=0.293, pruned_loss=0.04339, over 8107.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2804, pruned_loss=0.056, over 1614802.80 frames. ], batch size: 23, lr: 2.60e-03, grad_scale: 16.0 2023-02-09 01:57:43,233 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229187.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 01:58:03,961 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.9678, 1.6639, 3.2255, 1.5202, 2.4197, 3.5325, 3.7035, 3.0050], device='cuda:0'), covar=tensor([0.1295, 0.1791, 0.0409, 0.2231, 0.1061, 0.0268, 0.0600, 0.0595], device='cuda:0'), in_proj_covar=tensor([0.0314, 0.0330, 0.0299, 0.0327, 0.0331, 0.0282, 0.0451, 0.0314], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], device='cuda:0') 2023-02-09 01:58:05,222 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.341e+02 2.443e+02 3.095e+02 3.828e+02 9.615e+02, threshold=6.189e+02, percent-clipped=4.0 2023-02-09 01:58:07,411 INFO [train.py:901] (0/4) Epoch 29, batch 2900, loss[loss=0.1825, simple_loss=0.2752, pruned_loss=0.04493, over 8030.00 frames. ], tot_loss[loss=0.1954, simple_loss=0.2797, pruned_loss=0.0556, over 1613522.77 frames. ], batch size: 22, lr: 2.60e-03, grad_scale: 16.0 2023-02-09 01:58:21,868 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-02-09 01:58:23,785 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.9101, 1.4579, 1.6447, 1.4206, 0.9587, 1.5137, 1.7389, 1.4557], device='cuda:0'), covar=tensor([0.0554, 0.1334, 0.1762, 0.1509, 0.0645, 0.1552, 0.0718, 0.0722], device='cuda:0'), in_proj_covar=tensor([0.0101, 0.0155, 0.0191, 0.0163, 0.0103, 0.0165, 0.0114, 0.0148], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0007, 0.0008], device='cuda:0') 2023-02-09 01:58:44,222 INFO [train.py:901] (0/4) Epoch 29, batch 2950, loss[loss=0.1658, simple_loss=0.2527, pruned_loss=0.03947, over 7815.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2804, pruned_loss=0.05614, over 1616755.75 frames. ], batch size: 20, lr: 2.60e-03, grad_scale: 16.0 2023-02-09 01:58:47,089 WARNING [train.py:1067] (0/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-02-09 01:59:02,251 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=229297.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:59:17,270 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.610e+02 2.481e+02 3.005e+02 3.741e+02 9.617e+02, threshold=6.010e+02, percent-clipped=3.0 2023-02-09 01:59:17,395 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229318.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 01:59:19,280 INFO [train.py:901] (0/4) Epoch 29, batch 3000, loss[loss=0.1777, simple_loss=0.2523, pruned_loss=0.05156, over 7423.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2814, pruned_loss=0.05671, over 1618580.77 frames. ], batch size: 17, lr: 2.60e-03, grad_scale: 16.0 2023-02-09 01:59:19,280 INFO [train.py:926] (0/4) Computing validation loss 2023-02-09 01:59:34,615 INFO [train.py:935] (0/4) Epoch 29, validation: loss=0.17, simple_loss=0.2699, pruned_loss=0.03504, over 944034.00 frames. 2023-02-09 01:59:34,615 INFO [train.py:936] (0/4) Maximum memory allocated so far is 6641MB 2023-02-09 01:59:54,815 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([3.8301, 3.7935, 3.4467, 1.8664, 3.3768, 3.4806, 3.3639, 3.3052], device='cuda:0'), covar=tensor([0.1006, 0.0697, 0.1241, 0.5341, 0.1088, 0.1230, 0.1506, 0.0959], device='cuda:0'), in_proj_covar=tensor([0.0550, 0.0464, 0.0457, 0.0568, 0.0446, 0.0471, 0.0447, 0.0415], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 01:59:58,445 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.8194, 1.7237, 2.4974, 1.5656, 1.4234, 2.4261, 0.5007, 1.5305], device='cuda:0'), covar=tensor([0.1746, 0.1166, 0.0333, 0.1164, 0.2216, 0.0372, 0.1898, 0.1228], device='cuda:0'), in_proj_covar=tensor([0.0205, 0.0208, 0.0139, 0.0225, 0.0281, 0.0149, 0.0174, 0.0202], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 02:00:09,411 INFO [train.py:901] (0/4) Epoch 29, batch 3050, loss[loss=0.1716, simple_loss=0.2652, pruned_loss=0.03902, over 8103.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2821, pruned_loss=0.05696, over 1618155.04 frames. ], batch size: 23, lr: 2.60e-03, grad_scale: 16.0 2023-02-09 02:00:40,598 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=229412.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:00:44,468 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.402e+02 2.505e+02 2.815e+02 3.570e+02 7.212e+02, threshold=5.630e+02, percent-clipped=4.0 2023-02-09 02:00:46,507 INFO [train.py:901] (0/4) Epoch 29, batch 3100, loss[loss=0.2054, simple_loss=0.2906, pruned_loss=0.06005, over 7969.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2817, pruned_loss=0.05681, over 1617732.86 frames. ], batch size: 21, lr: 2.60e-03, grad_scale: 16.0 2023-02-09 02:01:15,197 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229461.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:01:21,910 INFO [train.py:901] (0/4) Epoch 29, batch 3150, loss[loss=0.2013, simple_loss=0.2903, pruned_loss=0.05612, over 8283.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.283, pruned_loss=0.05754, over 1613936.09 frames. ], batch size: 23, lr: 2.60e-03, grad_scale: 16.0 2023-02-09 02:01:37,529 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.14 vs. limit=2.0 2023-02-09 02:01:39,330 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229495.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:01:43,780 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.1657, 2.0005, 2.5793, 2.1926, 2.5352, 2.2728, 2.1193, 1.4613], device='cuda:0'), covar=tensor([0.6413, 0.5623, 0.2282, 0.4061, 0.2689, 0.3436, 0.1971, 0.5787], device='cuda:0'), in_proj_covar=tensor([0.0969, 0.1034, 0.0838, 0.1004, 0.1028, 0.0940, 0.0776, 0.0856], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 02:01:56,472 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.645e+02 2.412e+02 3.144e+02 3.923e+02 1.015e+03, threshold=6.289e+02, percent-clipped=11.0 2023-02-09 02:01:58,573 INFO [train.py:901] (0/4) Epoch 29, batch 3200, loss[loss=0.2266, simple_loss=0.2979, pruned_loss=0.07764, over 8081.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2828, pruned_loss=0.05767, over 1614515.47 frames. ], batch size: 21, lr: 2.60e-03, grad_scale: 16.0 2023-02-09 02:02:02,164 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.9003, 1.9884, 1.7518, 2.6117, 1.3230, 1.6360, 1.9810, 2.0676], device='cuda:0'), covar=tensor([0.0718, 0.0779, 0.0825, 0.0350, 0.1027, 0.1255, 0.0741, 0.0755], device='cuda:0'), in_proj_covar=tensor([0.0231, 0.0195, 0.0244, 0.0212, 0.0203, 0.0246, 0.0250, 0.0205], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], device='cuda:0') 2023-02-09 02:02:04,247 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([5.9435, 1.9362, 6.0733, 2.3791, 5.4891, 5.1090, 5.5977, 5.4760], device='cuda:0'), covar=tensor([0.0442, 0.4398, 0.0362, 0.3520, 0.0867, 0.0896, 0.0438, 0.0495], device='cuda:0'), in_proj_covar=tensor([0.0680, 0.0659, 0.0739, 0.0655, 0.0740, 0.0633, 0.0639, 0.0714], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 02:02:05,805 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.6805, 1.9662, 2.0296, 1.4225, 2.1144, 1.5831, 0.5426, 1.9460], device='cuda:0'), covar=tensor([0.0667, 0.0387, 0.0368, 0.0630, 0.0437, 0.1000, 0.1010, 0.0309], device='cuda:0'), in_proj_covar=tensor([0.0479, 0.0413, 0.0371, 0.0463, 0.0399, 0.0555, 0.0407, 0.0444], device='cuda:0'), out_proj_covar=tensor([1.2680e-04, 1.0684e-04, 9.6549e-05, 1.2098e-04, 1.0451e-04, 1.5445e-04, 1.0855e-04, 1.1610e-04], device='cuda:0') 2023-02-09 02:02:35,462 INFO [train.py:901] (0/4) Epoch 29, batch 3250, loss[loss=0.2153, simple_loss=0.2992, pruned_loss=0.06563, over 8195.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2828, pruned_loss=0.05755, over 1613288.28 frames. ], batch size: 23, lr: 2.60e-03, grad_scale: 16.0 2023-02-09 02:02:49,927 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=2.10 vs. limit=5.0 2023-02-09 02:02:52,613 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229595.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:02:58,035 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-02-09 02:03:09,567 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.589e+02 2.257e+02 2.746e+02 3.374e+02 9.131e+02, threshold=5.492e+02, percent-clipped=1.0 2023-02-09 02:03:11,682 INFO [train.py:901] (0/4) Epoch 29, batch 3300, loss[loss=0.2102, simple_loss=0.2945, pruned_loss=0.06297, over 8615.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.282, pruned_loss=0.05668, over 1616198.63 frames. ], batch size: 31, lr: 2.60e-03, grad_scale: 16.0 2023-02-09 02:03:20,264 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-02-09 02:03:41,131 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=229662.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:03:45,542 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229668.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:03:47,452 INFO [train.py:901] (0/4) Epoch 29, batch 3350, loss[loss=0.2328, simple_loss=0.3109, pruned_loss=0.07737, over 7221.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.281, pruned_loss=0.05619, over 1614235.40 frames. ], batch size: 71, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 02:03:52,168 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229677.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 02:04:03,326 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229693.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:04:20,879 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.769e+02 2.561e+02 2.994e+02 3.779e+02 7.703e+02, threshold=5.989e+02, percent-clipped=7.0 2023-02-09 02:04:22,315 INFO [train.py:901] (0/4) Epoch 29, batch 3400, loss[loss=0.1724, simple_loss=0.2662, pruned_loss=0.03935, over 8565.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2806, pruned_loss=0.05598, over 1614757.85 frames. ], batch size: 31, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 02:04:59,066 INFO [train.py:901] (0/4) Epoch 29, batch 3450, loss[loss=0.2145, simple_loss=0.2988, pruned_loss=0.06507, over 8438.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2812, pruned_loss=0.05658, over 1615343.96 frames. ], batch size: 27, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 02:05:00,795 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.2770, 2.1802, 2.6897, 2.3675, 2.6671, 2.4068, 2.2436, 1.5763], device='cuda:0'), covar=tensor([0.5981, 0.5102, 0.2181, 0.3502, 0.2405, 0.3131, 0.1971, 0.5338], device='cuda:0'), in_proj_covar=tensor([0.0969, 0.1033, 0.0840, 0.1004, 0.1028, 0.0941, 0.0776, 0.0856], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 02:05:03,507 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=229777.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:05:23,702 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=229805.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:05:33,250 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.613e+02 2.314e+02 2.735e+02 3.470e+02 1.051e+03, threshold=5.470e+02, percent-clipped=3.0 2023-02-09 02:05:34,602 INFO [train.py:901] (0/4) Epoch 29, batch 3500, loss[loss=0.2011, simple_loss=0.262, pruned_loss=0.07006, over 7540.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2805, pruned_loss=0.05609, over 1614768.14 frames. ], batch size: 18, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 02:05:37,546 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.2398, 2.1019, 2.6357, 2.2972, 2.6471, 2.3653, 2.2057, 1.4828], device='cuda:0'), covar=tensor([0.5964, 0.5248, 0.2156, 0.3933, 0.2708, 0.3188, 0.1859, 0.5687], device='cuda:0'), in_proj_covar=tensor([0.0970, 0.1034, 0.0841, 0.1005, 0.1029, 0.0941, 0.0776, 0.0857], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 02:05:38,249 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.7366, 2.0826, 2.0836, 1.4098, 2.3027, 1.5722, 0.7556, 2.0377], device='cuda:0'), covar=tensor([0.0699, 0.0390, 0.0374, 0.0709, 0.0379, 0.0983, 0.0987, 0.0334], device='cuda:0'), in_proj_covar=tensor([0.0482, 0.0416, 0.0374, 0.0466, 0.0402, 0.0558, 0.0410, 0.0448], device='cuda:0'), out_proj_covar=tensor([1.2768e-04, 1.0758e-04, 9.7337e-05, 1.2193e-04, 1.0517e-04, 1.5537e-04, 1.0928e-04, 1.1737e-04], device='cuda:0') 2023-02-09 02:05:46,931 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=229839.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:05:58,502 WARNING [train.py:1067] (0/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-02-09 02:06:12,022 INFO [train.py:901] (0/4) Epoch 29, batch 3550, loss[loss=0.2026, simple_loss=0.2969, pruned_loss=0.05416, over 8356.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2816, pruned_loss=0.05678, over 1613501.36 frames. ], batch size: 24, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 02:06:41,736 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229912.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:06:46,517 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.708e+02 2.466e+02 3.015e+02 3.666e+02 8.686e+02, threshold=6.030e+02, percent-clipped=2.0 2023-02-09 02:06:47,440 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=229920.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:06:47,948 INFO [train.py:901] (0/4) Epoch 29, batch 3600, loss[loss=0.2126, simple_loss=0.2878, pruned_loss=0.06871, over 7967.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2814, pruned_loss=0.05679, over 1610248.24 frames. ], batch size: 21, lr: 2.60e-03, grad_scale: 8.0 2023-02-09 02:07:00,991 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=229939.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:07:11,738 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=229954.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:07:24,209 INFO [train.py:901] (0/4) Epoch 29, batch 3650, loss[loss=0.2239, simple_loss=0.3069, pruned_loss=0.07047, over 8540.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2813, pruned_loss=0.05618, over 1616181.17 frames. ], batch size: 49, lr: 2.60e-03, grad_scale: 4.0 2023-02-09 02:07:45,869 INFO [checkpoint.py:75] (0/4) Saving checkpoint to pruned_transducer_stateless7_streaming/exp/v1/checkpoint-230000.pt 2023-02-09 02:07:50,399 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230005.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:08:00,967 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.417e+02 2.398e+02 2.862e+02 3.372e+02 7.881e+02, threshold=5.724e+02, percent-clipped=3.0 2023-02-09 02:08:01,714 INFO [train.py:901] (0/4) Epoch 29, batch 3700, loss[loss=0.1704, simple_loss=0.2449, pruned_loss=0.04798, over 7235.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.2804, pruned_loss=0.05569, over 1608890.94 frames. ], batch size: 16, lr: 2.60e-03, grad_scale: 4.0 2023-02-09 02:08:02,482 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=230021.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 02:08:06,541 WARNING [train.py:1067] (0/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-02-09 02:08:10,879 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230033.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:08:25,550 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230054.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:08:28,467 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230058.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:08:37,280 INFO [train.py:901] (0/4) Epoch 29, batch 3750, loss[loss=0.2031, simple_loss=0.2854, pruned_loss=0.06039, over 7236.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2814, pruned_loss=0.05633, over 1610345.77 frames. ], batch size: 16, lr: 2.60e-03, grad_scale: 4.0 2023-02-09 02:09:13,043 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.639e+02 2.469e+02 3.110e+02 3.966e+02 1.066e+03, threshold=6.219e+02, percent-clipped=4.0 2023-02-09 02:09:13,768 INFO [train.py:901] (0/4) Epoch 29, batch 3800, loss[loss=0.1867, simple_loss=0.2883, pruned_loss=0.04253, over 8576.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2817, pruned_loss=0.05677, over 1610188.18 frames. ], batch size: 31, lr: 2.60e-03, grad_scale: 4.0 2023-02-09 02:09:24,893 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230136.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 02:09:47,387 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230168.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:09:49,391 INFO [train.py:901] (0/4) Epoch 29, batch 3850, loss[loss=0.1723, simple_loss=0.2667, pruned_loss=0.03894, over 8457.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2806, pruned_loss=0.05629, over 1610577.67 frames. ], batch size: 29, lr: 2.60e-03, grad_scale: 4.0 2023-02-09 02:09:53,108 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230176.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:10:10,924 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([5.8438, 6.0499, 5.1340, 2.6813, 5.2436, 5.6243, 5.4981, 5.4166], device='cuda:0'), covar=tensor([0.0503, 0.0333, 0.0950, 0.4047, 0.0781, 0.0760, 0.1049, 0.0571], device='cuda:0'), in_proj_covar=tensor([0.0549, 0.0464, 0.0457, 0.0567, 0.0448, 0.0470, 0.0446, 0.0415], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 02:10:11,021 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230201.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:10:12,898 WARNING [train.py:1067] (0/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-02-09 02:10:17,573 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230210.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:10:24,398 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.322e+02 2.511e+02 3.297e+02 3.973e+02 1.066e+03, threshold=6.594e+02, percent-clipped=6.0 2023-02-09 02:10:25,133 INFO [train.py:901] (0/4) Epoch 29, batch 3900, loss[loss=0.1933, simple_loss=0.2699, pruned_loss=0.05832, over 8230.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.2805, pruned_loss=0.05586, over 1614249.97 frames. ], batch size: 22, lr: 2.59e-03, grad_scale: 4.0 2023-02-09 02:10:36,235 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230235.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:10:47,182 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=3.86 vs. limit=5.0 2023-02-09 02:10:51,860 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=230256.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:10:54,125 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230259.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:11:02,671 INFO [train.py:901] (0/4) Epoch 29, batch 3950, loss[loss=0.2125, simple_loss=0.293, pruned_loss=0.06599, over 7973.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2813, pruned_loss=0.05676, over 1610478.31 frames. ], batch size: 21, lr: 2.59e-03, grad_scale: 4.0 2023-02-09 02:11:30,515 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230310.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:11:37,968 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.749e+02 2.564e+02 3.192e+02 4.107e+02 9.729e+02, threshold=6.384e+02, percent-clipped=2.0 2023-02-09 02:11:38,729 INFO [train.py:901] (0/4) Epoch 29, batch 4000, loss[loss=0.1533, simple_loss=0.2348, pruned_loss=0.03591, over 7803.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2813, pruned_loss=0.05654, over 1615244.40 frames. ], batch size: 19, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:11:49,766 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230335.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:11:59,726 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=230349.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:12:16,083 INFO [train.py:901] (0/4) Epoch 29, batch 4050, loss[loss=0.1578, simple_loss=0.243, pruned_loss=0.03629, over 6410.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2811, pruned_loss=0.05639, over 1615089.15 frames. ], batch size: 14, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:12:16,266 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230371.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:12:31,119 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230392.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 02:12:48,570 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230417.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 02:12:50,325 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.613e+02 2.354e+02 2.823e+02 3.644e+02 9.834e+02, threshold=5.645e+02, percent-clipped=2.0 2023-02-09 02:12:51,009 INFO [train.py:901] (0/4) Epoch 29, batch 4100, loss[loss=0.2335, simple_loss=0.3244, pruned_loss=0.07134, over 8406.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.28, pruned_loss=0.05548, over 1612408.25 frames. ], batch size: 49, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:13:22,445 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230464.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:13:26,979 INFO [train.py:901] (0/4) Epoch 29, batch 4150, loss[loss=0.2079, simple_loss=0.2983, pruned_loss=0.05875, over 8445.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2808, pruned_loss=0.05589, over 1611139.02 frames. ], batch size: 29, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:13:38,479 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230486.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:13:56,241 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=230512.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:14:01,802 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.657e+02 2.471e+02 3.120e+02 4.398e+02 1.014e+03, threshold=6.241e+02, percent-clipped=11.0 2023-02-09 02:14:02,507 INFO [train.py:901] (0/4) Epoch 29, batch 4200, loss[loss=0.1973, simple_loss=0.2904, pruned_loss=0.05214, over 8499.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2815, pruned_loss=0.05665, over 1614681.91 frames. ], batch size: 28, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:14:16,946 WARNING [train.py:1067] (0/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-02-09 02:14:17,803 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230543.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:14:24,154 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230551.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:14:38,390 INFO [train.py:901] (0/4) Epoch 29, batch 4250, loss[loss=0.2273, simple_loss=0.3099, pruned_loss=0.07236, over 8453.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2827, pruned_loss=0.05684, over 1619119.17 frames. ], batch size: 27, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:14:41,209 WARNING [train.py:1067] (0/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-02-09 02:15:01,588 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=230603.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:15:13,978 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.550e+02 2.532e+02 3.108e+02 3.832e+02 7.900e+02, threshold=6.217e+02, percent-clipped=4.0 2023-02-09 02:15:14,745 INFO [train.py:901] (0/4) Epoch 29, batch 4300, loss[loss=0.1876, simple_loss=0.2747, pruned_loss=0.05024, over 8454.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2822, pruned_loss=0.05691, over 1608757.92 frames. ], batch size: 25, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:15:19,507 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230627.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:15:19,559 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230627.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:15:37,561 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230652.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:15:51,740 INFO [train.py:901] (0/4) Epoch 29, batch 4350, loss[loss=0.1967, simple_loss=0.2813, pruned_loss=0.05599, over 8472.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2809, pruned_loss=0.05629, over 1607732.78 frames. ], batch size: 27, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:15:55,720 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.7986, 1.6106, 2.4034, 1.6173, 1.3628, 2.3109, 0.5958, 1.5157], device='cuda:0'), covar=tensor([0.1599, 0.1222, 0.0353, 0.1084, 0.2189, 0.0404, 0.1785, 0.1208], device='cuda:0'), in_proj_covar=tensor([0.0203, 0.0206, 0.0138, 0.0225, 0.0281, 0.0149, 0.0174, 0.0202], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 02:16:17,542 WARNING [train.py:1067] (0/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-02-09 02:16:27,839 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230718.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:16:29,092 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.602e+02 2.462e+02 3.042e+02 3.743e+02 1.027e+03, threshold=6.085e+02, percent-clipped=1.0 2023-02-09 02:16:29,367 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230720.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:16:29,857 INFO [train.py:901] (0/4) Epoch 29, batch 4400, loss[loss=0.1851, simple_loss=0.272, pruned_loss=0.04906, over 8197.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2817, pruned_loss=0.0567, over 1608827.72 frames. ], batch size: 23, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:16:47,649 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230745.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:16:57,348 WARNING [train.py:1067] (0/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-02-09 02:17:06,612 INFO [train.py:901] (0/4) Epoch 29, batch 4450, loss[loss=0.208, simple_loss=0.2836, pruned_loss=0.06623, over 7792.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.281, pruned_loss=0.05642, over 1611355.12 frames. ], batch size: 19, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:17:36,261 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230810.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:17:43,247 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.628e+02 2.373e+02 2.913e+02 3.542e+02 8.795e+02, threshold=5.826e+02, percent-clipped=1.0 2023-02-09 02:17:43,993 INFO [train.py:901] (0/4) Epoch 29, batch 4500, loss[loss=0.2012, simple_loss=0.2946, pruned_loss=0.05389, over 8362.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2811, pruned_loss=0.05676, over 1606534.47 frames. ], batch size: 24, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:17:49,324 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.8671, 1.3384, 3.4779, 1.5047, 2.5216, 3.8311, 3.9474, 3.3121], device='cuda:0'), covar=tensor([0.1298, 0.2140, 0.0336, 0.2223, 0.1034, 0.0228, 0.0514, 0.0550], device='cuda:0'), in_proj_covar=tensor([0.0311, 0.0328, 0.0296, 0.0325, 0.0328, 0.0279, 0.0447, 0.0311], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 02:17:51,408 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=230830.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:17:54,914 WARNING [train.py:1067] (0/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-02-09 02:18:19,720 INFO [train.py:901] (0/4) Epoch 29, batch 4550, loss[loss=0.2397, simple_loss=0.33, pruned_loss=0.07473, over 8568.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2821, pruned_loss=0.05745, over 1608830.36 frames. ], batch size: 31, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:18:28,427 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230883.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:18:31,629 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=230887.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:18:37,203 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=230895.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:18:46,429 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230908.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:18:46,496 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230908.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:18:55,063 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.629e+02 2.340e+02 2.935e+02 3.730e+02 9.176e+02, threshold=5.869e+02, percent-clipped=5.0 2023-02-09 02:18:55,765 INFO [train.py:901] (0/4) Epoch 29, batch 4600, loss[loss=0.1911, simple_loss=0.2873, pruned_loss=0.04744, over 8252.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2816, pruned_loss=0.05703, over 1608952.06 frames. ], batch size: 24, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:19:13,598 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230945.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:19:31,775 INFO [train.py:901] (0/4) Epoch 29, batch 4650, loss[loss=0.1691, simple_loss=0.2494, pruned_loss=0.04444, over 7431.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2823, pruned_loss=0.05725, over 1611780.75 frames. ], batch size: 17, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:19:33,916 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230974.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:19:42,745 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([5.8410, 6.0267, 5.2571, 2.2642, 5.2201, 5.6602, 5.5304, 5.4150], device='cuda:0'), covar=tensor([0.0522, 0.0335, 0.0859, 0.4343, 0.0672, 0.0671, 0.0980, 0.0587], device='cuda:0'), in_proj_covar=tensor([0.0544, 0.0460, 0.0451, 0.0563, 0.0442, 0.0466, 0.0441, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 02:19:51,180 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230999.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:19:53,303 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231002.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:19:59,428 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231010.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:20:06,388 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.812e+02 2.522e+02 3.157e+02 3.845e+02 7.559e+02, threshold=6.314e+02, percent-clipped=7.0 2023-02-09 02:20:07,123 INFO [train.py:901] (0/4) Epoch 29, batch 4700, loss[loss=0.1945, simple_loss=0.2754, pruned_loss=0.05687, over 8199.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2821, pruned_loss=0.05697, over 1609069.62 frames. ], batch size: 23, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:20:43,873 INFO [train.py:901] (0/4) Epoch 29, batch 4750, loss[loss=0.1709, simple_loss=0.2528, pruned_loss=0.04451, over 7438.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2818, pruned_loss=0.05719, over 1603534.91 frames. ], batch size: 17, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:21:00,622 WARNING [train.py:1067] (0/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-02-09 02:21:02,795 WARNING [train.py:1067] (0/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-02-09 02:21:09,910 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 2023-02-09 02:21:18,722 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.727e+02 2.440e+02 2.924e+02 3.615e+02 6.392e+02, threshold=5.847e+02, percent-clipped=1.0 2023-02-09 02:21:19,439 INFO [train.py:901] (0/4) Epoch 29, batch 4800, loss[loss=0.172, simple_loss=0.2689, pruned_loss=0.03753, over 8353.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2807, pruned_loss=0.05646, over 1608578.33 frames. ], batch size: 24, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:21:43,701 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=231154.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:21:55,946 INFO [train.py:901] (0/4) Epoch 29, batch 4850, loss[loss=0.2115, simple_loss=0.2924, pruned_loss=0.06528, over 8124.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2803, pruned_loss=0.05617, over 1609018.06 frames. ], batch size: 22, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:21:55,955 WARNING [train.py:1067] (0/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-02-09 02:22:17,890 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231201.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:22:31,032 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.570e+02 2.490e+02 3.133e+02 4.186e+02 8.287e+02, threshold=6.266e+02, percent-clipped=6.0 2023-02-09 02:22:31,779 INFO [train.py:901] (0/4) Epoch 29, batch 4900, loss[loss=0.164, simple_loss=0.2505, pruned_loss=0.03871, over 8073.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2803, pruned_loss=0.05621, over 1608271.40 frames. ], batch size: 21, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:22:35,629 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231226.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:22:54,370 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=231252.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:22:59,240 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231258.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:23:04,620 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231266.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:23:06,612 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=231269.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:23:06,651 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231269.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:23:07,930 INFO [train.py:901] (0/4) Epoch 29, batch 4950, loss[loss=0.206, simple_loss=0.3014, pruned_loss=0.05535, over 8105.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2813, pruned_loss=0.05647, over 1608515.98 frames. ], batch size: 23, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:23:17,277 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231283.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:23:22,765 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231291.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:23:43,161 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.636e+02 2.476e+02 2.910e+02 3.581e+02 7.956e+02, threshold=5.820e+02, percent-clipped=2.0 2023-02-09 02:23:43,855 INFO [train.py:901] (0/4) Epoch 29, batch 5000, loss[loss=0.1778, simple_loss=0.2606, pruned_loss=0.04747, over 7555.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2813, pruned_loss=0.05638, over 1612126.84 frames. ], batch size: 18, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:23:45,426 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.4399, 1.8375, 4.5518, 2.3178, 2.6304, 5.2358, 5.2794, 4.5416], device='cuda:0'), covar=tensor([0.1094, 0.1770, 0.0226, 0.1735, 0.1120, 0.0149, 0.0321, 0.0542], device='cuda:0'), in_proj_covar=tensor([0.0313, 0.0330, 0.0299, 0.0327, 0.0328, 0.0281, 0.0450, 0.0312], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 02:24:17,067 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231367.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:24:19,695 INFO [train.py:901] (0/4) Epoch 29, batch 5050, loss[loss=0.1998, simple_loss=0.2902, pruned_loss=0.05468, over 8517.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2821, pruned_loss=0.05682, over 1610772.05 frames. ], batch size: 26, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:24:37,813 WARNING [train.py:1067] (0/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-02-09 02:24:55,717 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.552e+02 2.315e+02 2.860e+02 3.491e+02 8.708e+02, threshold=5.721e+02, percent-clipped=8.0 2023-02-09 02:24:56,378 INFO [train.py:901] (0/4) Epoch 29, batch 5100, loss[loss=0.2494, simple_loss=0.3194, pruned_loss=0.08976, over 8461.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2818, pruned_loss=0.05691, over 1607182.84 frames. ], batch size: 29, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:25:32,302 INFO [train.py:901] (0/4) Epoch 29, batch 5150, loss[loss=0.1885, simple_loss=0.2859, pruned_loss=0.0455, over 8283.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2816, pruned_loss=0.05695, over 1608446.74 frames. ], batch size: 23, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:25:52,008 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-02-09 02:26:07,354 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 2023-02-09 02:26:08,165 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.585e+02 2.313e+02 2.816e+02 3.592e+02 7.666e+02, threshold=5.632e+02, percent-clipped=6.0 2023-02-09 02:26:08,945 INFO [train.py:901] (0/4) Epoch 29, batch 5200, loss[loss=0.1736, simple_loss=0.2587, pruned_loss=0.0442, over 8077.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2805, pruned_loss=0.05632, over 1608953.10 frames. ], batch size: 21, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:26:12,110 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231525.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:26:30,191 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231550.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:26:39,129 WARNING [train.py:1067] (0/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-02-09 02:26:44,750 INFO [train.py:901] (0/4) Epoch 29, batch 5250, loss[loss=0.2057, simple_loss=0.2896, pruned_loss=0.06089, over 7918.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2802, pruned_loss=0.0565, over 1610060.70 frames. ], batch size: 20, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:27:15,801 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=231613.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:27:20,504 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.520e+02 2.485e+02 2.947e+02 3.559e+02 7.815e+02, threshold=5.893e+02, percent-clipped=2.0 2023-02-09 02:27:21,231 INFO [train.py:901] (0/4) Epoch 29, batch 5300, loss[loss=0.1826, simple_loss=0.2701, pruned_loss=0.04757, over 8187.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2808, pruned_loss=0.05641, over 1612536.56 frames. ], batch size: 23, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:27:22,862 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231623.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:27:41,027 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231648.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:27:46,378 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([5.9348, 1.5509, 6.0300, 2.2123, 5.4737, 5.0706, 5.5846, 5.4680], device='cuda:0'), covar=tensor([0.0426, 0.5158, 0.0371, 0.3952, 0.0939, 0.0917, 0.0450, 0.0493], device='cuda:0'), in_proj_covar=tensor([0.0682, 0.0660, 0.0742, 0.0653, 0.0739, 0.0632, 0.0640, 0.0717], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 02:27:47,757 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.0527, 2.1687, 1.8955, 2.9144, 1.2611, 1.6214, 2.0182, 2.1699], device='cuda:0'), covar=tensor([0.0768, 0.0842, 0.0849, 0.0382, 0.1107, 0.1417, 0.0920, 0.0725], device='cuda:0'), in_proj_covar=tensor([0.0230, 0.0195, 0.0243, 0.0213, 0.0202, 0.0245, 0.0250, 0.0204], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], device='cuda:0') 2023-02-09 02:27:53,033 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=231665.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:27:54,370 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([4.1555, 1.4116, 4.3720, 1.6290, 3.8591, 3.6706, 3.9315, 3.8396], device='cuda:0'), covar=tensor([0.0718, 0.4746, 0.0576, 0.4121, 0.1126, 0.0956, 0.0689, 0.0750], device='cuda:0'), in_proj_covar=tensor([0.0682, 0.0660, 0.0742, 0.0653, 0.0739, 0.0632, 0.0640, 0.0717], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 02:27:57,017 INFO [train.py:901] (0/4) Epoch 29, batch 5350, loss[loss=0.1959, simple_loss=0.2908, pruned_loss=0.05048, over 8492.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2814, pruned_loss=0.05665, over 1614671.44 frames. ], batch size: 29, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:28:32,683 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.493e+02 2.535e+02 3.138e+02 3.956e+02 6.651e+02, threshold=6.276e+02, percent-clipped=5.0 2023-02-09 02:28:33,426 INFO [train.py:901] (0/4) Epoch 29, batch 5400, loss[loss=0.1893, simple_loss=0.2788, pruned_loss=0.04987, over 8038.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2826, pruned_loss=0.05733, over 1617545.11 frames. ], batch size: 22, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:28:38,412 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231728.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:29:09,678 INFO [train.py:901] (0/4) Epoch 29, batch 5450, loss[loss=0.2117, simple_loss=0.3007, pruned_loss=0.06132, over 8426.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2816, pruned_loss=0.05723, over 1614468.91 frames. ], batch size: 49, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:29:24,014 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.1350, 3.5783, 2.4751, 3.0847, 2.8372, 2.2351, 2.8074, 3.1271], device='cuda:0'), covar=tensor([0.2013, 0.0427, 0.1149, 0.0718, 0.0855, 0.1469, 0.1215, 0.1208], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0244, 0.0343, 0.0315, 0.0302, 0.0348, 0.0351, 0.0320], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 02:29:29,509 WARNING [train.py:1067] (0/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-02-09 02:29:43,744 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.7857, 2.0096, 2.1026, 1.4492, 2.1911, 1.5023, 0.7864, 2.0391], device='cuda:0'), covar=tensor([0.0733, 0.0476, 0.0359, 0.0705, 0.0572, 0.0904, 0.1031, 0.0362], device='cuda:0'), in_proj_covar=tensor([0.0476, 0.0417, 0.0371, 0.0465, 0.0400, 0.0556, 0.0408, 0.0445], device='cuda:0'), out_proj_covar=tensor([1.2593e-04, 1.0787e-04, 9.6390e-05, 1.2157e-04, 1.0474e-04, 1.5474e-04, 1.0873e-04, 1.1633e-04], device='cuda:0') 2023-02-09 02:29:44,874 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.390e+02 2.416e+02 2.826e+02 3.521e+02 6.915e+02, threshold=5.653e+02, percent-clipped=1.0 2023-02-09 02:29:45,630 INFO [train.py:901] (0/4) Epoch 29, batch 5500, loss[loss=0.189, simple_loss=0.2784, pruned_loss=0.04981, over 8485.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2815, pruned_loss=0.05688, over 1617484.67 frames. ], batch size: 27, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:29:48,666 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.6459, 1.6687, 2.0279, 1.6645, 0.9578, 1.7234, 2.1791, 2.0699], device='cuda:0'), covar=tensor([0.0485, 0.1241, 0.1621, 0.1457, 0.0587, 0.1405, 0.0649, 0.0599], device='cuda:0'), in_proj_covar=tensor([0.0100, 0.0155, 0.0191, 0.0163, 0.0102, 0.0165, 0.0114, 0.0148], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0007, 0.0008], device='cuda:0') 2023-02-09 02:30:21,379 INFO [train.py:901] (0/4) Epoch 29, batch 5550, loss[loss=0.2045, simple_loss=0.2816, pruned_loss=0.06371, over 8091.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2806, pruned_loss=0.0566, over 1614144.66 frames. ], batch size: 21, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:30:49,882 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 2023-02-09 02:30:56,573 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.629e+02 2.336e+02 2.917e+02 3.506e+02 1.057e+03, threshold=5.834e+02, percent-clipped=5.0 2023-02-09 02:30:57,331 INFO [train.py:901] (0/4) Epoch 29, batch 5600, loss[loss=0.1897, simple_loss=0.2814, pruned_loss=0.04903, over 8192.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2811, pruned_loss=0.05681, over 1611912.04 frames. ], batch size: 23, lr: 2.59e-03, grad_scale: 8.0 2023-02-09 02:31:24,827 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.45 vs. limit=2.0 2023-02-09 02:31:34,506 INFO [train.py:901] (0/4) Epoch 29, batch 5650, loss[loss=0.2293, simple_loss=0.3146, pruned_loss=0.07205, over 8503.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2814, pruned_loss=0.05726, over 1611143.00 frames. ], batch size: 28, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:31:38,829 WARNING [train.py:1067] (0/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-02-09 02:31:43,813 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231984.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:31:54,968 INFO [checkpoint.py:75] (0/4) Saving checkpoint to pruned_transducer_stateless7_streaming/exp/v1/checkpoint-232000.pt 2023-02-09 02:32:02,829 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=232009.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:32:02,968 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=232009.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:32:10,314 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.554e+02 2.351e+02 2.719e+02 3.536e+02 6.635e+02, threshold=5.438e+02, percent-clipped=1.0 2023-02-09 02:32:11,022 INFO [train.py:901] (0/4) Epoch 29, batch 5700, loss[loss=0.204, simple_loss=0.2847, pruned_loss=0.06161, over 8345.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2821, pruned_loss=0.05745, over 1613788.31 frames. ], batch size: 26, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:32:45,200 WARNING [train.py:1067] (0/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-02-09 02:32:47,267 INFO [train.py:901] (0/4) Epoch 29, batch 5750, loss[loss=0.2152, simple_loss=0.294, pruned_loss=0.06823, over 8039.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2819, pruned_loss=0.05727, over 1614977.14 frames. ], batch size: 22, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:32:54,536 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.6225, 1.9772, 2.0584, 1.2776, 2.1144, 1.4829, 0.7621, 1.8773], device='cuda:0'), covar=tensor([0.0979, 0.0453, 0.0405, 0.0936, 0.0609, 0.1149, 0.1185, 0.0453], device='cuda:0'), in_proj_covar=tensor([0.0478, 0.0418, 0.0372, 0.0466, 0.0401, 0.0559, 0.0408, 0.0446], device='cuda:0'), out_proj_covar=tensor([1.2665e-04, 1.0797e-04, 9.6710e-05, 1.2186e-04, 1.0495e-04, 1.5560e-04, 1.0867e-04, 1.1668e-04], device='cuda:0') 2023-02-09 02:33:07,936 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.7668, 1.6595, 2.6190, 2.0065, 2.3291, 1.8366, 1.5825, 1.1863], device='cuda:0'), covar=tensor([0.7853, 0.6698, 0.2371, 0.4515, 0.3585, 0.4947, 0.3152, 0.6126], device='cuda:0'), in_proj_covar=tensor([0.0971, 0.1035, 0.0840, 0.1003, 0.1028, 0.0939, 0.0777, 0.0856], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 02:33:23,816 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.487e+02 2.417e+02 3.013e+02 3.730e+02 1.097e+03, threshold=6.026e+02, percent-clipped=6.0 2023-02-09 02:33:24,549 INFO [train.py:901] (0/4) Epoch 29, batch 5800, loss[loss=0.2019, simple_loss=0.2659, pruned_loss=0.06893, over 7667.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2806, pruned_loss=0.05636, over 1612646.63 frames. ], batch size: 19, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:33:26,837 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=232124.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:33:38,337 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.0390, 1.5291, 3.4891, 1.5293, 2.5792, 3.8269, 3.9890, 3.1680], device='cuda:0'), covar=tensor([0.1215, 0.1931, 0.0395, 0.2192, 0.1059, 0.0280, 0.0616, 0.0656], device='cuda:0'), in_proj_covar=tensor([0.0310, 0.0328, 0.0295, 0.0325, 0.0326, 0.0279, 0.0446, 0.0310], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 02:33:46,045 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.7572, 2.0132, 2.1030, 1.3522, 2.2454, 1.5057, 0.7929, 1.9807], device='cuda:0'), covar=tensor([0.0850, 0.0451, 0.0392, 0.0822, 0.0466, 0.1088, 0.1124, 0.0451], device='cuda:0'), in_proj_covar=tensor([0.0477, 0.0417, 0.0371, 0.0465, 0.0400, 0.0558, 0.0407, 0.0445], device='cuda:0'), out_proj_covar=tensor([1.2632e-04, 1.0779e-04, 9.6523e-05, 1.2158e-04, 1.0470e-04, 1.5528e-04, 1.0856e-04, 1.1647e-04], device='cuda:0') 2023-02-09 02:33:59,553 INFO [train.py:901] (0/4) Epoch 29, batch 5850, loss[loss=0.1926, simple_loss=0.2895, pruned_loss=0.04791, over 8108.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2805, pruned_loss=0.05595, over 1613049.27 frames. ], batch size: 23, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:34:31,421 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.1689, 1.5936, 1.8187, 1.5007, 1.0339, 1.5623, 1.8069, 1.5332], device='cuda:0'), covar=tensor([0.0516, 0.1262, 0.1584, 0.1449, 0.0584, 0.1472, 0.0709, 0.0698], device='cuda:0'), in_proj_covar=tensor([0.0100, 0.0155, 0.0191, 0.0163, 0.0102, 0.0165, 0.0115, 0.0148], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0008, 0.0008], device='cuda:0') 2023-02-09 02:34:34,651 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.378e+02 2.421e+02 2.869e+02 3.503e+02 7.290e+02, threshold=5.737e+02, percent-clipped=3.0 2023-02-09 02:34:35,349 INFO [train.py:901] (0/4) Epoch 29, batch 5900, loss[loss=0.1741, simple_loss=0.2639, pruned_loss=0.04217, over 8177.00 frames. ], tot_loss[loss=0.196, simple_loss=0.2804, pruned_loss=0.05583, over 1611903.02 frames. ], batch size: 23, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:34:44,583 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([5.8963, 1.3841, 6.0628, 2.0948, 5.4078, 5.1272, 5.5627, 5.4423], device='cuda:0'), covar=tensor([0.0552, 0.5445, 0.0414, 0.4368, 0.1186, 0.0918, 0.0576, 0.0625], device='cuda:0'), in_proj_covar=tensor([0.0688, 0.0667, 0.0748, 0.0659, 0.0747, 0.0637, 0.0643, 0.0724], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 02:35:06,599 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=232265.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:35:10,669 INFO [train.py:901] (0/4) Epoch 29, batch 5950, loss[loss=0.2252, simple_loss=0.3077, pruned_loss=0.07138, over 8105.00 frames. ], tot_loss[loss=0.1948, simple_loss=0.2791, pruned_loss=0.0553, over 1611126.03 frames. ], batch size: 23, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:35:26,282 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-02-09 02:35:46,611 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.609e+02 2.503e+02 2.837e+02 3.700e+02 9.228e+02, threshold=5.675e+02, percent-clipped=4.0 2023-02-09 02:35:47,368 INFO [train.py:901] (0/4) Epoch 29, batch 6000, loss[loss=0.191, simple_loss=0.2784, pruned_loss=0.05175, over 8461.00 frames. ], tot_loss[loss=0.1948, simple_loss=0.2792, pruned_loss=0.05519, over 1611239.32 frames. ], batch size: 27, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:35:47,369 INFO [train.py:926] (0/4) Computing validation loss 2023-02-09 02:36:01,206 INFO [train.py:935] (0/4) Epoch 29, validation: loss=0.1708, simple_loss=0.2701, pruned_loss=0.03577, over 944034.00 frames. 2023-02-09 02:36:01,207 INFO [train.py:936] (0/4) Maximum memory allocated so far is 6641MB 2023-02-09 02:36:11,926 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([5.8985, 1.5618, 5.9886, 2.2296, 5.4040, 4.9144, 5.4415, 5.3674], device='cuda:0'), covar=tensor([0.0437, 0.5182, 0.0448, 0.3955, 0.1056, 0.1077, 0.0529, 0.0551], device='cuda:0'), in_proj_covar=tensor([0.0687, 0.0666, 0.0747, 0.0658, 0.0745, 0.0637, 0.0641, 0.0724], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 02:36:21,626 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.9643, 1.5858, 2.8302, 1.4927, 2.2298, 3.0208, 3.1826, 2.6053], device='cuda:0'), covar=tensor([0.1058, 0.1584, 0.0359, 0.2044, 0.0930, 0.0300, 0.0611, 0.0543], device='cuda:0'), in_proj_covar=tensor([0.0311, 0.0328, 0.0297, 0.0326, 0.0326, 0.0279, 0.0447, 0.0311], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 02:36:37,560 INFO [train.py:901] (0/4) Epoch 29, batch 6050, loss[loss=0.1926, simple_loss=0.284, pruned_loss=0.05064, over 8664.00 frames. ], tot_loss[loss=0.1949, simple_loss=0.2792, pruned_loss=0.05528, over 1614339.02 frames. ], batch size: 34, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:36:44,021 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=232380.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:36:57,858 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=232399.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:37:02,410 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=232405.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:37:12,732 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.752e+02 2.608e+02 3.073e+02 4.031e+02 7.869e+02, threshold=6.145e+02, percent-clipped=3.0 2023-02-09 02:37:13,453 INFO [train.py:901] (0/4) Epoch 29, batch 6100, loss[loss=0.1855, simple_loss=0.2781, pruned_loss=0.04646, over 8185.00 frames. ], tot_loss[loss=0.1946, simple_loss=0.2791, pruned_loss=0.05502, over 1612608.25 frames. ], batch size: 23, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:37:27,130 WARNING [train.py:1067] (0/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-02-09 02:37:50,003 INFO [train.py:901] (0/4) Epoch 29, batch 6150, loss[loss=0.1788, simple_loss=0.2769, pruned_loss=0.04032, over 8468.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.2802, pruned_loss=0.05561, over 1617715.54 frames. ], batch size: 27, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:38:00,015 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.8321, 1.7823, 2.3212, 1.4778, 1.4698, 2.2391, 0.5455, 1.4776], device='cuda:0'), covar=tensor([0.1389, 0.1026, 0.0289, 0.0891, 0.1963, 0.0496, 0.1514, 0.1068], device='cuda:0'), in_proj_covar=tensor([0.0203, 0.0205, 0.0137, 0.0224, 0.0279, 0.0149, 0.0172, 0.0201], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 02:38:06,448 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=232494.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:38:25,294 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.683e+02 2.548e+02 2.901e+02 3.636e+02 6.365e+02, threshold=5.801e+02, percent-clipped=1.0 2023-02-09 02:38:25,874 INFO [train.py:901] (0/4) Epoch 29, batch 6200, loss[loss=0.2931, simple_loss=0.3529, pruned_loss=0.1167, over 8107.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2811, pruned_loss=0.05603, over 1613681.94 frames. ], batch size: 23, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:38:36,674 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=232536.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:38:53,542 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.6834, 2.4804, 3.1373, 2.6399, 3.0118, 2.6251, 2.6283, 2.3612], device='cuda:0'), covar=tensor([0.4261, 0.4103, 0.1884, 0.3291, 0.2402, 0.2799, 0.1634, 0.4422], device='cuda:0'), in_proj_covar=tensor([0.0974, 0.1036, 0.0843, 0.1005, 0.1030, 0.0939, 0.0779, 0.0857], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 02:39:02,789 INFO [train.py:901] (0/4) Epoch 29, batch 6250, loss[loss=0.2336, simple_loss=0.3014, pruned_loss=0.08291, over 6832.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.2802, pruned_loss=0.05573, over 1611208.02 frames. ], batch size: 71, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:39:07,150 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.1018, 1.5152, 1.7611, 1.4383, 1.0908, 1.4858, 1.9594, 1.9014], device='cuda:0'), covar=tensor([0.0553, 0.1243, 0.1600, 0.1465, 0.0590, 0.1496, 0.0700, 0.0588], device='cuda:0'), in_proj_covar=tensor([0.0100, 0.0155, 0.0190, 0.0162, 0.0102, 0.0164, 0.0114, 0.0148], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0007, 0.0008], device='cuda:0') 2023-02-09 02:39:16,237 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.4373, 1.6263, 4.5153, 2.0979, 2.5329, 5.1199, 5.2045, 4.3642], device='cuda:0'), covar=tensor([0.1184, 0.2015, 0.0260, 0.1948, 0.1136, 0.0171, 0.0558, 0.0587], device='cuda:0'), in_proj_covar=tensor([0.0310, 0.0327, 0.0295, 0.0326, 0.0325, 0.0278, 0.0445, 0.0309], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 02:39:23,936 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.7354, 1.3364, 2.8664, 1.4242, 2.2266, 3.0868, 3.2614, 2.6324], device='cuda:0'), covar=tensor([0.1256, 0.1814, 0.0401, 0.2243, 0.0933, 0.0321, 0.0748, 0.0587], device='cuda:0'), in_proj_covar=tensor([0.0310, 0.0326, 0.0295, 0.0326, 0.0325, 0.0278, 0.0445, 0.0309], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 02:39:30,142 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=232609.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:39:37,831 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.584e+02 2.332e+02 2.762e+02 3.491e+02 8.673e+02, threshold=5.523e+02, percent-clipped=4.0 2023-02-09 02:39:39,195 INFO [train.py:901] (0/4) Epoch 29, batch 6300, loss[loss=0.2006, simple_loss=0.2929, pruned_loss=0.05414, over 8460.00 frames. ], tot_loss[loss=0.196, simple_loss=0.2805, pruned_loss=0.05572, over 1615247.96 frames. ], batch size: 27, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:39:56,315 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.9437, 1.7693, 2.2406, 1.8796, 2.2462, 2.0471, 1.9275, 1.1763], device='cuda:0'), covar=tensor([0.6623, 0.5491, 0.2343, 0.4522, 0.2864, 0.3611, 0.2213, 0.5940], device='cuda:0'), in_proj_covar=tensor([0.0974, 0.1036, 0.0844, 0.1006, 0.1030, 0.0940, 0.0780, 0.0858], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 02:40:12,366 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.6234, 1.9754, 2.8556, 1.5141, 2.0966, 2.0544, 1.7364, 2.2115], device='cuda:0'), covar=tensor([0.2121, 0.2827, 0.1104, 0.4944, 0.2318, 0.3426, 0.2661, 0.2441], device='cuda:0'), in_proj_covar=tensor([0.0544, 0.0641, 0.0565, 0.0674, 0.0667, 0.0614, 0.0569, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 02:40:14,953 INFO [train.py:901] (0/4) Epoch 29, batch 6350, loss[loss=0.185, simple_loss=0.2755, pruned_loss=0.04729, over 8581.00 frames. ], tot_loss[loss=0.1956, simple_loss=0.2798, pruned_loss=0.05568, over 1614151.32 frames. ], batch size: 49, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:40:50,679 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.474e+02 2.465e+02 3.018e+02 3.778e+02 1.284e+03, threshold=6.036e+02, percent-clipped=3.0 2023-02-09 02:40:51,328 INFO [train.py:901] (0/4) Epoch 29, batch 6400, loss[loss=0.2403, simple_loss=0.3069, pruned_loss=0.08691, over 6730.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.2801, pruned_loss=0.05577, over 1611370.33 frames. ], batch size: 72, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:40:53,617 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=232724.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:41:08,038 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=232743.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:41:28,304 INFO [train.py:901] (0/4) Epoch 29, batch 6450, loss[loss=0.1535, simple_loss=0.2384, pruned_loss=0.03429, over 7448.00 frames. ], tot_loss[loss=0.196, simple_loss=0.2805, pruned_loss=0.05575, over 1615133.99 frames. ], batch size: 17, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:42:05,017 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.603e+02 2.358e+02 3.084e+02 4.266e+02 9.590e+02, threshold=6.168e+02, percent-clipped=5.0 2023-02-09 02:42:05,757 INFO [train.py:901] (0/4) Epoch 29, batch 6500, loss[loss=0.1683, simple_loss=0.247, pruned_loss=0.04475, over 7662.00 frames. ], tot_loss[loss=0.195, simple_loss=0.2794, pruned_loss=0.05528, over 1611046.93 frames. ], batch size: 19, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:42:17,434 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=232838.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:42:24,002 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.76 vs. limit=2.0 2023-02-09 02:42:24,518 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.6978, 2.5387, 1.7848, 2.3300, 2.3528, 1.6027, 2.2563, 2.1677], device='cuda:0'), covar=tensor([0.1570, 0.0475, 0.1430, 0.0753, 0.0696, 0.1683, 0.1039, 0.1103], device='cuda:0'), in_proj_covar=tensor([0.0358, 0.0246, 0.0347, 0.0316, 0.0303, 0.0351, 0.0352, 0.0321], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 02:42:31,256 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=232858.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:42:31,462 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-02-09 02:42:40,789 INFO [train.py:901] (0/4) Epoch 29, batch 6550, loss[loss=0.2322, simple_loss=0.3147, pruned_loss=0.07484, over 8550.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.2796, pruned_loss=0.05572, over 1614820.02 frames. ], batch size: 39, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:42:47,141 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=232880.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:42:47,799 WARNING [train.py:1067] (0/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-02-09 02:43:06,958 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.1684, 3.4627, 2.2495, 2.9697, 2.8099, 2.0810, 2.7673, 3.0802], device='cuda:0'), covar=tensor([0.1832, 0.0449, 0.1261, 0.0771, 0.0780, 0.1577, 0.1105, 0.1121], device='cuda:0'), in_proj_covar=tensor([0.0359, 0.0247, 0.0348, 0.0317, 0.0304, 0.0352, 0.0353, 0.0322], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 02:43:08,169 WARNING [train.py:1067] (0/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-02-09 02:43:16,623 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.606e+02 2.398e+02 2.846e+02 3.696e+02 7.042e+02, threshold=5.692e+02, percent-clipped=2.0 2023-02-09 02:43:17,344 INFO [train.py:901] (0/4) Epoch 29, batch 6600, loss[loss=0.2123, simple_loss=0.3004, pruned_loss=0.06215, over 8513.00 frames. ], tot_loss[loss=0.195, simple_loss=0.2794, pruned_loss=0.05533, over 1616344.77 frames. ], batch size: 26, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:43:40,641 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=232953.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:43:52,846 INFO [train.py:901] (0/4) Epoch 29, batch 6650, loss[loss=0.1813, simple_loss=0.2672, pruned_loss=0.04769, over 7236.00 frames. ], tot_loss[loss=0.1946, simple_loss=0.279, pruned_loss=0.0551, over 1612117.80 frames. ], batch size: 16, lr: 2.58e-03, grad_scale: 16.0 2023-02-09 02:43:59,988 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=232980.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:44:10,273 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=232995.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:44:17,502 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233005.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:44:19,561 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233008.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:44:28,483 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.750e+02 2.388e+02 2.820e+02 3.481e+02 6.998e+02, threshold=5.640e+02, percent-clipped=2.0 2023-02-09 02:44:28,503 INFO [train.py:901] (0/4) Epoch 29, batch 6700, loss[loss=0.2007, simple_loss=0.2792, pruned_loss=0.06109, over 7974.00 frames. ], tot_loss[loss=0.195, simple_loss=0.2793, pruned_loss=0.05531, over 1612187.66 frames. ], batch size: 21, lr: 2.58e-03, grad_scale: 8.0 2023-02-09 02:44:40,632 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233036.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:44:55,145 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.8108, 1.8096, 2.8235, 2.1158, 2.5325, 1.8888, 1.6550, 1.4097], device='cuda:0'), covar=tensor([0.8017, 0.6309, 0.2340, 0.4629, 0.3372, 0.4954, 0.3270, 0.6080], device='cuda:0'), in_proj_covar=tensor([0.0971, 0.1034, 0.0842, 0.1003, 0.1027, 0.0939, 0.0777, 0.0856], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 02:45:06,286 INFO [train.py:901] (0/4) Epoch 29, batch 6750, loss[loss=0.1714, simple_loss=0.2617, pruned_loss=0.04053, over 7659.00 frames. ], tot_loss[loss=0.1951, simple_loss=0.2793, pruned_loss=0.05551, over 1611923.77 frames. ], batch size: 19, lr: 2.58e-03, grad_scale: 8.0 2023-02-09 02:45:30,771 WARNING [train.py:1067] (0/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-02-09 02:45:38,442 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233114.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:45:43,345 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.753e+02 2.536e+02 3.042e+02 3.988e+02 8.675e+02, threshold=6.084e+02, percent-clipped=8.0 2023-02-09 02:45:43,372 INFO [train.py:901] (0/4) Epoch 29, batch 6800, loss[loss=0.2106, simple_loss=0.29, pruned_loss=0.0656, over 8657.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.2803, pruned_loss=0.05599, over 1612065.93 frames. ], batch size: 34, lr: 2.58e-03, grad_scale: 8.0 2023-02-09 02:45:56,017 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233139.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:46:18,739 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.1090, 2.2280, 1.8983, 2.7773, 1.5059, 1.7463, 2.1091, 2.3079], device='cuda:0'), covar=tensor([0.0710, 0.0769, 0.0841, 0.0456, 0.1009, 0.1261, 0.0830, 0.0745], device='cuda:0'), in_proj_covar=tensor([0.0233, 0.0196, 0.0246, 0.0215, 0.0203, 0.0248, 0.0250, 0.0206], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], device='cuda:0') 2023-02-09 02:46:19,939 INFO [train.py:901] (0/4) Epoch 29, batch 6850, loss[loss=0.161, simple_loss=0.2431, pruned_loss=0.03949, over 7431.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.281, pruned_loss=0.05631, over 1611796.98 frames. ], batch size: 17, lr: 2.58e-03, grad_scale: 8.0 2023-02-09 02:46:22,005 WARNING [train.py:1067] (0/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-02-09 02:46:34,886 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.8695, 1.3047, 3.1834, 1.5968, 2.3883, 3.4260, 3.5257, 2.9721], device='cuda:0'), covar=tensor([0.1117, 0.1905, 0.0306, 0.1954, 0.0949, 0.0253, 0.0597, 0.0499], device='cuda:0'), in_proj_covar=tensor([0.0311, 0.0328, 0.0297, 0.0327, 0.0328, 0.0280, 0.0449, 0.0311], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 02:46:45,157 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.6611, 2.5672, 1.7731, 2.3940, 2.1715, 1.5173, 2.1325, 2.2996], device='cuda:0'), covar=tensor([0.1594, 0.0432, 0.1457, 0.0701, 0.0849, 0.1839, 0.1067, 0.0936], device='cuda:0'), in_proj_covar=tensor([0.0358, 0.0246, 0.0346, 0.0315, 0.0302, 0.0350, 0.0351, 0.0321], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 02:46:46,494 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233209.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:46:48,583 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.2040, 1.9926, 2.4490, 2.0699, 2.4608, 2.3378, 2.1502, 1.3418], device='cuda:0'), covar=tensor([0.5866, 0.5315, 0.2338, 0.4053, 0.2614, 0.3197, 0.1959, 0.5776], device='cuda:0'), in_proj_covar=tensor([0.0975, 0.1037, 0.0844, 0.1010, 0.1030, 0.0943, 0.0780, 0.0859], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 02:46:55,106 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.759e+02 2.470e+02 3.068e+02 3.817e+02 7.038e+02, threshold=6.136e+02, percent-clipped=5.0 2023-02-09 02:46:55,128 INFO [train.py:901] (0/4) Epoch 29, batch 6900, loss[loss=0.1869, simple_loss=0.2802, pruned_loss=0.04678, over 8473.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2805, pruned_loss=0.05616, over 1608805.96 frames. ], batch size: 25, lr: 2.58e-03, grad_scale: 8.0 2023-02-09 02:47:00,415 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.2685, 2.1874, 2.6720, 2.2348, 2.7323, 2.4309, 2.1829, 1.5836], device='cuda:0'), covar=tensor([0.5902, 0.5219, 0.2310, 0.4312, 0.2792, 0.3090, 0.1997, 0.5656], device='cuda:0'), in_proj_covar=tensor([0.0974, 0.1037, 0.0844, 0.1009, 0.1030, 0.0943, 0.0779, 0.0858], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 02:47:04,474 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233234.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:47:16,287 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233251.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:47:31,602 INFO [train.py:901] (0/4) Epoch 29, batch 6950, loss[loss=0.2078, simple_loss=0.2987, pruned_loss=0.05846, over 8446.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2808, pruned_loss=0.05654, over 1609964.06 frames. ], batch size: 27, lr: 2.58e-03, grad_scale: 8.0 2023-02-09 02:47:33,688 WARNING [train.py:1067] (0/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-02-09 02:47:35,282 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233276.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:47:35,910 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233277.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:48:07,355 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.648e+02 2.388e+02 2.860e+02 3.725e+02 6.106e+02, threshold=5.720e+02, percent-clipped=0.0 2023-02-09 02:48:07,382 INFO [train.py:901] (0/4) Epoch 29, batch 7000, loss[loss=0.1918, simple_loss=0.2676, pruned_loss=0.058, over 8234.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.281, pruned_loss=0.05687, over 1611101.31 frames. ], batch size: 22, lr: 2.58e-03, grad_scale: 8.0 2023-02-09 02:48:30,364 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=233352.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:48:44,316 INFO [train.py:901] (0/4) Epoch 29, batch 7050, loss[loss=0.2151, simple_loss=0.289, pruned_loss=0.07055, over 8128.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2807, pruned_loss=0.05651, over 1610248.08 frames. ], batch size: 22, lr: 2.58e-03, grad_scale: 8.0 2023-02-09 02:48:51,048 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=233380.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:49:21,590 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.6380, 2.1303, 3.2595, 1.4988, 2.5805, 2.0586, 1.7797, 2.5283], device='cuda:0'), covar=tensor([0.2020, 0.2825, 0.0941, 0.5056, 0.1923, 0.3437, 0.2642, 0.2416], device='cuda:0'), in_proj_covar=tensor([0.0543, 0.0641, 0.0566, 0.0676, 0.0664, 0.0615, 0.0568, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 02:49:22,018 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.696e+02 2.400e+02 3.088e+02 3.796e+02 6.683e+02, threshold=6.176e+02, percent-clipped=2.0 2023-02-09 02:49:22,039 INFO [train.py:901] (0/4) Epoch 29, batch 7100, loss[loss=0.1762, simple_loss=0.268, pruned_loss=0.04224, over 8367.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2816, pruned_loss=0.05681, over 1611682.62 frames. ], batch size: 24, lr: 2.58e-03, grad_scale: 8.0 2023-02-09 02:49:54,761 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233467.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:49:57,422 INFO [train.py:901] (0/4) Epoch 29, batch 7150, loss[loss=0.1942, simple_loss=0.2743, pruned_loss=0.05704, over 8084.00 frames. ], tot_loss[loss=0.198, simple_loss=0.282, pruned_loss=0.05698, over 1611316.74 frames. ], batch size: 21, lr: 2.58e-03, grad_scale: 8.0 2023-02-09 02:50:09,678 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.9558, 2.0340, 1.7538, 2.5798, 1.2078, 1.5703, 1.8853, 2.0208], device='cuda:0'), covar=tensor([0.0662, 0.0787, 0.0877, 0.0377, 0.1039, 0.1229, 0.0817, 0.0790], device='cuda:0'), in_proj_covar=tensor([0.0231, 0.0195, 0.0245, 0.0214, 0.0202, 0.0246, 0.0249, 0.0206], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], device='cuda:0') 2023-02-09 02:50:14,767 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233495.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:50:23,099 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([3.0871, 2.9746, 2.8020, 1.6875, 2.7092, 2.8329, 2.7283, 2.7069], device='cuda:0'), covar=tensor([0.1212, 0.0877, 0.1226, 0.4502, 0.1196, 0.1345, 0.1726, 0.1139], device='cuda:0'), in_proj_covar=tensor([0.0549, 0.0462, 0.0454, 0.0563, 0.0446, 0.0470, 0.0445, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 02:50:23,183 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.1792, 1.3147, 1.5209, 1.2748, 0.8641, 1.3342, 1.2380, 1.0709], device='cuda:0'), covar=tensor([0.0664, 0.1218, 0.1603, 0.1419, 0.0573, 0.1419, 0.0698, 0.0683], device='cuda:0'), in_proj_covar=tensor([0.0101, 0.0155, 0.0191, 0.0163, 0.0102, 0.0165, 0.0114, 0.0148], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0008, 0.0008], device='cuda:0') 2023-02-09 02:50:34,193 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.538e+02 2.385e+02 2.900e+02 3.377e+02 5.605e+02, threshold=5.800e+02, percent-clipped=0.0 2023-02-09 02:50:34,214 INFO [train.py:901] (0/4) Epoch 29, batch 7200, loss[loss=0.1804, simple_loss=0.2637, pruned_loss=0.04854, over 8086.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2805, pruned_loss=0.0565, over 1613352.73 frames. ], batch size: 21, lr: 2.58e-03, grad_scale: 8.0 2023-02-09 02:51:10,566 INFO [train.py:901] (0/4) Epoch 29, batch 7250, loss[loss=0.2027, simple_loss=0.289, pruned_loss=0.05822, over 8025.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2813, pruned_loss=0.05646, over 1616070.20 frames. ], batch size: 22, lr: 2.58e-03, grad_scale: 8.0 2023-02-09 02:51:11,761 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 2023-02-09 02:51:36,858 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.8373, 2.3145, 3.7364, 1.8545, 1.9429, 3.6822, 0.7785, 2.1054], device='cuda:0'), covar=tensor([0.1255, 0.1081, 0.0202, 0.1577, 0.2131, 0.0261, 0.1958, 0.1315], device='cuda:0'), in_proj_covar=tensor([0.0204, 0.0207, 0.0138, 0.0227, 0.0281, 0.0149, 0.0174, 0.0202], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 02:51:46,283 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.552e+02 2.384e+02 2.959e+02 3.398e+02 1.041e+03, threshold=5.918e+02, percent-clipped=4.0 2023-02-09 02:51:46,303 INFO [train.py:901] (0/4) Epoch 29, batch 7300, loss[loss=0.1774, simple_loss=0.2607, pruned_loss=0.04699, over 7714.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.281, pruned_loss=0.05662, over 1615240.84 frames. ], batch size: 18, lr: 2.58e-03, grad_scale: 8.0 2023-02-09 02:51:46,371 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=233621.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:52:13,025 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233657.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:52:22,374 INFO [train.py:901] (0/4) Epoch 29, batch 7350, loss[loss=0.1858, simple_loss=0.2756, pruned_loss=0.04805, over 7971.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2798, pruned_loss=0.05625, over 1613875.80 frames. ], batch size: 21, lr: 2.58e-03, grad_scale: 8.0 2023-02-09 02:52:27,578 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.4210, 1.6763, 2.1933, 1.3726, 1.5647, 1.6876, 1.5445, 1.6111], device='cuda:0'), covar=tensor([0.2107, 0.2678, 0.1037, 0.4803, 0.2158, 0.3642, 0.2574, 0.2363], device='cuda:0'), in_proj_covar=tensor([0.0546, 0.0644, 0.0569, 0.0678, 0.0667, 0.0619, 0.0570, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:0') 2023-02-09 02:52:28,055 WARNING [train.py:1067] (0/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-02-09 02:52:42,955 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([6.0339, 1.8179, 6.2189, 2.2718, 5.6506, 5.1838, 5.7236, 5.6503], device='cuda:0'), covar=tensor([0.0476, 0.4696, 0.0321, 0.4225, 0.1005, 0.0938, 0.0490, 0.0527], device='cuda:0'), in_proj_covar=tensor([0.0695, 0.0671, 0.0756, 0.0663, 0.0755, 0.0642, 0.0650, 0.0733], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 02:52:48,257 WARNING [train.py:1067] (0/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-02-09 02:52:58,118 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.683e+02 2.663e+02 3.074e+02 4.100e+02 9.512e+02, threshold=6.147e+02, percent-clipped=7.0 2023-02-09 02:52:58,138 INFO [train.py:901] (0/4) Epoch 29, batch 7400, loss[loss=0.1905, simple_loss=0.2707, pruned_loss=0.05514, over 7971.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.2798, pruned_loss=0.05623, over 1610530.89 frames. ], batch size: 21, lr: 2.58e-03, grad_scale: 8.0 2023-02-09 02:52:59,689 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233723.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:53:03,233 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233728.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:53:08,960 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233736.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:53:18,939 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233748.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:53:21,107 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233751.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:53:32,423 WARNING [train.py:1067] (0/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-02-09 02:53:36,092 INFO [train.py:901] (0/4) Epoch 29, batch 7450, loss[loss=0.2089, simple_loss=0.3072, pruned_loss=0.05528, over 8106.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2799, pruned_loss=0.05658, over 1605796.04 frames. ], batch size: 23, lr: 2.58e-03, grad_scale: 8.0 2023-02-09 02:53:39,820 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233776.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:53:42,697 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.7129, 1.7525, 2.0571, 1.7955, 1.1489, 1.7548, 2.2841, 1.9774], device='cuda:0'), covar=tensor([0.0514, 0.1224, 0.1613, 0.1374, 0.0594, 0.1446, 0.0617, 0.0627], device='cuda:0'), in_proj_covar=tensor([0.0101, 0.0155, 0.0191, 0.0163, 0.0102, 0.0165, 0.0115, 0.0148], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0008, 0.0008], device='cuda:0') 2023-02-09 02:54:12,667 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.619e+02 2.228e+02 2.637e+02 3.402e+02 7.399e+02, threshold=5.273e+02, percent-clipped=2.0 2023-02-09 02:54:12,687 INFO [train.py:901] (0/4) Epoch 29, batch 7500, loss[loss=0.2013, simple_loss=0.2905, pruned_loss=0.05606, over 8467.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2806, pruned_loss=0.05673, over 1608222.29 frames. ], batch size: 25, lr: 2.57e-03, grad_scale: 8.0 2023-02-09 02:54:48,750 INFO [train.py:901] (0/4) Epoch 29, batch 7550, loss[loss=0.1721, simple_loss=0.2625, pruned_loss=0.04089, over 8039.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2802, pruned_loss=0.0565, over 1606517.10 frames. ], batch size: 22, lr: 2.57e-03, grad_scale: 8.0 2023-02-09 02:55:00,149 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.0178, 2.1833, 1.8192, 2.7821, 1.3613, 1.6806, 2.0979, 2.1963], device='cuda:0'), covar=tensor([0.0702, 0.0789, 0.0869, 0.0324, 0.1055, 0.1221, 0.0734, 0.0710], device='cuda:0'), in_proj_covar=tensor([0.0231, 0.0195, 0.0244, 0.0213, 0.0202, 0.0246, 0.0249, 0.0205], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], device='cuda:0') 2023-02-09 02:55:24,356 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.668e+02 2.424e+02 2.935e+02 3.534e+02 7.288e+02, threshold=5.870e+02, percent-clipped=3.0 2023-02-09 02:55:24,377 INFO [train.py:901] (0/4) Epoch 29, batch 7600, loss[loss=0.2332, simple_loss=0.3064, pruned_loss=0.08003, over 7135.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2807, pruned_loss=0.05633, over 1611218.29 frames. ], batch size: 71, lr: 2.57e-03, grad_scale: 8.0 2023-02-09 02:55:24,646 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.3400, 2.2195, 3.2535, 2.4921, 2.9282, 2.1850, 2.1814, 2.2245], device='cuda:0'), covar=tensor([0.5841, 0.5798, 0.2265, 0.4531, 0.3170, 0.4357, 0.2550, 0.5149], device='cuda:0'), in_proj_covar=tensor([0.0974, 0.1037, 0.0845, 0.1009, 0.1033, 0.0945, 0.0781, 0.0859], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 02:56:01,032 INFO [train.py:901] (0/4) Epoch 29, batch 7650, loss[loss=0.1686, simple_loss=0.2547, pruned_loss=0.04129, over 7918.00 frames. ], tot_loss[loss=0.196, simple_loss=0.2803, pruned_loss=0.05587, over 1612928.38 frames. ], batch size: 20, lr: 2.57e-03, grad_scale: 8.0 2023-02-09 02:56:12,976 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([3.8693, 1.4151, 4.0154, 1.4108, 3.6252, 3.3274, 3.6614, 3.5706], device='cuda:0'), covar=tensor([0.0646, 0.4566, 0.0606, 0.4501, 0.1052, 0.1074, 0.0661, 0.0704], device='cuda:0'), in_proj_covar=tensor([0.0693, 0.0670, 0.0754, 0.0662, 0.0751, 0.0641, 0.0649, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 02:56:16,585 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233992.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:56:22,137 INFO [checkpoint.py:75] (0/4) Saving checkpoint to pruned_transducer_stateless7_streaming/exp/v1/checkpoint-234000.pt 2023-02-09 02:56:23,675 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=234001.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:56:35,521 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234017.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:56:38,037 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.742e+02 2.208e+02 2.731e+02 3.331e+02 6.993e+02, threshold=5.462e+02, percent-clipped=2.0 2023-02-09 02:56:38,057 INFO [train.py:901] (0/4) Epoch 29, batch 7700, loss[loss=0.1589, simple_loss=0.246, pruned_loss=0.0359, over 7451.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2811, pruned_loss=0.05596, over 1611192.28 frames. ], batch size: 17, lr: 2.57e-03, grad_scale: 8.0 2023-02-09 02:56:49,496 WARNING [train.py:1067] (0/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-02-09 02:56:53,148 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234043.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:57:12,864 INFO [train.py:901] (0/4) Epoch 29, batch 7750, loss[loss=0.1715, simple_loss=0.2449, pruned_loss=0.04905, over 5995.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2804, pruned_loss=0.056, over 1610937.67 frames. ], batch size: 13, lr: 2.57e-03, grad_scale: 8.0 2023-02-09 02:57:13,646 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=234072.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:57:45,241 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234116.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:57:48,454 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.737e+02 2.355e+02 2.809e+02 3.505e+02 7.382e+02, threshold=5.617e+02, percent-clipped=2.0 2023-02-09 02:57:48,495 INFO [train.py:901] (0/4) Epoch 29, batch 7800, loss[loss=0.1906, simple_loss=0.2827, pruned_loss=0.04924, over 8288.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2804, pruned_loss=0.05611, over 1608957.17 frames. ], batch size: 23, lr: 2.57e-03, grad_scale: 8.0 2023-02-09 02:58:24,432 INFO [train.py:901] (0/4) Epoch 29, batch 7850, loss[loss=0.1637, simple_loss=0.2528, pruned_loss=0.03729, over 7965.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.2798, pruned_loss=0.05577, over 1607945.35 frames. ], batch size: 21, lr: 2.57e-03, grad_scale: 8.0 2023-02-09 02:58:36,023 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234187.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:58:58,932 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234220.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:58:59,506 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.576e+02 2.264e+02 2.745e+02 3.500e+02 8.048e+02, threshold=5.490e+02, percent-clipped=4.0 2023-02-09 02:58:59,527 INFO [train.py:901] (0/4) Epoch 29, batch 7900, loss[loss=0.2015, simple_loss=0.2842, pruned_loss=0.0594, over 8661.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.2799, pruned_loss=0.05592, over 1605850.28 frames. ], batch size: 34, lr: 2.57e-03, grad_scale: 8.0 2023-02-09 02:59:19,152 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234249.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 02:59:34,070 INFO [train.py:901] (0/4) Epoch 29, batch 7950, loss[loss=0.2022, simple_loss=0.2935, pruned_loss=0.05548, over 8454.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.28, pruned_loss=0.05637, over 1608682.20 frames. ], batch size: 27, lr: 2.57e-03, grad_scale: 8.0 2023-02-09 02:59:40,221 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=2.25 vs. limit=5.0 2023-02-09 03:00:10,399 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.560e+02 2.366e+02 2.676e+02 3.633e+02 8.832e+02, threshold=5.352e+02, percent-clipped=6.0 2023-02-09 03:00:10,419 INFO [train.py:901] (0/4) Epoch 29, batch 8000, loss[loss=0.1868, simple_loss=0.2726, pruned_loss=0.05044, over 8197.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2804, pruned_loss=0.05648, over 1610456.57 frames. ], batch size: 23, lr: 2.57e-03, grad_scale: 8.0 2023-02-09 03:00:44,627 INFO [train.py:901] (0/4) Epoch 29, batch 8050, loss[loss=0.2566, simple_loss=0.3258, pruned_loss=0.09373, over 7192.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.2789, pruned_loss=0.05623, over 1587524.57 frames. ], batch size: 73, lr: 2.57e-03, grad_scale: 8.0 2023-02-09 03:00:45,534 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234372.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:00:55,734 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=234387.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:01:02,586 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234397.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:01:07,577 INFO [checkpoint.py:75] (0/4) Saving checkpoint to pruned_transducer_stateless7_streaming/exp/v1/epoch-29.pt 2023-02-09 03:01:20,130 WARNING [train.py:1067] (0/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 2023-02-09 03:01:23,764 INFO [train.py:901] (0/4) Epoch 30, batch 0, loss[loss=0.2073, simple_loss=0.2937, pruned_loss=0.06046, over 8037.00 frames. ], tot_loss[loss=0.2073, simple_loss=0.2937, pruned_loss=0.06046, over 8037.00 frames. ], batch size: 22, lr: 2.53e-03, grad_scale: 8.0 2023-02-09 03:01:23,764 INFO [train.py:926] (0/4) Computing validation loss 2023-02-09 03:01:35,943 INFO [train.py:935] (0/4) Epoch 30, validation: loss=0.1704, simple_loss=0.27, pruned_loss=0.03537, over 944034.00 frames. 2023-02-09 03:01:35,944 INFO [train.py:936] (0/4) Maximum memory allocated so far is 6641MB 2023-02-09 03:01:47,823 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.533e+02 2.332e+02 2.743e+02 3.464e+02 7.498e+02, threshold=5.486e+02, percent-clipped=3.0 2023-02-09 03:01:51,469 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.9942, 2.3435, 1.8958, 2.8816, 1.4111, 1.6784, 2.0272, 2.2555], device='cuda:0'), covar=tensor([0.0718, 0.0695, 0.0856, 0.0328, 0.1090, 0.1231, 0.0846, 0.0751], device='cuda:0'), in_proj_covar=tensor([0.0232, 0.0196, 0.0246, 0.0214, 0.0203, 0.0247, 0.0251, 0.0206], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], device='cuda:0') 2023-02-09 03:01:52,029 WARNING [train.py:1067] (0/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 2023-02-09 03:02:04,490 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234443.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:02:12,501 INFO [train.py:901] (0/4) Epoch 30, batch 50, loss[loss=0.1856, simple_loss=0.266, pruned_loss=0.05259, over 8090.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2799, pruned_loss=0.05627, over 362826.06 frames. ], batch size: 21, lr: 2.53e-03, grad_scale: 8.0 2023-02-09 03:02:23,117 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234468.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:02:25,364 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.3674, 1.5841, 1.5956, 1.0792, 1.6665, 1.2743, 0.2578, 1.5735], device='cuda:0'), covar=tensor([0.0579, 0.0468, 0.0412, 0.0612, 0.0491, 0.1130, 0.1053, 0.0340], device='cuda:0'), in_proj_covar=tensor([0.0477, 0.0418, 0.0373, 0.0466, 0.0401, 0.0558, 0.0407, 0.0445], device='cuda:0'), out_proj_covar=tensor([1.2634e-04, 1.0819e-04, 9.7132e-05, 1.2162e-04, 1.0500e-04, 1.5531e-04, 1.0858e-04, 1.1626e-04], device='cuda:0') 2023-02-09 03:02:28,117 WARNING [train.py:1067] (0/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 2023-02-09 03:02:49,607 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234502.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:02:50,908 INFO [train.py:901] (0/4) Epoch 30, batch 100, loss[loss=0.1854, simple_loss=0.2796, pruned_loss=0.04561, over 8505.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.2808, pruned_loss=0.05533, over 643515.93 frames. ], batch size: 28, lr: 2.53e-03, grad_scale: 8.0 2023-02-09 03:02:54,593 WARNING [train.py:1067] (0/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 2023-02-09 03:03:03,349 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.539e+02 2.294e+02 2.794e+02 3.449e+02 7.855e+02, threshold=5.588e+02, percent-clipped=7.0 2023-02-09 03:03:28,104 INFO [train.py:901] (0/4) Epoch 30, batch 150, loss[loss=0.2164, simple_loss=0.3061, pruned_loss=0.06339, over 8495.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2815, pruned_loss=0.05617, over 862624.66 frames. ], batch size: 26, lr: 2.53e-03, grad_scale: 8.0 2023-02-09 03:03:29,553 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 2023-02-09 03:03:35,358 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=234564.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:03:56,912 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=234593.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:04:04,596 INFO [train.py:901] (0/4) Epoch 30, batch 200, loss[loss=0.2299, simple_loss=0.3148, pruned_loss=0.0725, over 8601.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2822, pruned_loss=0.05598, over 1032107.50 frames. ], batch size: 34, lr: 2.53e-03, grad_scale: 8.0 2023-02-09 03:04:07,725 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 2023-02-09 03:04:16,947 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.375e+02 2.329e+02 2.797e+02 3.759e+02 1.341e+03, threshold=5.593e+02, percent-clipped=8.0 2023-02-09 03:04:40,212 INFO [train.py:901] (0/4) Epoch 30, batch 250, loss[loss=0.191, simple_loss=0.2654, pruned_loss=0.05833, over 7977.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.2809, pruned_loss=0.05543, over 1161714.73 frames. ], batch size: 21, lr: 2.53e-03, grad_scale: 8.0 2023-02-09 03:04:48,444 WARNING [train.py:1067] (0/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 2023-02-09 03:04:50,788 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.7406, 1.5806, 2.4708, 1.3913, 1.2110, 2.3883, 0.4361, 1.4633], device='cuda:0'), covar=tensor([0.1579, 0.1298, 0.0327, 0.1261, 0.2591, 0.0420, 0.1909, 0.1376], device='cuda:0'), in_proj_covar=tensor([0.0203, 0.0208, 0.0139, 0.0227, 0.0280, 0.0149, 0.0174, 0.0203], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 03:04:57,629 WARNING [train.py:1067] (0/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 2023-02-09 03:04:58,551 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234679.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:05:04,648 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234687.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:05:16,625 INFO [train.py:901] (0/4) Epoch 30, batch 300, loss[loss=0.1753, simple_loss=0.2529, pruned_loss=0.04884, over 7714.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2822, pruned_loss=0.05646, over 1267197.18 frames. ], batch size: 18, lr: 2.53e-03, grad_scale: 8.0 2023-02-09 03:05:19,753 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234708.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:05:29,382 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.648e+02 2.381e+02 2.888e+02 3.640e+02 7.253e+02, threshold=5.776e+02, percent-clipped=4.0 2023-02-09 03:05:45,734 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234744.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:05:46,523 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.8849, 2.1276, 2.1313, 1.5118, 2.2766, 1.6825, 0.6870, 2.1398], device='cuda:0'), covar=tensor([0.0623, 0.0365, 0.0374, 0.0656, 0.0497, 0.0976, 0.1016, 0.0339], device='cuda:0'), in_proj_covar=tensor([0.0477, 0.0416, 0.0372, 0.0465, 0.0400, 0.0555, 0.0405, 0.0443], device='cuda:0'), out_proj_covar=tensor([1.2618e-04, 1.0768e-04, 9.6869e-05, 1.2151e-04, 1.0482e-04, 1.5420e-04, 1.0793e-04, 1.1575e-04], device='cuda:0') 2023-02-09 03:05:52,647 INFO [train.py:901] (0/4) Epoch 30, batch 350, loss[loss=0.2019, simple_loss=0.2934, pruned_loss=0.0552, over 8354.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2814, pruned_loss=0.05701, over 1340121.49 frames. ], batch size: 24, lr: 2.53e-03, grad_scale: 8.0 2023-02-09 03:05:55,588 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234758.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:06:14,224 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234783.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:06:27,364 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234800.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:06:30,086 INFO [train.py:901] (0/4) Epoch 30, batch 400, loss[loss=0.172, simple_loss=0.261, pruned_loss=0.04149, over 8129.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.282, pruned_loss=0.05721, over 1402342.04 frames. ], batch size: 22, lr: 2.53e-03, grad_scale: 8.0 2023-02-09 03:06:42,231 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.545e+02 2.513e+02 2.926e+02 3.697e+02 1.204e+03, threshold=5.852e+02, percent-clipped=7.0 2023-02-09 03:07:06,512 INFO [train.py:901] (0/4) Epoch 30, batch 450, loss[loss=0.2098, simple_loss=0.2947, pruned_loss=0.06246, over 8235.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2828, pruned_loss=0.05726, over 1449261.75 frames. ], batch size: 22, lr: 2.53e-03, grad_scale: 8.0 2023-02-09 03:07:38,964 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234900.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:07:42,332 INFO [train.py:901] (0/4) Epoch 30, batch 500, loss[loss=0.176, simple_loss=0.2614, pruned_loss=0.04532, over 8109.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2822, pruned_loss=0.057, over 1490830.31 frames. ], batch size: 23, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:07:54,773 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.531e+02 2.393e+02 2.948e+02 3.833e+02 6.284e+02, threshold=5.896e+02, percent-clipped=1.0 2023-02-09 03:08:04,868 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234935.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:08:18,296 INFO [train.py:901] (0/4) Epoch 30, batch 550, loss[loss=0.1888, simple_loss=0.2802, pruned_loss=0.04866, over 8030.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2821, pruned_loss=0.05719, over 1518835.36 frames. ], batch size: 22, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:08:22,859 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234960.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:08:26,380 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234964.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:08:34,795 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234976.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:08:36,247 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234978.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:08:44,125 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234989.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:08:54,540 INFO [train.py:901] (0/4) Epoch 30, batch 600, loss[loss=0.1781, simple_loss=0.2587, pruned_loss=0.04879, over 7530.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2828, pruned_loss=0.05754, over 1541779.07 frames. ], batch size: 18, lr: 2.52e-03, grad_scale: 16.0 2023-02-09 03:09:06,006 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.582e+02 2.485e+02 2.961e+02 3.544e+02 6.861e+02, threshold=5.922e+02, percent-clipped=1.0 2023-02-09 03:09:10,820 WARNING [train.py:1067] (0/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 2023-02-09 03:09:13,646 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=235031.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:09:29,975 INFO [train.py:901] (0/4) Epoch 30, batch 650, loss[loss=0.2099, simple_loss=0.2892, pruned_loss=0.06531, over 8676.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2811, pruned_loss=0.05663, over 1554121.66 frames. ], batch size: 39, lr: 2.52e-03, grad_scale: 16.0 2023-02-09 03:09:54,561 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=235088.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:10:05,667 INFO [train.py:901] (0/4) Epoch 30, batch 700, loss[loss=0.1978, simple_loss=0.2859, pruned_loss=0.05491, over 8445.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2815, pruned_loss=0.05693, over 1565056.16 frames. ], batch size: 27, lr: 2.52e-03, grad_scale: 16.0 2023-02-09 03:10:12,457 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0 2023-02-09 03:10:17,681 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.543e+02 2.477e+02 3.055e+02 3.959e+02 7.285e+02, threshold=6.109e+02, percent-clipped=6.0 2023-02-09 03:10:33,428 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=235144.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:10:34,922 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235146.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:10:41,067 INFO [train.py:901] (0/4) Epoch 30, batch 750, loss[loss=0.1833, simple_loss=0.2652, pruned_loss=0.05067, over 8068.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2816, pruned_loss=0.05689, over 1581164.82 frames. ], batch size: 21, lr: 2.52e-03, grad_scale: 16.0 2023-02-09 03:10:59,030 WARNING [train.py:1067] (0/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 2023-02-09 03:10:59,130 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([3.8029, 3.7904, 3.4769, 1.9211, 3.3639, 3.6310, 3.3456, 3.4628], device='cuda:0'), covar=tensor([0.0924, 0.0681, 0.1081, 0.4458, 0.1032, 0.0976, 0.1442, 0.0870], device='cuda:0'), in_proj_covar=tensor([0.0551, 0.0461, 0.0451, 0.0562, 0.0445, 0.0470, 0.0446, 0.0414], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 03:11:08,059 WARNING [train.py:1067] (0/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 2023-02-09 03:11:17,105 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235203.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:11:17,666 INFO [train.py:901] (0/4) Epoch 30, batch 800, loss[loss=0.1642, simple_loss=0.2657, pruned_loss=0.03138, over 8477.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2818, pruned_loss=0.05712, over 1586358.78 frames. ], batch size: 25, lr: 2.52e-03, grad_scale: 16.0 2023-02-09 03:11:30,476 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.721e+02 2.417e+02 2.836e+02 3.328e+02 8.160e+02, threshold=5.671e+02, percent-clipped=2.0 2023-02-09 03:11:46,746 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=235244.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:11:53,481 INFO [train.py:901] (0/4) Epoch 30, batch 850, loss[loss=0.1765, simple_loss=0.2616, pruned_loss=0.04569, over 8447.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2808, pruned_loss=0.05665, over 1592312.02 frames. ], batch size: 25, lr: 2.52e-03, grad_scale: 16.0 2023-02-09 03:11:57,047 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235259.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:12:07,386 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235273.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:12:20,601 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([4.4006, 4.4626, 4.0228, 1.9658, 3.9069, 4.1098, 4.0051, 3.9094], device='cuda:0'), covar=tensor([0.0909, 0.0628, 0.1258, 0.4720, 0.1050, 0.0941, 0.1362, 0.0753], device='cuda:0'), in_proj_covar=tensor([0.0547, 0.0458, 0.0448, 0.0559, 0.0443, 0.0467, 0.0443, 0.0411], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 03:12:25,088 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 2023-02-09 03:12:29,564 INFO [train.py:901] (0/4) Epoch 30, batch 900, loss[loss=0.1756, simple_loss=0.2462, pruned_loss=0.05247, over 7700.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.2795, pruned_loss=0.056, over 1595774.43 frames. ], batch size: 18, lr: 2.52e-03, grad_scale: 16.0 2023-02-09 03:12:41,045 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=235320.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:12:41,595 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.572e+02 2.438e+02 3.006e+02 3.865e+02 6.238e+02, threshold=6.012e+02, percent-clipped=6.0 2023-02-09 03:12:42,348 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=235322.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:13:05,247 INFO [train.py:901] (0/4) Epoch 30, batch 950, loss[loss=0.171, simple_loss=0.2557, pruned_loss=0.04315, over 7803.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.2793, pruned_loss=0.05583, over 1599378.95 frames. ], batch size: 20, lr: 2.52e-03, grad_scale: 16.0 2023-02-09 03:13:08,753 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235359.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:13:21,825 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235378.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 03:13:32,455 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.5605, 2.4003, 1.8674, 2.2590, 2.0533, 1.6181, 1.9828, 2.0866], device='cuda:0'), covar=tensor([0.1565, 0.0465, 0.1280, 0.0626, 0.0822, 0.1757, 0.1118, 0.1038], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0243, 0.0344, 0.0314, 0.0302, 0.0349, 0.0350, 0.0321], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 03:13:32,976 WARNING [train.py:1067] (0/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 2023-02-09 03:13:39,419 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235402.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:13:40,150 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.5449, 2.4417, 3.1056, 2.6228, 3.1045, 2.6498, 2.5314, 2.0022], device='cuda:0'), covar=tensor([0.5552, 0.5187, 0.2350, 0.4245, 0.2808, 0.3328, 0.1941, 0.6041], device='cuda:0'), in_proj_covar=tensor([0.0978, 0.1045, 0.0853, 0.1015, 0.1038, 0.0949, 0.0782, 0.0865], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 03:13:40,609 INFO [train.py:901] (0/4) Epoch 30, batch 1000, loss[loss=0.1707, simple_loss=0.2535, pruned_loss=0.04399, over 7792.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.2796, pruned_loss=0.05626, over 1602075.43 frames. ], batch size: 19, lr: 2.52e-03, grad_scale: 16.0 2023-02-09 03:13:52,258 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.648e+02 2.479e+02 3.055e+02 4.205e+02 7.814e+02, threshold=6.110e+02, percent-clipped=3.0 2023-02-09 03:13:56,469 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235427.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:14:02,538 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235435.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:14:03,948 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235437.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:14:08,428 WARNING [train.py:1067] (0/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 2023-02-09 03:14:15,931 INFO [train.py:901] (0/4) Epoch 30, batch 1050, loss[loss=0.1591, simple_loss=0.2449, pruned_loss=0.03661, over 8075.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2806, pruned_loss=0.05653, over 1606725.03 frames. ], batch size: 21, lr: 2.52e-03, grad_scale: 16.0 2023-02-09 03:14:19,405 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235459.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:14:21,148 WARNING [train.py:1067] (0/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 2023-02-09 03:14:36,648 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235484.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:14:41,816 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 2023-02-09 03:14:50,593 INFO [train.py:901] (0/4) Epoch 30, batch 1100, loss[loss=0.1506, simple_loss=0.2313, pruned_loss=0.03493, over 7791.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2806, pruned_loss=0.05654, over 1610569.37 frames. ], batch size: 19, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:14:59,101 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235515.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:15:03,872 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.639e+02 2.444e+02 3.135e+02 3.900e+02 6.752e+02, threshold=6.270e+02, percent-clipped=2.0 2023-02-09 03:15:04,840 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.7765, 2.0389, 2.0618, 1.3847, 2.1555, 1.6180, 0.5849, 1.9683], device='cuda:0'), covar=tensor([0.0674, 0.0452, 0.0398, 0.0683, 0.0504, 0.1057, 0.1118, 0.0321], device='cuda:0'), in_proj_covar=tensor([0.0483, 0.0421, 0.0375, 0.0470, 0.0404, 0.0561, 0.0410, 0.0448], device='cuda:0'), out_proj_covar=tensor([1.2799e-04, 1.0897e-04, 9.7569e-05, 1.2288e-04, 1.0573e-04, 1.5611e-04, 1.0930e-04, 1.1726e-04], device='cuda:0') 2023-02-09 03:15:09,972 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=2.32 vs. limit=5.0 2023-02-09 03:15:12,933 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=3.87 vs. limit=5.0 2023-02-09 03:15:17,351 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235540.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:15:26,820 INFO [train.py:901] (0/4) Epoch 30, batch 1150, loss[loss=0.2247, simple_loss=0.2992, pruned_loss=0.07514, over 8195.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.2804, pruned_loss=0.05593, over 1617035.56 frames. ], batch size: 48, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:15:35,846 WARNING [train.py:1067] (0/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 2023-02-09 03:15:39,059 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.67 vs. limit=2.0 2023-02-09 03:16:03,360 INFO [train.py:901] (0/4) Epoch 30, batch 1200, loss[loss=0.1983, simple_loss=0.2877, pruned_loss=0.05443, over 8378.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.281, pruned_loss=0.0562, over 1618357.61 frames. ], batch size: 48, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:16:11,480 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235615.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:16:12,739 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=235617.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:16:16,062 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.630e+02 2.378e+02 2.945e+02 3.640e+02 8.540e+02, threshold=5.890e+02, percent-clipped=4.0 2023-02-09 03:16:30,043 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235640.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:16:32,813 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.3656, 2.1501, 2.6754, 2.3381, 2.6840, 2.3647, 2.2672, 1.5579], device='cuda:0'), covar=tensor([0.5541, 0.4734, 0.1961, 0.3552, 0.2332, 0.3175, 0.1937, 0.5086], device='cuda:0'), in_proj_covar=tensor([0.0970, 0.1037, 0.0846, 0.1008, 0.1030, 0.0943, 0.0777, 0.0860], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 03:16:39,494 INFO [train.py:901] (0/4) Epoch 30, batch 1250, loss[loss=0.1926, simple_loss=0.2728, pruned_loss=0.05621, over 7239.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2819, pruned_loss=0.05668, over 1617900.65 frames. ], batch size: 16, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:17:06,201 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235691.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:17:07,630 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235693.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:17:10,373 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235697.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:17:15,251 INFO [train.py:901] (0/4) Epoch 30, batch 1300, loss[loss=0.1986, simple_loss=0.2974, pruned_loss=0.04992, over 8486.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2823, pruned_loss=0.05683, over 1619721.29 frames. ], batch size: 29, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:17:23,814 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235716.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:17:25,211 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235718.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:17:27,378 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.2181, 2.1280, 2.6664, 2.2846, 2.6427, 2.3201, 2.1744, 1.5727], device='cuda:0'), covar=tensor([0.6022, 0.5076, 0.2208, 0.3950, 0.2786, 0.3349, 0.2016, 0.5665], device='cuda:0'), in_proj_covar=tensor([0.0964, 0.1030, 0.0841, 0.1002, 0.1025, 0.0939, 0.0773, 0.0855], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 03:17:27,735 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.629e+02 2.444e+02 2.774e+02 3.314e+02 6.214e+02, threshold=5.548e+02, percent-clipped=2.0 2023-02-09 03:17:27,829 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=235722.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 03:17:34,617 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235732.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:17:35,440 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=4.22 vs. limit=5.0 2023-02-09 03:17:50,119 INFO [train.py:901] (0/4) Epoch 30, batch 1350, loss[loss=0.1677, simple_loss=0.2434, pruned_loss=0.04593, over 7549.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2828, pruned_loss=0.05682, over 1621018.32 frames. ], batch size: 18, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:18:26,589 INFO [train.py:901] (0/4) Epoch 30, batch 1400, loss[loss=0.1669, simple_loss=0.2449, pruned_loss=0.04442, over 7660.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2809, pruned_loss=0.0563, over 1618278.91 frames. ], batch size: 19, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:18:30,253 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235809.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:18:39,093 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.499e+02 2.338e+02 2.741e+02 3.583e+02 7.907e+02, threshold=5.482e+02, percent-clipped=6.0 2023-02-09 03:18:49,570 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235837.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 03:19:00,849 INFO [train.py:901] (0/4) Epoch 30, batch 1450, loss[loss=0.1797, simple_loss=0.2774, pruned_loss=0.04097, over 8278.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2806, pruned_loss=0.05593, over 1616056.42 frames. ], batch size: 23, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:19:07,593 WARNING [train.py:1067] (0/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 2023-02-09 03:19:22,545 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 2023-02-09 03:19:38,202 INFO [train.py:901] (0/4) Epoch 30, batch 1500, loss[loss=0.1915, simple_loss=0.2818, pruned_loss=0.05059, over 8435.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2809, pruned_loss=0.05601, over 1617228.40 frames. ], batch size: 27, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:19:42,088 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.26 vs. limit=2.0 2023-02-09 03:19:51,220 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.723e+02 2.306e+02 2.900e+02 3.560e+02 8.272e+02, threshold=5.801e+02, percent-clipped=7.0 2023-02-09 03:19:56,742 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 2023-02-09 03:20:14,193 INFO [train.py:901] (0/4) Epoch 30, batch 1550, loss[loss=0.1904, simple_loss=0.282, pruned_loss=0.04938, over 8646.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2814, pruned_loss=0.05597, over 1620647.57 frames. ], batch size: 34, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:20:17,116 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.2851, 1.6362, 4.3037, 1.8736, 2.4658, 4.8549, 5.0367, 4.2033], device='cuda:0'), covar=tensor([0.1123, 0.1985, 0.0280, 0.2017, 0.1256, 0.0197, 0.0552, 0.0587], device='cuda:0'), in_proj_covar=tensor([0.0309, 0.0329, 0.0296, 0.0329, 0.0327, 0.0282, 0.0447, 0.0310], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 03:20:22,061 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-02-09 03:20:27,562 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.8766, 1.8553, 2.5143, 1.5957, 1.4477, 2.5130, 0.4707, 1.5506], device='cuda:0'), covar=tensor([0.1473, 0.1063, 0.0310, 0.1190, 0.2225, 0.0302, 0.1829, 0.1366], device='cuda:0'), in_proj_covar=tensor([0.0203, 0.0208, 0.0139, 0.0226, 0.0280, 0.0149, 0.0175, 0.0202], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 03:20:38,887 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235988.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:20:47,546 INFO [checkpoint.py:75] (0/4) Saving checkpoint to pruned_transducer_stateless7_streaming/exp/v1/checkpoint-236000.pt 2023-02-09 03:20:51,300 INFO [train.py:901] (0/4) Epoch 30, batch 1600, loss[loss=0.189, simple_loss=0.2697, pruned_loss=0.05422, over 8077.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2813, pruned_loss=0.05574, over 1621820.12 frames. ], batch size: 21, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:20:57,863 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=236013.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:21:00,115 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.2784, 2.0693, 1.6197, 1.9605, 1.7270, 1.4842, 1.6713, 1.7173], device='cuda:0'), covar=tensor([0.1399, 0.0472, 0.1401, 0.0581, 0.0808, 0.1577, 0.0982, 0.0964], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0246, 0.0349, 0.0317, 0.0304, 0.0352, 0.0354, 0.0325], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 03:21:04,933 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.669e+02 2.693e+02 3.134e+02 4.092e+02 8.333e+02, threshold=6.267e+02, percent-clipped=7.0 2023-02-09 03:21:15,461 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=236036.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:21:19,095 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=236041.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:21:28,393 INFO [train.py:901] (0/4) Epoch 30, batch 1650, loss[loss=0.1616, simple_loss=0.2419, pruned_loss=0.04066, over 7452.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.2807, pruned_loss=0.05559, over 1619205.66 frames. ], batch size: 17, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:21:56,809 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=236093.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 03:22:04,824 INFO [train.py:901] (0/4) Epoch 30, batch 1700, loss[loss=0.1749, simple_loss=0.2515, pruned_loss=0.04917, over 7541.00 frames. ], tot_loss[loss=0.1952, simple_loss=0.2802, pruned_loss=0.05504, over 1619674.91 frames. ], batch size: 18, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:22:15,388 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=236118.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 03:22:17,735 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.648e+02 2.438e+02 2.804e+02 3.459e+02 5.840e+02, threshold=5.608e+02, percent-clipped=0.0 2023-02-09 03:22:34,518 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=2.59 vs. limit=5.0 2023-02-09 03:22:40,513 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=236153.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:22:41,159 INFO [train.py:901] (0/4) Epoch 30, batch 1750, loss[loss=0.19, simple_loss=0.2806, pruned_loss=0.04972, over 8537.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.2809, pruned_loss=0.05539, over 1623601.95 frames. ], batch size: 49, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:22:42,651 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=236156.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:23:07,926 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.1638, 2.0365, 2.4921, 2.1803, 2.5490, 2.2743, 2.1450, 1.5526], device='cuda:0'), covar=tensor([0.5919, 0.5270, 0.2284, 0.4119, 0.2696, 0.3352, 0.2027, 0.5365], device='cuda:0'), in_proj_covar=tensor([0.0972, 0.1037, 0.0848, 0.1008, 0.1032, 0.0946, 0.0778, 0.0860], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 03:23:16,218 INFO [train.py:901] (0/4) Epoch 30, batch 1800, loss[loss=0.2484, simple_loss=0.327, pruned_loss=0.08493, over 8506.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.282, pruned_loss=0.05615, over 1624405.25 frames. ], batch size: 26, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:23:29,387 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.538e+02 2.503e+02 3.165e+02 3.852e+02 7.294e+02, threshold=6.329e+02, percent-clipped=5.0 2023-02-09 03:23:52,507 INFO [train.py:901] (0/4) Epoch 30, batch 1850, loss[loss=0.2371, simple_loss=0.3321, pruned_loss=0.07103, over 8249.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2812, pruned_loss=0.05629, over 1619653.95 frames. ], batch size: 24, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:24:03,974 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=236268.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:24:28,813 INFO [train.py:901] (0/4) Epoch 30, batch 1900, loss[loss=0.1686, simple_loss=0.2388, pruned_loss=0.04919, over 7436.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.281, pruned_loss=0.05616, over 1620042.86 frames. ], batch size: 17, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:24:41,170 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.66 vs. limit=2.0 2023-02-09 03:24:41,427 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.517e+02 2.325e+02 3.004e+02 3.832e+02 8.674e+02, threshold=6.008e+02, percent-clipped=3.0 2023-02-09 03:25:01,722 WARNING [train.py:1067] (0/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 2023-02-09 03:25:05,199 INFO [train.py:901] (0/4) Epoch 30, batch 1950, loss[loss=0.1892, simple_loss=0.2793, pruned_loss=0.04955, over 8244.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2811, pruned_loss=0.05618, over 1617187.41 frames. ], batch size: 24, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:25:13,610 WARNING [train.py:1067] (0/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 2023-02-09 03:25:24,198 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=236380.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:25:25,020 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.5628, 1.7792, 1.8273, 1.1939, 1.8609, 1.4281, 0.4719, 1.7322], device='cuda:0'), covar=tensor([0.0629, 0.0423, 0.0342, 0.0643, 0.0561, 0.0976, 0.1039, 0.0322], device='cuda:0'), in_proj_covar=tensor([0.0477, 0.0418, 0.0373, 0.0465, 0.0400, 0.0555, 0.0405, 0.0442], device='cuda:0'), out_proj_covar=tensor([1.2627e-04, 1.0814e-04, 9.7134e-05, 1.2130e-04, 1.0478e-04, 1.5435e-04, 1.0794e-04, 1.1561e-04], device='cuda:0') 2023-02-09 03:25:25,042 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.4645, 1.9320, 2.8471, 1.4609, 2.0689, 1.9098, 1.6111, 2.1767], device='cuda:0'), covar=tensor([0.2133, 0.2703, 0.0985, 0.4923, 0.2147, 0.3563, 0.2628, 0.2339], device='cuda:0'), in_proj_covar=tensor([0.0546, 0.0646, 0.0569, 0.0681, 0.0670, 0.0622, 0.0573, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:0') 2023-02-09 03:25:33,332 WARNING [train.py:1067] (0/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 2023-02-09 03:25:41,050 INFO [train.py:901] (0/4) Epoch 30, batch 2000, loss[loss=0.1937, simple_loss=0.2997, pruned_loss=0.04389, over 8291.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2811, pruned_loss=0.05649, over 1613739.73 frames. ], batch size: 23, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:25:43,360 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.2414, 2.0935, 2.6512, 2.2318, 2.6299, 2.3472, 2.1872, 1.6319], device='cuda:0'), covar=tensor([0.5935, 0.5443, 0.2208, 0.4291, 0.2860, 0.3388, 0.2090, 0.5733], device='cuda:0'), in_proj_covar=tensor([0.0970, 0.1036, 0.0846, 0.1007, 0.1032, 0.0947, 0.0778, 0.0857], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 03:25:46,893 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=236412.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:25:53,562 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.624e+02 2.406e+02 2.956e+02 3.756e+02 9.982e+02, threshold=5.913e+02, percent-clipped=8.0 2023-02-09 03:26:04,554 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=236437.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:26:16,760 INFO [train.py:901] (0/4) Epoch 30, batch 2050, loss[loss=0.1881, simple_loss=0.2679, pruned_loss=0.05419, over 8079.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2808, pruned_loss=0.05651, over 1612990.03 frames. ], batch size: 21, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:26:47,589 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=236495.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:26:53,533 INFO [train.py:901] (0/4) Epoch 30, batch 2100, loss[loss=0.1941, simple_loss=0.277, pruned_loss=0.0556, over 8071.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2811, pruned_loss=0.05636, over 1617949.22 frames. ], batch size: 21, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:27:06,244 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.478e+02 2.477e+02 3.089e+02 3.892e+02 8.089e+02, threshold=6.178e+02, percent-clipped=3.0 2023-02-09 03:27:07,847 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=236524.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:27:25,249 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=236549.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:27:28,498 INFO [train.py:901] (0/4) Epoch 30, batch 2150, loss[loss=0.1924, simple_loss=0.2883, pruned_loss=0.04822, over 8499.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2806, pruned_loss=0.05603, over 1616884.01 frames. ], batch size: 28, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:27:43,268 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([5.8322, 5.9470, 5.2509, 2.5706, 5.2709, 5.6565, 5.2700, 5.5096], device='cuda:0'), covar=tensor([0.0647, 0.0373, 0.0925, 0.4341, 0.0784, 0.0874, 0.1249, 0.0535], device='cuda:0'), in_proj_covar=tensor([0.0551, 0.0461, 0.0450, 0.0564, 0.0446, 0.0473, 0.0449, 0.0413], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 03:27:57,013 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.4373, 2.3682, 1.7829, 2.1054, 1.9725, 1.5737, 1.8140, 1.8930], device='cuda:0'), covar=tensor([0.1506, 0.0406, 0.1184, 0.0617, 0.0772, 0.1504, 0.1053, 0.1066], device='cuda:0'), in_proj_covar=tensor([0.0361, 0.0245, 0.0349, 0.0317, 0.0304, 0.0353, 0.0355, 0.0325], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 03:28:04,589 INFO [train.py:901] (0/4) Epoch 30, batch 2200, loss[loss=0.168, simple_loss=0.2365, pruned_loss=0.04972, over 7241.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.2803, pruned_loss=0.05568, over 1611760.66 frames. ], batch size: 16, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:28:18,769 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.663e+02 2.435e+02 2.816e+02 3.564e+02 9.413e+02, threshold=5.632e+02, percent-clipped=3.0 2023-02-09 03:28:30,761 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([4.3249, 1.3216, 4.5482, 1.6845, 4.0209, 3.7824, 4.0983, 4.0139], device='cuda:0'), covar=tensor([0.0805, 0.5347, 0.0633, 0.4716, 0.1310, 0.1136, 0.0724, 0.0796], device='cuda:0'), in_proj_covar=tensor([0.0695, 0.0674, 0.0754, 0.0668, 0.0759, 0.0646, 0.0653, 0.0730], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 03:28:40,462 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.0502, 2.2312, 1.7870, 2.8310, 1.4212, 1.5935, 2.1260, 2.2545], device='cuda:0'), covar=tensor([0.0727, 0.0754, 0.0928, 0.0315, 0.1112, 0.1364, 0.0801, 0.0732], device='cuda:0'), in_proj_covar=tensor([0.0232, 0.0196, 0.0245, 0.0216, 0.0202, 0.0248, 0.0252, 0.0206], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], device='cuda:0') 2023-02-09 03:28:40,976 INFO [train.py:901] (0/4) Epoch 30, batch 2250, loss[loss=0.1927, simple_loss=0.2959, pruned_loss=0.04477, over 8309.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.2807, pruned_loss=0.05555, over 1617382.61 frames. ], batch size: 25, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:29:04,196 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.68 vs. limit=2.0 2023-02-09 03:29:16,927 INFO [train.py:901] (0/4) Epoch 30, batch 2300, loss[loss=0.1818, simple_loss=0.2809, pruned_loss=0.04136, over 8587.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2804, pruned_loss=0.05604, over 1613803.23 frames. ], batch size: 34, lr: 2.52e-03, grad_scale: 8.0 2023-02-09 03:29:29,253 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.819e+02 2.489e+02 3.036e+02 4.215e+02 7.962e+02, threshold=6.071e+02, percent-clipped=6.0 2023-02-09 03:29:50,634 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=236751.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:29:52,586 INFO [train.py:901] (0/4) Epoch 30, batch 2350, loss[loss=0.181, simple_loss=0.2738, pruned_loss=0.04412, over 8244.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2807, pruned_loss=0.05628, over 1614675.40 frames. ], batch size: 22, lr: 2.51e-03, grad_scale: 8.0 2023-02-09 03:30:08,618 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=236776.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:30:29,061 INFO [train.py:901] (0/4) Epoch 30, batch 2400, loss[loss=0.1693, simple_loss=0.2594, pruned_loss=0.03958, over 7807.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.28, pruned_loss=0.05622, over 1610348.22 frames. ], batch size: 20, lr: 2.51e-03, grad_scale: 8.0 2023-02-09 03:30:42,227 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.757e+02 2.281e+02 2.685e+02 3.729e+02 8.099e+02, threshold=5.371e+02, percent-clipped=9.0 2023-02-09 03:31:05,197 INFO [train.py:901] (0/4) Epoch 30, batch 2450, loss[loss=0.1892, simple_loss=0.2803, pruned_loss=0.04902, over 8198.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2805, pruned_loss=0.05669, over 1614812.19 frames. ], batch size: 23, lr: 2.51e-03, grad_scale: 8.0 2023-02-09 03:31:39,710 INFO [train.py:901] (0/4) Epoch 30, batch 2500, loss[loss=0.228, simple_loss=0.3087, pruned_loss=0.07368, over 8356.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2795, pruned_loss=0.0564, over 1612544.47 frames. ], batch size: 26, lr: 2.51e-03, grad_scale: 8.0 2023-02-09 03:31:43,945 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=236910.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 03:31:52,883 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.501e+02 2.373e+02 3.086e+02 3.767e+02 7.222e+02, threshold=6.171e+02, percent-clipped=6.0 2023-02-09 03:32:13,826 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.6575, 2.3598, 1.8584, 2.2895, 2.1930, 1.5829, 2.0645, 2.0152], device='cuda:0'), covar=tensor([0.1281, 0.0433, 0.1259, 0.0541, 0.0715, 0.1630, 0.0951, 0.0933], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0242, 0.0345, 0.0314, 0.0302, 0.0350, 0.0351, 0.0320], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 03:32:16,587 INFO [train.py:901] (0/4) Epoch 30, batch 2550, loss[loss=0.1975, simple_loss=0.2813, pruned_loss=0.05686, over 8422.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2797, pruned_loss=0.05655, over 1613259.32 frames. ], batch size: 27, lr: 2.51e-03, grad_scale: 8.0 2023-02-09 03:32:54,193 INFO [train.py:901] (0/4) Epoch 30, batch 2600, loss[loss=0.1602, simple_loss=0.2512, pruned_loss=0.03457, over 8105.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2801, pruned_loss=0.05635, over 1615964.14 frames. ], batch size: 23, lr: 2.51e-03, grad_scale: 8.0 2023-02-09 03:33:06,914 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.551e+02 2.458e+02 3.021e+02 3.974e+02 8.394e+02, threshold=6.042e+02, percent-clipped=5.0 2023-02-09 03:33:12,331 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=3.50 vs. limit=5.0 2023-02-09 03:33:30,302 INFO [train.py:901] (0/4) Epoch 30, batch 2650, loss[loss=0.1914, simple_loss=0.2805, pruned_loss=0.05113, over 8644.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2815, pruned_loss=0.05706, over 1617523.94 frames. ], batch size: 27, lr: 2.51e-03, grad_scale: 8.0 2023-02-09 03:34:06,427 INFO [train.py:901] (0/4) Epoch 30, batch 2700, loss[loss=0.1813, simple_loss=0.2535, pruned_loss=0.05461, over 7936.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2815, pruned_loss=0.0572, over 1612550.81 frames. ], batch size: 20, lr: 2.51e-03, grad_scale: 8.0 2023-02-09 03:34:07,587 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 2023-02-09 03:34:18,973 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.680e+02 2.464e+02 3.015e+02 4.068e+02 7.247e+02, threshold=6.030e+02, percent-clipped=1.0 2023-02-09 03:34:41,482 INFO [train.py:901] (0/4) Epoch 30, batch 2750, loss[loss=0.1886, simple_loss=0.2637, pruned_loss=0.05677, over 7925.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2805, pruned_loss=0.05669, over 1610995.49 frames. ], batch size: 20, lr: 2.51e-03, grad_scale: 8.0 2023-02-09 03:35:18,240 INFO [train.py:901] (0/4) Epoch 30, batch 2800, loss[loss=0.2242, simple_loss=0.3071, pruned_loss=0.07064, over 8468.00 frames. ], tot_loss[loss=0.196, simple_loss=0.2795, pruned_loss=0.05622, over 1607240.44 frames. ], batch size: 49, lr: 2.51e-03, grad_scale: 8.0 2023-02-09 03:35:20,505 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.6027, 1.3965, 1.7360, 1.2951, 0.9700, 1.4766, 1.5414, 1.3398], device='cuda:0'), covar=tensor([0.0601, 0.1289, 0.1586, 0.1492, 0.0584, 0.1494, 0.0720, 0.0711], device='cuda:0'), in_proj_covar=tensor([0.0101, 0.0155, 0.0192, 0.0163, 0.0102, 0.0165, 0.0114, 0.0149], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0008, 0.0008], device='cuda:0') 2023-02-09 03:35:31,338 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.593e+02 2.300e+02 2.824e+02 3.573e+02 8.919e+02, threshold=5.648e+02, percent-clipped=3.0 2023-02-09 03:35:45,285 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([6.0150, 1.7386, 6.1547, 2.2430, 5.5787, 5.1944, 5.6744, 5.5627], device='cuda:0'), covar=tensor([0.0495, 0.4598, 0.0369, 0.3831, 0.1004, 0.0840, 0.0523, 0.0543], device='cuda:0'), in_proj_covar=tensor([0.0694, 0.0671, 0.0752, 0.0670, 0.0756, 0.0645, 0.0655, 0.0729], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 03:35:48,282 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 2023-02-09 03:35:53,019 INFO [train.py:901] (0/4) Epoch 30, batch 2850, loss[loss=0.1942, simple_loss=0.281, pruned_loss=0.05374, over 8332.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2796, pruned_loss=0.05645, over 1606593.71 frames. ], batch size: 26, lr: 2.51e-03, grad_scale: 8.0 2023-02-09 03:35:53,086 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=237254.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 03:36:18,934 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.88 vs. limit=2.0 2023-02-09 03:36:27,598 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.57 vs. limit=2.0 2023-02-09 03:36:29,228 INFO [train.py:901] (0/4) Epoch 30, batch 2900, loss[loss=0.2138, simple_loss=0.305, pruned_loss=0.06134, over 8456.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2801, pruned_loss=0.05695, over 1608026.24 frames. ], batch size: 29, lr: 2.51e-03, grad_scale: 8.0 2023-02-09 03:36:42,593 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.704e+02 2.592e+02 3.021e+02 4.387e+02 8.419e+02, threshold=6.042e+02, percent-clipped=5.0 2023-02-09 03:37:04,020 WARNING [train.py:1067] (0/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 2023-02-09 03:37:05,370 INFO [train.py:901] (0/4) Epoch 30, batch 2950, loss[loss=0.2294, simple_loss=0.3099, pruned_loss=0.07448, over 7549.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.28, pruned_loss=0.05641, over 1609751.87 frames. ], batch size: 18, lr: 2.51e-03, grad_scale: 8.0 2023-02-09 03:37:15,785 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=237369.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 03:37:36,198 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([5.8772, 6.0635, 5.2368, 2.4435, 5.3370, 5.7064, 5.4105, 5.5050], device='cuda:0'), covar=tensor([0.0463, 0.0258, 0.0775, 0.3980, 0.0660, 0.0648, 0.0880, 0.0400], device='cuda:0'), in_proj_covar=tensor([0.0550, 0.0461, 0.0450, 0.0562, 0.0445, 0.0473, 0.0447, 0.0415], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 03:37:40,335 INFO [train.py:901] (0/4) Epoch 30, batch 3000, loss[loss=0.2191, simple_loss=0.2994, pruned_loss=0.06944, over 8732.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2806, pruned_loss=0.05684, over 1611549.79 frames. ], batch size: 40, lr: 2.51e-03, grad_scale: 8.0 2023-02-09 03:37:40,336 INFO [train.py:926] (0/4) Computing validation loss 2023-02-09 03:37:54,060 INFO [train.py:935] (0/4) Epoch 30, validation: loss=0.1704, simple_loss=0.2697, pruned_loss=0.0356, over 944034.00 frames. 2023-02-09 03:37:54,061 INFO [train.py:936] (0/4) Maximum memory allocated so far is 6641MB 2023-02-09 03:38:07,363 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.697e+02 2.369e+02 2.918e+02 3.560e+02 6.316e+02, threshold=5.836e+02, percent-clipped=1.0 2023-02-09 03:38:31,177 INFO [train.py:901] (0/4) Epoch 30, batch 3050, loss[loss=0.2181, simple_loss=0.2995, pruned_loss=0.06835, over 8527.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2806, pruned_loss=0.05682, over 1614406.21 frames. ], batch size: 28, lr: 2.51e-03, grad_scale: 8.0 2023-02-09 03:39:07,128 INFO [train.py:901] (0/4) Epoch 30, batch 3100, loss[loss=0.1846, simple_loss=0.2776, pruned_loss=0.04577, over 8479.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2809, pruned_loss=0.05636, over 1618555.37 frames. ], batch size: 29, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:39:10,835 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=237509.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:39:15,046 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.3600, 1.4252, 1.3633, 1.8380, 0.6558, 1.2817, 1.2733, 1.4887], device='cuda:0'), covar=tensor([0.0977, 0.0846, 0.1007, 0.0533, 0.1157, 0.1325, 0.0756, 0.0725], device='cuda:0'), in_proj_covar=tensor([0.0230, 0.0194, 0.0245, 0.0215, 0.0202, 0.0246, 0.0249, 0.0204], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], device='cuda:0') 2023-02-09 03:39:19,648 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.408e+02 2.429e+02 3.016e+02 3.485e+02 6.483e+02, threshold=6.032e+02, percent-clipped=4.0 2023-02-09 03:39:31,253 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.7721, 1.6754, 2.7015, 2.0637, 2.4307, 1.7977, 1.6028, 1.2677], device='cuda:0'), covar=tensor([0.8015, 0.7085, 0.2508, 0.4520, 0.3450, 0.5048, 0.3163, 0.6399], device='cuda:0'), in_proj_covar=tensor([0.0967, 0.1036, 0.0847, 0.1009, 0.1030, 0.0948, 0.0778, 0.0860], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 03:39:43,983 INFO [train.py:901] (0/4) Epoch 30, batch 3150, loss[loss=0.1635, simple_loss=0.2472, pruned_loss=0.03985, over 7800.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.28, pruned_loss=0.05569, over 1619032.72 frames. ], batch size: 19, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:40:03,671 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=237581.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 03:40:21,021 INFO [train.py:901] (0/4) Epoch 30, batch 3200, loss[loss=0.1826, simple_loss=0.2698, pruned_loss=0.04771, over 8473.00 frames. ], tot_loss[loss=0.1952, simple_loss=0.2796, pruned_loss=0.05543, over 1615685.68 frames. ], batch size: 29, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:40:33,294 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.611e+02 2.320e+02 2.861e+02 3.592e+02 8.186e+02, threshold=5.722e+02, percent-clipped=5.0 2023-02-09 03:40:33,442 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.7564, 1.5058, 3.0988, 1.4049, 2.4777, 3.3995, 3.5157, 2.8282], device='cuda:0'), covar=tensor([0.1255, 0.1869, 0.0366, 0.2265, 0.0906, 0.0309, 0.0556, 0.0706], device='cuda:0'), in_proj_covar=tensor([0.0309, 0.0329, 0.0298, 0.0328, 0.0330, 0.0283, 0.0449, 0.0311], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 03:40:35,513 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=237625.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 03:40:52,677 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=237650.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 03:40:56,035 INFO [train.py:901] (0/4) Epoch 30, batch 3250, loss[loss=0.1723, simple_loss=0.2598, pruned_loss=0.04242, over 7929.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.28, pruned_loss=0.05586, over 1612775.25 frames. ], batch size: 20, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:41:32,182 INFO [train.py:901] (0/4) Epoch 30, batch 3300, loss[loss=0.1962, simple_loss=0.2838, pruned_loss=0.05426, over 8462.00 frames. ], tot_loss[loss=0.1948, simple_loss=0.2789, pruned_loss=0.05537, over 1613520.21 frames. ], batch size: 49, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:41:45,796 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.739e+02 2.392e+02 2.907e+02 3.818e+02 6.093e+02, threshold=5.813e+02, percent-clipped=2.0 2023-02-09 03:42:03,991 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.0134, 1.6253, 1.3197, 1.5233, 1.3201, 1.2190, 1.2460, 1.2873], device='cuda:0'), covar=tensor([0.1121, 0.0510, 0.1413, 0.0571, 0.0716, 0.1560, 0.0955, 0.0812], device='cuda:0'), in_proj_covar=tensor([0.0354, 0.0242, 0.0343, 0.0314, 0.0299, 0.0347, 0.0348, 0.0319], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 03:42:07,978 INFO [train.py:901] (0/4) Epoch 30, batch 3350, loss[loss=0.1782, simple_loss=0.2682, pruned_loss=0.04415, over 8130.00 frames. ], tot_loss[loss=0.1944, simple_loss=0.2787, pruned_loss=0.05501, over 1613184.54 frames. ], batch size: 22, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:42:35,991 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.93 vs. limit=2.0 2023-02-09 03:42:40,705 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([0.9640, 1.6231, 1.3090, 1.4556, 1.2767, 1.1467, 1.1364, 1.2276], device='cuda:0'), covar=tensor([0.1416, 0.0616, 0.1577, 0.0759, 0.1001, 0.1981, 0.1280, 0.0998], device='cuda:0'), in_proj_covar=tensor([0.0356, 0.0243, 0.0344, 0.0314, 0.0300, 0.0348, 0.0349, 0.0320], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 03:42:44,185 INFO [train.py:901] (0/4) Epoch 30, batch 3400, loss[loss=0.1609, simple_loss=0.2467, pruned_loss=0.03755, over 8248.00 frames. ], tot_loss[loss=0.1945, simple_loss=0.2785, pruned_loss=0.05529, over 1614121.56 frames. ], batch size: 22, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:42:47,129 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=237808.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:42:57,408 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.758e+02 2.484e+02 3.245e+02 4.483e+02 9.283e+02, threshold=6.490e+02, percent-clipped=12.0 2023-02-09 03:43:19,448 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=237853.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:43:20,122 INFO [train.py:901] (0/4) Epoch 30, batch 3450, loss[loss=0.2164, simple_loss=0.3059, pruned_loss=0.06348, over 8103.00 frames. ], tot_loss[loss=0.1948, simple_loss=0.2788, pruned_loss=0.05543, over 1610162.67 frames. ], batch size: 23, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:43:56,091 INFO [train.py:901] (0/4) Epoch 30, batch 3500, loss[loss=0.19, simple_loss=0.2836, pruned_loss=0.04819, over 8248.00 frames. ], tot_loss[loss=0.195, simple_loss=0.2791, pruned_loss=0.05547, over 1612494.09 frames. ], batch size: 24, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:43:57,001 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.5042, 1.3965, 1.8392, 1.2246, 1.1329, 1.8150, 0.2806, 1.1662], device='cuda:0'), covar=tensor([0.1318, 0.1196, 0.0371, 0.0743, 0.2311, 0.0443, 0.1737, 0.1035], device='cuda:0'), in_proj_covar=tensor([0.0204, 0.0207, 0.0139, 0.0224, 0.0279, 0.0149, 0.0174, 0.0201], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 03:44:08,739 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.712e+02 2.505e+02 3.010e+02 3.725e+02 8.965e+02, threshold=6.019e+02, percent-clipped=4.0 2023-02-09 03:44:11,008 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=237925.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 03:44:14,972 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=237930.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:44:22,354 WARNING [train.py:1067] (0/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 2023-02-09 03:44:32,818 INFO [train.py:901] (0/4) Epoch 30, batch 3550, loss[loss=0.1828, simple_loss=0.2779, pruned_loss=0.04383, over 8506.00 frames. ], tot_loss[loss=0.1952, simple_loss=0.2797, pruned_loss=0.05538, over 1617647.16 frames. ], batch size: 28, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:44:43,175 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=237968.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:44:56,017 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.7688, 2.2935, 3.6842, 1.5662, 2.8009, 2.2381, 1.8036, 2.8550], device='cuda:0'), covar=tensor([0.1999, 0.2578, 0.1014, 0.4836, 0.1972, 0.3410, 0.2623, 0.2378], device='cuda:0'), in_proj_covar=tensor([0.0545, 0.0642, 0.0568, 0.0677, 0.0668, 0.0617, 0.0571, 0.0648], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:0') 2023-02-09 03:45:05,696 INFO [checkpoint.py:75] (0/4) Saving checkpoint to pruned_transducer_stateless7_streaming/exp/v1/checkpoint-238000.pt 2023-02-09 03:45:09,421 INFO [train.py:901] (0/4) Epoch 30, batch 3600, loss[loss=0.2271, simple_loss=0.3175, pruned_loss=0.06829, over 8499.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.2805, pruned_loss=0.05551, over 1619058.85 frames. ], batch size: 39, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:45:22,459 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.684e+02 2.286e+02 2.832e+02 3.360e+02 7.556e+02, threshold=5.664e+02, percent-clipped=2.0 2023-02-09 03:45:35,422 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238040.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 03:45:40,895 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([5.8419, 6.0660, 5.1922, 2.4631, 5.2583, 5.6974, 5.4899, 5.5048], device='cuda:0'), covar=tensor([0.0650, 0.0378, 0.0965, 0.4432, 0.0766, 0.0761, 0.1134, 0.0670], device='cuda:0'), in_proj_covar=tensor([0.0551, 0.0463, 0.0452, 0.0563, 0.0445, 0.0475, 0.0451, 0.0416], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 03:45:44,902 INFO [train.py:901] (0/4) Epoch 30, batch 3650, loss[loss=0.2574, simple_loss=0.3272, pruned_loss=0.09381, over 7279.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.2803, pruned_loss=0.05552, over 1619148.87 frames. ], batch size: 78, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:46:05,857 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238082.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:46:12,779 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238092.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:46:20,738 INFO [train.py:901] (0/4) Epoch 30, batch 3700, loss[loss=0.2346, simple_loss=0.321, pruned_loss=0.07413, over 8355.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2805, pruned_loss=0.05592, over 1615037.34 frames. ], batch size: 24, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:46:21,662 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.4945, 1.3938, 1.8343, 1.1742, 1.1071, 1.8042, 0.2364, 1.1560], device='cuda:0'), covar=tensor([0.1400, 0.1175, 0.0371, 0.0824, 0.2318, 0.0410, 0.1717, 0.1065], device='cuda:0'), in_proj_covar=tensor([0.0204, 0.0207, 0.0139, 0.0225, 0.0279, 0.0149, 0.0175, 0.0201], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 03:46:29,078 WARNING [train.py:1067] (0/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 2023-02-09 03:46:33,267 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.392e+02 2.352e+02 3.001e+02 3.686e+02 7.575e+02, threshold=6.003e+02, percent-clipped=3.0 2023-02-09 03:46:45,999 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.1812, 1.4480, 1.7390, 1.4084, 0.7422, 1.5291, 1.1660, 1.2058], device='cuda:0'), covar=tensor([0.0602, 0.1157, 0.1471, 0.1377, 0.0537, 0.1335, 0.0663, 0.0645], device='cuda:0'), in_proj_covar=tensor([0.0101, 0.0155, 0.0193, 0.0163, 0.0102, 0.0165, 0.0114, 0.0149], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0008, 0.0008], device='cuda:0') 2023-02-09 03:46:50,339 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238144.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:46:56,066 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=238152.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:46:57,373 INFO [train.py:901] (0/4) Epoch 30, batch 3750, loss[loss=0.2219, simple_loss=0.3072, pruned_loss=0.06834, over 8765.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2807, pruned_loss=0.05641, over 1615925.74 frames. ], batch size: 30, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:46:57,656 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.4308, 2.3509, 2.9914, 2.5296, 2.9241, 2.5248, 2.3755, 1.8210], device='cuda:0'), covar=tensor([0.5623, 0.5255, 0.2191, 0.4072, 0.2893, 0.3253, 0.1994, 0.5875], device='cuda:0'), in_proj_covar=tensor([0.0971, 0.1038, 0.0848, 0.1012, 0.1033, 0.0949, 0.0780, 0.0860], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 03:47:33,681 INFO [train.py:901] (0/4) Epoch 30, batch 3800, loss[loss=0.2026, simple_loss=0.286, pruned_loss=0.05962, over 7918.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2814, pruned_loss=0.05696, over 1612469.83 frames. ], batch size: 20, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:47:46,041 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.710e+02 2.427e+02 2.911e+02 3.474e+02 7.215e+02, threshold=5.821e+02, percent-clipped=2.0 2023-02-09 03:47:47,593 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238224.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:47:48,257 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.7788, 1.4433, 1.7020, 1.3485, 1.0641, 1.4630, 1.6871, 1.5041], device='cuda:0'), covar=tensor([0.0631, 0.1334, 0.1694, 0.1501, 0.0627, 0.1554, 0.0733, 0.0676], device='cuda:0'), in_proj_covar=tensor([0.0101, 0.0155, 0.0193, 0.0163, 0.0102, 0.0165, 0.0114, 0.0149], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0008, 0.0008], device='cuda:0') 2023-02-09 03:48:05,496 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238249.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:48:09,474 INFO [train.py:901] (0/4) Epoch 30, batch 3850, loss[loss=0.1917, simple_loss=0.2823, pruned_loss=0.0505, over 8192.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2824, pruned_loss=0.05742, over 1615671.95 frames. ], batch size: 23, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:48:18,890 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238267.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:48:23,697 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=238274.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:48:37,706 WARNING [train.py:1067] (0/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 2023-02-09 03:48:39,926 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238296.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 03:48:45,178 INFO [train.py:901] (0/4) Epoch 30, batch 3900, loss[loss=0.1725, simple_loss=0.2492, pruned_loss=0.04795, over 7787.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2822, pruned_loss=0.05682, over 1621772.59 frames. ], batch size: 19, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:48:57,810 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238321.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 03:48:58,293 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.463e+02 2.361e+02 2.887e+02 3.538e+02 6.169e+02, threshold=5.773e+02, percent-clipped=2.0 2023-02-09 03:49:12,284 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.52 vs. limit=2.0 2023-02-09 03:49:20,478 INFO [train.py:901] (0/4) Epoch 30, batch 3950, loss[loss=0.1946, simple_loss=0.2855, pruned_loss=0.05182, over 8322.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2817, pruned_loss=0.05668, over 1619176.25 frames. ], batch size: 25, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:49:45,699 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([3.2721, 1.4851, 3.4129, 1.2055, 3.0455, 2.8799, 3.1516, 3.0664], device='cuda:0'), covar=tensor([0.0805, 0.3704, 0.0760, 0.4123, 0.1302, 0.1040, 0.0716, 0.0855], device='cuda:0'), in_proj_covar=tensor([0.0695, 0.0670, 0.0755, 0.0667, 0.0755, 0.0644, 0.0653, 0.0728], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 03:49:46,435 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238389.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:49:56,967 INFO [train.py:901] (0/4) Epoch 30, batch 4000, loss[loss=0.1991, simple_loss=0.287, pruned_loss=0.05564, over 7978.00 frames. ], tot_loss[loss=0.197, simple_loss=0.281, pruned_loss=0.05654, over 1615781.56 frames. ], batch size: 21, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:50:08,209 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.8740, 1.9629, 1.9309, 1.6209, 2.0390, 1.7018, 1.2167, 1.9257], device='cuda:0'), covar=tensor([0.0509, 0.0365, 0.0290, 0.0514, 0.0396, 0.0694, 0.0827, 0.0312], device='cuda:0'), in_proj_covar=tensor([0.0484, 0.0424, 0.0380, 0.0470, 0.0407, 0.0564, 0.0411, 0.0450], device='cuda:0'), out_proj_covar=tensor([1.2813e-04, 1.0972e-04, 9.8935e-05, 1.2272e-04, 1.0644e-04, 1.5702e-04, 1.0953e-04, 1.1764e-04], device='cuda:0') 2023-02-09 03:50:09,962 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.496e+02 2.354e+02 2.920e+02 3.674e+02 8.815e+02, threshold=5.839e+02, percent-clipped=5.0 2023-02-09 03:50:12,702 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=238426.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:50:20,358 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=238436.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:50:21,775 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238438.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 03:50:32,675 INFO [train.py:901] (0/4) Epoch 30, batch 4050, loss[loss=0.1929, simple_loss=0.2854, pruned_loss=0.05023, over 8505.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2808, pruned_loss=0.05644, over 1616021.86 frames. ], batch size: 26, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:50:47,615 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([4.0585, 1.5911, 4.3878, 1.9620, 3.5370, 3.4464, 3.9099, 3.8893], device='cuda:0'), covar=tensor([0.1358, 0.6995, 0.1174, 0.5876, 0.2219, 0.1923, 0.1223, 0.1098], device='cuda:0'), in_proj_covar=tensor([0.0697, 0.0673, 0.0757, 0.0669, 0.0757, 0.0646, 0.0655, 0.0731], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 03:50:49,068 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=2.16 vs. limit=5.0 2023-02-09 03:50:57,402 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=238488.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:51:08,980 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-02-09 03:51:09,342 INFO [train.py:901] (0/4) Epoch 30, batch 4100, loss[loss=0.1655, simple_loss=0.2574, pruned_loss=0.03682, over 8348.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2803, pruned_loss=0.0562, over 1614098.89 frames. ], batch size: 24, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:51:21,790 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.512e+02 2.287e+02 2.925e+02 3.934e+02 1.031e+03, threshold=5.850e+02, percent-clipped=7.0 2023-02-09 03:51:22,750 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238523.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:51:35,966 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238541.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:51:41,378 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238548.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:51:43,266 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238551.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:51:45,014 INFO [train.py:901] (0/4) Epoch 30, batch 4150, loss[loss=0.2282, simple_loss=0.3147, pruned_loss=0.07084, over 8322.00 frames. ], tot_loss[loss=0.197, simple_loss=0.281, pruned_loss=0.05651, over 1616288.42 frames. ], batch size: 25, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:52:11,306 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([4.4225, 1.6885, 4.5892, 1.8184, 4.1228, 3.7819, 4.1051, 3.9935], device='cuda:0'), covar=tensor([0.0615, 0.4587, 0.0492, 0.4330, 0.0977, 0.0978, 0.0621, 0.0717], device='cuda:0'), in_proj_covar=tensor([0.0698, 0.0675, 0.0760, 0.0672, 0.0760, 0.0650, 0.0658, 0.0734], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 03:52:19,910 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238603.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:52:20,455 INFO [train.py:901] (0/4) Epoch 30, batch 4200, loss[loss=0.195, simple_loss=0.2654, pruned_loss=0.06226, over 7538.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2806, pruned_loss=0.05663, over 1616141.68 frames. ], batch size: 18, lr: 2.51e-03, grad_scale: 16.0 2023-02-09 03:52:33,715 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.657e+02 2.494e+02 3.256e+02 4.447e+02 1.288e+03, threshold=6.511e+02, percent-clipped=8.0 2023-02-09 03:52:41,508 WARNING [train.py:1067] (0/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 2023-02-09 03:52:50,657 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238645.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:52:56,587 INFO [train.py:901] (0/4) Epoch 30, batch 4250, loss[loss=0.1948, simple_loss=0.2901, pruned_loss=0.04978, over 8320.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2814, pruned_loss=0.05706, over 1622936.83 frames. ], batch size: 25, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 03:53:05,087 WARNING [train.py:1067] (0/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 2023-02-09 03:53:07,965 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238670.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:53:27,915 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238699.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:53:31,134 INFO [train.py:901] (0/4) Epoch 30, batch 4300, loss[loss=0.2745, simple_loss=0.3333, pruned_loss=0.1079, over 8617.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2806, pruned_loss=0.05696, over 1621512.84 frames. ], batch size: 31, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 03:53:44,805 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.604e+02 2.303e+02 2.743e+02 3.342e+02 6.438e+02, threshold=5.486e+02, percent-clipped=0.0 2023-02-09 03:54:06,887 INFO [train.py:901] (0/4) Epoch 30, batch 4350, loss[loss=0.1982, simple_loss=0.2894, pruned_loss=0.05354, over 8462.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2804, pruned_loss=0.05662, over 1622870.72 frames. ], batch size: 27, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 03:54:27,363 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=238782.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 03:54:36,398 WARNING [train.py:1067] (0/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 2023-02-09 03:54:37,987 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238797.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:54:42,555 INFO [train.py:901] (0/4) Epoch 30, batch 4400, loss[loss=0.2368, simple_loss=0.3177, pruned_loss=0.07796, over 8613.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2805, pruned_loss=0.05638, over 1621789.99 frames. ], batch size: 49, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 03:54:44,953 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238807.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:54:55,878 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.538e+02 2.573e+02 3.023e+02 3.983e+02 6.680e+02, threshold=6.046e+02, percent-clipped=2.0 2023-02-09 03:54:56,101 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238822.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:55:03,128 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238832.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:55:11,355 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.5269, 1.8455, 2.6670, 1.4934, 1.9359, 1.9563, 1.6229, 2.0488], device='cuda:0'), covar=tensor([0.1959, 0.2884, 0.1027, 0.4800, 0.2089, 0.3302, 0.2658, 0.2265], device='cuda:0'), in_proj_covar=tensor([0.0549, 0.0647, 0.0571, 0.0681, 0.0673, 0.0620, 0.0574, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:0') 2023-02-09 03:55:15,276 WARNING [train.py:1067] (0/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 2023-02-09 03:55:18,603 INFO [train.py:901] (0/4) Epoch 30, batch 4450, loss[loss=0.2153, simple_loss=0.2995, pruned_loss=0.06552, over 8194.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2817, pruned_loss=0.05667, over 1624724.27 frames. ], batch size: 23, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 03:55:18,810 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.4105, 2.7049, 2.2149, 3.8600, 1.6787, 2.1006, 2.4816, 2.6318], device='cuda:0'), covar=tensor([0.0722, 0.0788, 0.0842, 0.0271, 0.1081, 0.1178, 0.0944, 0.0894], device='cuda:0'), in_proj_covar=tensor([0.0233, 0.0195, 0.0246, 0.0215, 0.0203, 0.0248, 0.0250, 0.0206], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], device='cuda:0') 2023-02-09 03:55:22,519 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238859.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:55:40,886 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238884.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:55:50,458 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238897.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 03:55:55,020 INFO [train.py:901] (0/4) Epoch 30, batch 4500, loss[loss=0.2059, simple_loss=0.2804, pruned_loss=0.06566, over 8636.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.282, pruned_loss=0.05743, over 1616014.79 frames. ], batch size: 34, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 03:56:07,440 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.647e+02 2.335e+02 2.828e+02 3.474e+02 8.376e+02, threshold=5.656e+02, percent-clipped=3.0 2023-02-09 03:56:08,204 WARNING [train.py:1067] (0/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 2023-02-09 03:56:13,481 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=3.80 vs. limit=5.0 2023-02-09 03:56:30,985 INFO [train.py:901] (0/4) Epoch 30, batch 4550, loss[loss=0.1779, simple_loss=0.2642, pruned_loss=0.04586, over 8244.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2833, pruned_loss=0.05745, over 1618388.74 frames. ], batch size: 22, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 03:56:31,145 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238954.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 03:57:04,986 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.3302, 2.5069, 2.1409, 2.9067, 2.0551, 2.1385, 2.3347, 2.5417], device='cuda:0'), covar=tensor([0.0617, 0.0666, 0.0697, 0.0477, 0.0824, 0.0937, 0.0651, 0.0603], device='cuda:0'), in_proj_covar=tensor([0.0232, 0.0195, 0.0246, 0.0215, 0.0203, 0.0247, 0.0251, 0.0206], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], device='cuda:0') 2023-02-09 03:57:06,050 INFO [train.py:901] (0/4) Epoch 30, batch 4600, loss[loss=0.202, simple_loss=0.2791, pruned_loss=0.06244, over 7814.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.2837, pruned_loss=0.0579, over 1618426.68 frames. ], batch size: 20, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 03:57:09,651 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.5524, 1.8304, 1.8616, 1.2853, 1.9588, 1.3926, 0.4692, 1.7581], device='cuda:0'), covar=tensor([0.0718, 0.0463, 0.0345, 0.0687, 0.0503, 0.1229, 0.1126, 0.0359], device='cuda:0'), in_proj_covar=tensor([0.0483, 0.0421, 0.0379, 0.0468, 0.0403, 0.0562, 0.0408, 0.0450], device='cuda:0'), out_proj_covar=tensor([1.2768e-04, 1.0874e-04, 9.8625e-05, 1.2225e-04, 1.0537e-04, 1.5617e-04, 1.0883e-04, 1.1764e-04], device='cuda:0') 2023-02-09 03:57:19,156 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.552e+02 2.347e+02 2.832e+02 3.443e+02 5.144e+02, threshold=5.665e+02, percent-clipped=0.0 2023-02-09 03:57:34,216 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=239043.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:57:41,235 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.3116, 2.0913, 1.6625, 2.0456, 1.7298, 1.4090, 1.6878, 1.7573], device='cuda:0'), covar=tensor([0.1405, 0.0484, 0.1421, 0.0560, 0.0845, 0.1783, 0.1090, 0.0932], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0247, 0.0347, 0.0317, 0.0304, 0.0351, 0.0352, 0.0322], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 03:57:41,732 INFO [train.py:901] (0/4) Epoch 30, batch 4650, loss[loss=0.1749, simple_loss=0.2712, pruned_loss=0.03935, over 8188.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.285, pruned_loss=0.05864, over 1615969.62 frames. ], batch size: 23, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 03:57:43,630 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-02-09 03:57:47,058 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.78 vs. limit=2.0 2023-02-09 03:57:52,089 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.1778, 2.3195, 1.8647, 2.8788, 1.3582, 1.6882, 2.1708, 2.3730], device='cuda:0'), covar=tensor([0.0616, 0.0714, 0.0789, 0.0331, 0.1012, 0.1188, 0.0795, 0.0631], device='cuda:0'), in_proj_covar=tensor([0.0231, 0.0194, 0.0244, 0.0214, 0.0203, 0.0246, 0.0249, 0.0204], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], device='cuda:0') 2023-02-09 03:58:17,765 INFO [train.py:901] (0/4) Epoch 30, batch 4700, loss[loss=0.2061, simple_loss=0.2888, pruned_loss=0.06168, over 8280.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.284, pruned_loss=0.05828, over 1616589.96 frames. ], batch size: 23, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 03:58:30,953 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.697e+02 2.353e+02 2.866e+02 3.941e+02 8.957e+02, threshold=5.733e+02, percent-clipped=8.0 2023-02-09 03:58:52,470 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=239153.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 03:58:52,926 INFO [train.py:901] (0/4) Epoch 30, batch 4750, loss[loss=0.2249, simple_loss=0.3131, pruned_loss=0.06839, over 8513.00 frames. ], tot_loss[loss=0.2011, simple_loss=0.2846, pruned_loss=0.05882, over 1620103.88 frames. ], batch size: 49, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 03:58:55,963 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=239158.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 03:59:10,887 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=239178.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 03:59:12,057 WARNING [train.py:1067] (0/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 2023-02-09 03:59:14,175 WARNING [train.py:1067] (0/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 2023-02-09 03:59:28,649 INFO [train.py:901] (0/4) Epoch 30, batch 4800, loss[loss=0.2203, simple_loss=0.3034, pruned_loss=0.06855, over 8037.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2843, pruned_loss=0.05838, over 1621861.16 frames. ], batch size: 22, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 03:59:41,693 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.637e+02 2.393e+02 3.010e+02 3.751e+02 7.640e+02, threshold=6.020e+02, percent-clipped=2.0 2023-02-09 04:00:04,582 INFO [train.py:901] (0/4) Epoch 30, batch 4850, loss[loss=0.1998, simple_loss=0.2838, pruned_loss=0.05793, over 8499.00 frames. ], tot_loss[loss=0.1998, simple_loss=0.284, pruned_loss=0.05782, over 1619725.03 frames. ], batch size: 28, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 04:00:06,728 WARNING [train.py:1067] (0/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 2023-02-09 04:00:36,497 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=239298.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 04:00:40,440 INFO [train.py:901] (0/4) Epoch 30, batch 4900, loss[loss=0.2152, simple_loss=0.2918, pruned_loss=0.06934, over 8605.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2842, pruned_loss=0.05796, over 1620357.71 frames. ], batch size: 31, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 04:00:53,055 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.481e+02 2.412e+02 2.818e+02 3.519e+02 1.028e+03, threshold=5.635e+02, percent-clipped=4.0 2023-02-09 04:00:59,134 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-02-09 04:01:15,935 INFO [train.py:901] (0/4) Epoch 30, batch 4950, loss[loss=0.2206, simple_loss=0.3098, pruned_loss=0.06569, over 8347.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2832, pruned_loss=0.05771, over 1621328.00 frames. ], batch size: 25, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 04:01:51,710 INFO [train.py:901] (0/4) Epoch 30, batch 5000, loss[loss=0.2138, simple_loss=0.2962, pruned_loss=0.06566, over 7985.00 frames. ], tot_loss[loss=0.2007, simple_loss=0.2842, pruned_loss=0.05861, over 1617263.14 frames. ], batch size: 21, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 04:01:58,133 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=239413.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 04:01:58,923 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=239414.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:02:05,006 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.597e+02 2.513e+02 3.095e+02 3.810e+02 1.179e+03, threshold=6.190e+02, percent-clipped=9.0 2023-02-09 04:02:17,713 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=239439.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:02:29,049 INFO [train.py:901] (0/4) Epoch 30, batch 5050, loss[loss=0.1865, simple_loss=0.2533, pruned_loss=0.0598, over 7690.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2826, pruned_loss=0.05793, over 1614418.64 frames. ], batch size: 18, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 04:02:48,703 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.7304, 2.6137, 1.8918, 2.4601, 2.2640, 1.6293, 2.2536, 2.2942], device='cuda:0'), covar=tensor([0.1540, 0.0456, 0.1317, 0.0687, 0.0802, 0.1625, 0.0980, 0.0969], device='cuda:0'), in_proj_covar=tensor([0.0360, 0.0246, 0.0346, 0.0316, 0.0303, 0.0348, 0.0350, 0.0322], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 04:02:52,733 WARNING [train.py:1067] (0/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 2023-02-09 04:03:05,858 INFO [train.py:901] (0/4) Epoch 30, batch 5100, loss[loss=0.2029, simple_loss=0.2803, pruned_loss=0.0628, over 7531.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2817, pruned_loss=0.05716, over 1613945.50 frames. ], batch size: 18, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 04:03:20,038 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.557e+02 2.525e+02 3.230e+02 3.994e+02 1.175e+03, threshold=6.461e+02, percent-clipped=6.0 2023-02-09 04:03:27,188 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.5782, 1.7662, 2.1709, 1.4655, 1.6052, 1.8024, 1.6806, 1.5654], device='cuda:0'), covar=tensor([0.1925, 0.2588, 0.0955, 0.4532, 0.1954, 0.3480, 0.2415, 0.2159], device='cuda:0'), in_proj_covar=tensor([0.0545, 0.0645, 0.0568, 0.0680, 0.0669, 0.0619, 0.0571, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:0') 2023-02-09 04:03:42,222 INFO [train.py:901] (0/4) Epoch 30, batch 5150, loss[loss=0.1851, simple_loss=0.265, pruned_loss=0.05256, over 7649.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.282, pruned_loss=0.05754, over 1611879.85 frames. ], batch size: 19, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 04:04:12,495 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.0238, 1.5340, 1.8493, 1.4445, 1.0074, 1.5748, 1.7809, 1.5646], device='cuda:0'), covar=tensor([0.0592, 0.1241, 0.1611, 0.1422, 0.0620, 0.1415, 0.0687, 0.0674], device='cuda:0'), in_proj_covar=tensor([0.0101, 0.0155, 0.0192, 0.0163, 0.0102, 0.0165, 0.0114, 0.0149], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0008, 0.0008], device='cuda:0') 2023-02-09 04:04:18,739 INFO [train.py:901] (0/4) Epoch 30, batch 5200, loss[loss=0.1715, simple_loss=0.2556, pruned_loss=0.04371, over 7714.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2823, pruned_loss=0.05751, over 1608897.61 frames. ], batch size: 18, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 04:04:31,920 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.498e+02 2.360e+02 2.794e+02 3.430e+02 1.458e+03, threshold=5.587e+02, percent-clipped=2.0 2023-02-09 04:04:44,694 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=239640.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:04:54,718 INFO [train.py:901] (0/4) Epoch 30, batch 5250, loss[loss=0.2335, simple_loss=0.309, pruned_loss=0.07901, over 8506.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2818, pruned_loss=0.0571, over 1605094.33 frames. ], batch size: 49, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 04:04:56,133 WARNING [train.py:1067] (0/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 2023-02-09 04:05:05,211 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=239669.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 04:05:07,208 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.8868, 1.4030, 3.4899, 1.6036, 2.6044, 3.7883, 3.8965, 3.2898], device='cuda:0'), covar=tensor([0.1281, 0.2004, 0.0281, 0.2034, 0.0944, 0.0213, 0.0461, 0.0488], device='cuda:0'), in_proj_covar=tensor([0.0310, 0.0329, 0.0297, 0.0329, 0.0331, 0.0284, 0.0451, 0.0311], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 04:05:09,931 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([4.5703, 4.5905, 4.1255, 2.2682, 4.0198, 4.2254, 4.1512, 4.0960], device='cuda:0'), covar=tensor([0.0669, 0.0500, 0.1014, 0.4258, 0.0880, 0.1024, 0.1200, 0.0726], device='cuda:0'), in_proj_covar=tensor([0.0556, 0.0465, 0.0455, 0.0569, 0.0450, 0.0478, 0.0453, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 04:05:23,284 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=239694.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 04:05:29,843 INFO [train.py:901] (0/4) Epoch 30, batch 5300, loss[loss=0.1962, simple_loss=0.2887, pruned_loss=0.05192, over 8675.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2818, pruned_loss=0.05678, over 1604859.99 frames. ], batch size: 34, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 04:05:32,367 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 2023-02-09 04:05:43,734 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.572e+02 2.433e+02 2.937e+02 3.850e+02 7.663e+02, threshold=5.875e+02, percent-clipped=5.0 2023-02-09 04:06:04,875 INFO [train.py:901] (0/4) Epoch 30, batch 5350, loss[loss=0.1557, simple_loss=0.2437, pruned_loss=0.03385, over 7459.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2832, pruned_loss=0.05776, over 1610947.73 frames. ], batch size: 17, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 04:06:09,871 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.3142, 1.4518, 4.6537, 2.2860, 2.6634, 5.2752, 5.3445, 4.6383], device='cuda:0'), covar=tensor([0.1252, 0.2158, 0.0214, 0.1797, 0.1092, 0.0154, 0.0477, 0.0515], device='cuda:0'), in_proj_covar=tensor([0.0311, 0.0330, 0.0297, 0.0330, 0.0332, 0.0284, 0.0453, 0.0311], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 04:06:41,526 INFO [train.py:901] (0/4) Epoch 30, batch 5400, loss[loss=0.1883, simple_loss=0.2787, pruned_loss=0.04895, over 8649.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2831, pruned_loss=0.05792, over 1611206.72 frames. ], batch size: 34, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 04:06:43,090 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.8632, 1.6662, 2.5077, 1.6434, 1.3825, 2.4617, 0.5059, 1.5625], device='cuda:0'), covar=tensor([0.1309, 0.1159, 0.0409, 0.1020, 0.2189, 0.0393, 0.2094, 0.1164], device='cuda:0'), in_proj_covar=tensor([0.0204, 0.0207, 0.0138, 0.0224, 0.0280, 0.0149, 0.0176, 0.0202], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 04:06:55,033 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.816e+02 2.371e+02 2.881e+02 3.522e+02 8.420e+02, threshold=5.763e+02, percent-clipped=7.0 2023-02-09 04:06:59,886 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 2023-02-09 04:07:17,636 INFO [train.py:901] (0/4) Epoch 30, batch 5450, loss[loss=0.1582, simple_loss=0.2409, pruned_loss=0.03774, over 7694.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2829, pruned_loss=0.05765, over 1615932.82 frames. ], batch size: 18, lr: 2.50e-03, grad_scale: 16.0 2023-02-09 04:07:49,860 WARNING [train.py:1067] (0/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 2023-02-09 04:07:53,962 INFO [train.py:901] (0/4) Epoch 30, batch 5500, loss[loss=0.1565, simple_loss=0.2544, pruned_loss=0.02934, over 8507.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2827, pruned_loss=0.05734, over 1617039.42 frames. ], batch size: 26, lr: 2.50e-03, grad_scale: 8.0 2023-02-09 04:08:08,771 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.536e+02 2.378e+02 3.012e+02 4.037e+02 9.246e+02, threshold=6.023e+02, percent-clipped=5.0 2023-02-09 04:08:30,325 INFO [train.py:901] (0/4) Epoch 30, batch 5550, loss[loss=0.1765, simple_loss=0.2618, pruned_loss=0.04564, over 8031.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2827, pruned_loss=0.05737, over 1615188.12 frames. ], batch size: 22, lr: 2.50e-03, grad_scale: 8.0 2023-02-09 04:08:50,819 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=239984.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:09:02,725 INFO [checkpoint.py:75] (0/4) Saving checkpoint to pruned_transducer_stateless7_streaming/exp/v1/checkpoint-240000.pt 2023-02-09 04:09:07,150 INFO [train.py:901] (0/4) Epoch 30, batch 5600, loss[loss=0.1787, simple_loss=0.262, pruned_loss=0.04774, over 7931.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2822, pruned_loss=0.05706, over 1613533.10 frames. ], batch size: 20, lr: 2.50e-03, grad_scale: 8.0 2023-02-09 04:09:15,199 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 2023-02-09 04:09:21,043 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.553e+02 2.479e+02 2.939e+02 3.472e+02 8.474e+02, threshold=5.878e+02, percent-clipped=2.0 2023-02-09 04:09:21,981 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.7995, 1.6835, 2.1671, 1.3943, 1.3946, 2.0633, 0.3297, 1.3638], device='cuda:0'), covar=tensor([0.1124, 0.1018, 0.0332, 0.0828, 0.1944, 0.0450, 0.1709, 0.1205], device='cuda:0'), in_proj_covar=tensor([0.0204, 0.0206, 0.0138, 0.0224, 0.0279, 0.0149, 0.0175, 0.0202], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 04:09:42,594 INFO [train.py:901] (0/4) Epoch 30, batch 5650, loss[loss=0.1726, simple_loss=0.2556, pruned_loss=0.04482, over 8096.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2818, pruned_loss=0.05717, over 1611902.04 frames. ], batch size: 21, lr: 2.50e-03, grad_scale: 8.0 2023-02-09 04:09:44,739 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([5.9330, 6.2015, 5.4875, 2.8059, 5.5223, 5.8396, 5.6441, 5.8126], device='cuda:0'), covar=tensor([0.0640, 0.0366, 0.0890, 0.4272, 0.0770, 0.0689, 0.1033, 0.0521], device='cuda:0'), in_proj_covar=tensor([0.0558, 0.0466, 0.0456, 0.0572, 0.0453, 0.0480, 0.0456, 0.0419], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 04:09:59,638 WARNING [train.py:1067] (0/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 2023-02-09 04:10:14,064 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=240099.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:10:17,325 INFO [train.py:901] (0/4) Epoch 30, batch 5700, loss[loss=0.2179, simple_loss=0.2955, pruned_loss=0.07015, over 7919.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2815, pruned_loss=0.05725, over 1613377.69 frames. ], batch size: 20, lr: 2.50e-03, grad_scale: 8.0 2023-02-09 04:10:20,770 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.4495, 2.8418, 2.3499, 4.1239, 1.5418, 2.0853, 2.5841, 2.7046], device='cuda:0'), covar=tensor([0.0817, 0.0807, 0.0896, 0.0246, 0.1142, 0.1262, 0.0902, 0.0848], device='cuda:0'), in_proj_covar=tensor([0.0230, 0.0193, 0.0243, 0.0213, 0.0201, 0.0245, 0.0247, 0.0203], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], device='cuda:0') 2023-02-09 04:10:32,009 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.343e+02 2.506e+02 3.162e+02 4.194e+02 1.225e+03, threshold=6.325e+02, percent-clipped=8.0 2023-02-09 04:10:53,043 INFO [train.py:901] (0/4) Epoch 30, batch 5750, loss[loss=0.2035, simple_loss=0.2827, pruned_loss=0.06213, over 7974.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.282, pruned_loss=0.05759, over 1617643.34 frames. ], batch size: 21, lr: 2.50e-03, grad_scale: 8.0 2023-02-09 04:11:04,197 WARNING [train.py:1067] (0/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 2023-02-09 04:11:21,805 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.2912, 3.3446, 2.3786, 2.8542, 2.6306, 2.2258, 2.5845, 2.9498], device='cuda:0'), covar=tensor([0.1432, 0.0381, 0.1088, 0.0639, 0.0721, 0.1327, 0.0975, 0.0988], device='cuda:0'), in_proj_covar=tensor([0.0357, 0.0245, 0.0344, 0.0314, 0.0301, 0.0346, 0.0349, 0.0318], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 04:11:28,592 INFO [train.py:901] (0/4) Epoch 30, batch 5800, loss[loss=0.1569, simple_loss=0.2391, pruned_loss=0.03735, over 7548.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2812, pruned_loss=0.05733, over 1616714.26 frames. ], batch size: 18, lr: 2.50e-03, grad_scale: 8.0 2023-02-09 04:11:35,160 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=2.84 vs. limit=5.0 2023-02-09 04:11:39,945 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.4506, 2.7569, 2.1749, 3.9227, 1.6435, 1.9861, 2.4649, 2.6724], device='cuda:0'), covar=tensor([0.0660, 0.0763, 0.0766, 0.0229, 0.1041, 0.1176, 0.0849, 0.0755], device='cuda:0'), in_proj_covar=tensor([0.0230, 0.0194, 0.0244, 0.0213, 0.0201, 0.0245, 0.0247, 0.0203], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], device='cuda:0') 2023-02-09 04:11:42,453 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.461e+02 2.400e+02 2.667e+02 3.487e+02 8.848e+02, threshold=5.334e+02, percent-clipped=2.0 2023-02-09 04:12:04,254 INFO [train.py:901] (0/4) Epoch 30, batch 5850, loss[loss=0.1998, simple_loss=0.2893, pruned_loss=0.05516, over 8455.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.2796, pruned_loss=0.05614, over 1614462.56 frames. ], batch size: 27, lr: 2.50e-03, grad_scale: 8.0 2023-02-09 04:12:33,887 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.8867, 2.1861, 3.6173, 1.9044, 1.8993, 3.5346, 0.8307, 2.2529], device='cuda:0'), covar=tensor([0.1106, 0.1227, 0.0233, 0.1447, 0.2223, 0.0300, 0.1967, 0.1209], device='cuda:0'), in_proj_covar=tensor([0.0205, 0.0208, 0.0139, 0.0225, 0.0281, 0.0150, 0.0176, 0.0203], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 04:12:39,820 INFO [train.py:901] (0/4) Epoch 30, batch 5900, loss[loss=0.208, simple_loss=0.2904, pruned_loss=0.06287, over 8323.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2803, pruned_loss=0.05648, over 1613305.04 frames. ], batch size: 25, lr: 2.50e-03, grad_scale: 8.0 2023-02-09 04:12:52,667 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=2.86 vs. limit=5.0 2023-02-09 04:12:53,716 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.373e+02 2.330e+02 2.970e+02 3.920e+02 1.059e+03, threshold=5.939e+02, percent-clipped=6.0 2023-02-09 04:13:15,508 INFO [train.py:901] (0/4) Epoch 30, batch 5950, loss[loss=0.2329, simple_loss=0.3108, pruned_loss=0.07746, over 8444.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2803, pruned_loss=0.05631, over 1613540.00 frames. ], batch size: 27, lr: 2.50e-03, grad_scale: 8.0 2023-02-09 04:13:16,436 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=240355.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:13:18,461 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=240358.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:13:33,813 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=240380.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:13:50,597 INFO [train.py:901] (0/4) Epoch 30, batch 6000, loss[loss=0.1747, simple_loss=0.26, pruned_loss=0.0447, over 8034.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2804, pruned_loss=0.05628, over 1614785.18 frames. ], batch size: 22, lr: 2.50e-03, grad_scale: 8.0 2023-02-09 04:13:50,598 INFO [train.py:926] (0/4) Computing validation loss 2023-02-09 04:14:04,295 INFO [train.py:935] (0/4) Epoch 30, validation: loss=0.1701, simple_loss=0.2695, pruned_loss=0.03536, over 944034.00 frames. 2023-02-09 04:14:04,296 INFO [train.py:936] (0/4) Maximum memory allocated so far is 6641MB 2023-02-09 04:14:17,955 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.732e+02 2.377e+02 3.122e+02 3.554e+02 6.850e+02, threshold=6.243e+02, percent-clipped=2.0 2023-02-09 04:14:29,600 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([3.2262, 3.1078, 2.9233, 1.6731, 2.8112, 2.9317, 2.7988, 2.8134], device='cuda:0'), covar=tensor([0.1154, 0.0803, 0.1242, 0.4299, 0.1290, 0.1324, 0.1709, 0.1022], device='cuda:0'), in_proj_covar=tensor([0.0556, 0.0464, 0.0455, 0.0568, 0.0451, 0.0479, 0.0453, 0.0417], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 04:14:29,703 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.4059, 1.4855, 1.3513, 1.7625, 0.7112, 1.2594, 1.2877, 1.4817], device='cuda:0'), covar=tensor([0.0852, 0.0766, 0.0981, 0.0493, 0.1120, 0.1327, 0.0768, 0.0745], device='cuda:0'), in_proj_covar=tensor([0.0231, 0.0194, 0.0244, 0.0214, 0.0202, 0.0246, 0.0248, 0.0203], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], device='cuda:0') 2023-02-09 04:14:39,922 INFO [train.py:901] (0/4) Epoch 30, batch 6050, loss[loss=0.1992, simple_loss=0.2776, pruned_loss=0.0604, over 8086.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2812, pruned_loss=0.05652, over 1618848.16 frames. ], batch size: 21, lr: 2.50e-03, grad_scale: 8.0 2023-02-09 04:15:16,413 INFO [train.py:901] (0/4) Epoch 30, batch 6100, loss[loss=0.1644, simple_loss=0.2542, pruned_loss=0.03733, over 8104.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2813, pruned_loss=0.05638, over 1619769.23 frames. ], batch size: 23, lr: 2.50e-03, grad_scale: 8.0 2023-02-09 04:15:30,262 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.619e+02 2.423e+02 2.992e+02 3.767e+02 7.583e+02, threshold=5.983e+02, percent-clipped=4.0 2023-02-09 04:15:38,795 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.2124, 1.0458, 1.3049, 1.0218, 0.9789, 1.2801, 0.1294, 0.9607], device='cuda:0'), covar=tensor([0.1461, 0.1330, 0.0504, 0.0630, 0.2353, 0.0564, 0.1859, 0.1062], device='cuda:0'), in_proj_covar=tensor([0.0205, 0.0207, 0.0139, 0.0224, 0.0280, 0.0149, 0.0175, 0.0202], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 04:15:40,577 WARNING [train.py:1067] (0/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 2023-02-09 04:15:51,551 INFO [train.py:901] (0/4) Epoch 30, batch 6150, loss[loss=0.1554, simple_loss=0.2268, pruned_loss=0.04194, over 7536.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2813, pruned_loss=0.05634, over 1621960.83 frames. ], batch size: 18, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:16:28,475 INFO [train.py:901] (0/4) Epoch 30, batch 6200, loss[loss=0.2063, simple_loss=0.2744, pruned_loss=0.06913, over 7654.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.281, pruned_loss=0.05627, over 1620316.36 frames. ], batch size: 19, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:16:44,278 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.619e+02 2.533e+02 2.968e+02 3.901e+02 6.917e+02, threshold=5.935e+02, percent-clipped=4.0 2023-02-09 04:16:58,423 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.2286, 2.0045, 2.6286, 2.2474, 2.6961, 2.3272, 2.2072, 1.6493], device='cuda:0'), covar=tensor([0.6129, 0.5663, 0.2464, 0.4353, 0.2871, 0.3653, 0.2032, 0.5807], device='cuda:0'), in_proj_covar=tensor([0.0970, 0.1037, 0.0852, 0.1010, 0.1033, 0.0947, 0.0779, 0.0858], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 04:17:01,144 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.0755, 2.2803, 1.9065, 2.8856, 1.3662, 1.8010, 2.1214, 2.2552], device='cuda:0'), covar=tensor([0.0712, 0.0711, 0.0831, 0.0330, 0.1082, 0.1185, 0.0829, 0.0734], device='cuda:0'), in_proj_covar=tensor([0.0232, 0.0194, 0.0245, 0.0214, 0.0203, 0.0247, 0.0248, 0.0205], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], device='cuda:0') 2023-02-09 04:17:05,856 INFO [train.py:901] (0/4) Epoch 30, batch 6250, loss[loss=0.218, simple_loss=0.3021, pruned_loss=0.06696, over 8498.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.2799, pruned_loss=0.05575, over 1620593.53 frames. ], batch size: 28, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:17:26,883 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([3.2720, 3.2049, 2.9396, 1.7279, 2.8540, 2.9502, 2.7816, 2.8207], device='cuda:0'), covar=tensor([0.1027, 0.0736, 0.1148, 0.3931, 0.1145, 0.1194, 0.1504, 0.0991], device='cuda:0'), in_proj_covar=tensor([0.0556, 0.0465, 0.0455, 0.0568, 0.0451, 0.0480, 0.0451, 0.0418], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 04:17:39,922 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=240702.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:17:41,233 INFO [train.py:901] (0/4) Epoch 30, batch 6300, loss[loss=0.2136, simple_loss=0.3027, pruned_loss=0.06229, over 8035.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2808, pruned_loss=0.05611, over 1617510.13 frames. ], batch size: 22, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:17:54,934 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.808e+02 2.569e+02 3.101e+02 4.376e+02 1.063e+03, threshold=6.203e+02, percent-clipped=9.0 2023-02-09 04:18:17,094 INFO [train.py:901] (0/4) Epoch 30, batch 6350, loss[loss=0.1506, simple_loss=0.2377, pruned_loss=0.0318, over 7802.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.2797, pruned_loss=0.05586, over 1616166.45 frames. ], batch size: 19, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:18:53,328 INFO [train.py:901] (0/4) Epoch 30, batch 6400, loss[loss=0.2055, simple_loss=0.2947, pruned_loss=0.05819, over 8613.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.2797, pruned_loss=0.05597, over 1608919.95 frames. ], batch size: 34, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:19:02,609 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=240817.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:19:07,065 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.453e+02 2.422e+02 2.800e+02 3.642e+02 5.918e+02, threshold=5.600e+02, percent-clipped=0.0 2023-02-09 04:19:28,687 INFO [train.py:901] (0/4) Epoch 30, batch 6450, loss[loss=0.1955, simple_loss=0.2909, pruned_loss=0.05001, over 8240.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.2797, pruned_loss=0.05592, over 1613119.12 frames. ], batch size: 24, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:19:33,400 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.0338, 1.3229, 3.4877, 1.6750, 2.5026, 3.8481, 4.0302, 3.3386], device='cuda:0'), covar=tensor([0.1149, 0.2003, 0.0279, 0.1925, 0.1027, 0.0211, 0.0442, 0.0495], device='cuda:0'), in_proj_covar=tensor([0.0308, 0.0329, 0.0295, 0.0328, 0.0329, 0.0283, 0.0450, 0.0308], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 04:20:03,777 INFO [train.py:901] (0/4) Epoch 30, batch 6500, loss[loss=0.1888, simple_loss=0.2842, pruned_loss=0.04667, over 8350.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.28, pruned_loss=0.05587, over 1616188.71 frames. ], batch size: 26, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:20:17,923 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.512e+02 2.582e+02 3.161e+02 3.840e+02 1.025e+03, threshold=6.322e+02, percent-clipped=7.0 2023-02-09 04:20:24,408 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.7075, 1.5771, 1.9053, 1.5589, 1.1230, 1.6901, 2.2240, 1.9365], device='cuda:0'), covar=tensor([0.0517, 0.1278, 0.1622, 0.1452, 0.0612, 0.1441, 0.0675, 0.0651], device='cuda:0'), in_proj_covar=tensor([0.0101, 0.0154, 0.0191, 0.0162, 0.0102, 0.0164, 0.0113, 0.0148], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0007, 0.0008], device='cuda:0') 2023-02-09 04:20:36,288 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=240950.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 04:20:38,871 INFO [train.py:901] (0/4) Epoch 30, batch 6550, loss[loss=0.1846, simple_loss=0.2598, pruned_loss=0.05469, over 7245.00 frames. ], tot_loss[loss=0.1952, simple_loss=0.2789, pruned_loss=0.05575, over 1609999.80 frames. ], batch size: 16, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:21:00,578 WARNING [train.py:1067] (0/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 2023-02-09 04:21:15,891 INFO [train.py:901] (0/4) Epoch 30, batch 6600, loss[loss=0.1919, simple_loss=0.2855, pruned_loss=0.04917, over 8583.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2803, pruned_loss=0.05621, over 1613655.08 frames. ], batch size: 39, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:21:16,038 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=241004.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:21:20,066 WARNING [train.py:1067] (0/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 2023-02-09 04:21:29,847 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.473e+02 2.233e+02 3.026e+02 3.930e+02 1.368e+03, threshold=6.053e+02, percent-clipped=4.0 2023-02-09 04:21:45,179 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([4.1935, 1.5310, 4.3732, 1.6733, 3.9083, 3.6316, 3.9550, 3.8905], device='cuda:0'), covar=tensor([0.0704, 0.4504, 0.0543, 0.4375, 0.1052, 0.1089, 0.0627, 0.0689], device='cuda:0'), in_proj_covar=tensor([0.0704, 0.0671, 0.0757, 0.0676, 0.0760, 0.0649, 0.0658, 0.0735], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 04:21:51,501 INFO [train.py:901] (0/4) Epoch 30, batch 6650, loss[loss=0.2216, simple_loss=0.2917, pruned_loss=0.07571, over 8234.00 frames. ], tot_loss[loss=0.1956, simple_loss=0.2796, pruned_loss=0.05584, over 1609782.68 frames. ], batch size: 22, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:21:58,701 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.3520, 3.5420, 2.3212, 3.1787, 3.0676, 2.0436, 2.9109, 3.0457], device='cuda:0'), covar=tensor([0.1703, 0.0451, 0.1314, 0.0653, 0.0678, 0.1656, 0.1003, 0.1150], device='cuda:0'), in_proj_covar=tensor([0.0364, 0.0248, 0.0349, 0.0318, 0.0305, 0.0352, 0.0353, 0.0323], device='cuda:0'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], device='cuda:0') 2023-02-09 04:22:04,833 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=241073.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:22:23,575 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=241098.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:22:27,626 INFO [train.py:901] (0/4) Epoch 30, batch 6700, loss[loss=0.2475, simple_loss=0.3359, pruned_loss=0.07956, over 8506.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2802, pruned_loss=0.05626, over 1608365.89 frames. ], batch size: 28, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:22:42,108 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.506e+02 2.246e+02 2.834e+02 3.422e+02 9.903e+02, threshold=5.667e+02, percent-clipped=4.0 2023-02-09 04:23:04,033 INFO [train.py:901] (0/4) Epoch 30, batch 6750, loss[loss=0.1977, simple_loss=0.2842, pruned_loss=0.05564, over 8440.00 frames. ], tot_loss[loss=0.1956, simple_loss=0.2798, pruned_loss=0.05572, over 1608974.12 frames. ], batch size: 27, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:23:14,120 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 2023-02-09 04:23:15,321 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.5599, 1.8399, 2.7557, 1.4883, 2.0376, 1.9285, 1.5869, 2.1018], device='cuda:0'), covar=tensor([0.2121, 0.2897, 0.0927, 0.4976, 0.1976, 0.3534, 0.2672, 0.2111], device='cuda:0'), in_proj_covar=tensor([0.0548, 0.0650, 0.0570, 0.0681, 0.0675, 0.0624, 0.0576, 0.0653], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], device='cuda:0') 2023-02-09 04:23:39,181 INFO [train.py:901] (0/4) Epoch 30, batch 6800, loss[loss=0.1884, simple_loss=0.2823, pruned_loss=0.04729, over 8471.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2808, pruned_loss=0.05613, over 1612483.20 frames. ], batch size: 49, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:23:42,695 WARNING [train.py:1067] (0/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 2023-02-09 04:23:43,123 INFO [scaling.py:679] (0/4) Whitening: num_groups=1, num_channels=256, metric=3.52 vs. limit=5.0 2023-02-09 04:23:53,781 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.574e+02 2.178e+02 2.682e+02 3.526e+02 7.087e+02, threshold=5.364e+02, percent-clipped=2.0 2023-02-09 04:24:15,461 INFO [train.py:901] (0/4) Epoch 30, batch 6850, loss[loss=0.1656, simple_loss=0.2451, pruned_loss=0.04306, over 7794.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2804, pruned_loss=0.05619, over 1607364.90 frames. ], batch size: 19, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:24:34,765 WARNING [train.py:1067] (0/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 2023-02-09 04:24:35,653 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.9604, 2.4268, 3.8549, 2.0509, 2.0966, 3.8153, 0.9057, 2.2975], device='cuda:0'), covar=tensor([0.1245, 0.0918, 0.0219, 0.1352, 0.1796, 0.0262, 0.1686, 0.1042], device='cuda:0'), in_proj_covar=tensor([0.0203, 0.0207, 0.0138, 0.0223, 0.0277, 0.0149, 0.0174, 0.0201], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 04:24:43,868 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=241294.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 04:24:47,961 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.9932, 1.4838, 3.5528, 1.7244, 2.5799, 3.9107, 4.0319, 3.3728], device='cuda:0'), covar=tensor([0.1268, 0.1956, 0.0302, 0.2063, 0.0982, 0.0212, 0.0586, 0.0525], device='cuda:0'), in_proj_covar=tensor([0.0311, 0.0331, 0.0299, 0.0331, 0.0332, 0.0286, 0.0455, 0.0311], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], device='cuda:0') 2023-02-09 04:24:50,577 INFO [train.py:901] (0/4) Epoch 30, batch 6900, loss[loss=0.232, simple_loss=0.3029, pruned_loss=0.08059, over 7800.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.28, pruned_loss=0.0565, over 1606188.68 frames. ], batch size: 20, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:25:05,718 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.702e+02 2.527e+02 3.094e+02 3.969e+02 8.004e+02, threshold=6.188e+02, percent-clipped=9.0 2023-02-09 04:25:17,363 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.8215, 2.1144, 2.1104, 1.3454, 2.3089, 1.5704, 0.7396, 2.0135], device='cuda:0'), covar=tensor([0.0763, 0.0439, 0.0368, 0.0791, 0.0511, 0.1069, 0.1140, 0.0416], device='cuda:0'), in_proj_covar=tensor([0.0483, 0.0418, 0.0376, 0.0466, 0.0404, 0.0560, 0.0406, 0.0447], device='cuda:0'), out_proj_covar=tensor([1.2769e-04, 1.0758e-04, 9.7909e-05, 1.2150e-04, 1.0537e-04, 1.5562e-04, 1.0836e-04, 1.1663e-04], device='cuda:0') 2023-02-09 04:25:22,104 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=241348.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:25:26,916 INFO [train.py:901] (0/4) Epoch 30, batch 6950, loss[loss=0.191, simple_loss=0.2788, pruned_loss=0.05158, over 8470.00 frames. ], tot_loss[loss=0.195, simple_loss=0.2787, pruned_loss=0.05568, over 1608616.37 frames. ], batch size: 27, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:25:33,886 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.0900, 2.2476, 1.8765, 2.7252, 1.2627, 1.7681, 2.0609, 2.1980], device='cuda:0'), covar=tensor([0.0679, 0.0674, 0.0845, 0.0358, 0.1099, 0.1259, 0.0798, 0.0725], device='cuda:0'), in_proj_covar=tensor([0.0230, 0.0193, 0.0244, 0.0213, 0.0201, 0.0246, 0.0247, 0.0203], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], device='cuda:0') 2023-02-09 04:25:41,892 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.5253, 1.4284, 1.7510, 1.3720, 0.8982, 1.4763, 1.4893, 1.4174], device='cuda:0'), covar=tensor([0.0651, 0.1276, 0.1635, 0.1473, 0.0587, 0.1460, 0.0716, 0.0689], device='cuda:0'), in_proj_covar=tensor([0.0101, 0.0155, 0.0191, 0.0163, 0.0102, 0.0164, 0.0114, 0.0149], device='cuda:0'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0008, 0.0008], device='cuda:0') 2023-02-09 04:25:46,583 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 2023-02-09 04:25:46,725 WARNING [train.py:1067] (0/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 2023-02-09 04:26:04,049 INFO [train.py:901] (0/4) Epoch 30, batch 7000, loss[loss=0.2221, simple_loss=0.3056, pruned_loss=0.06927, over 8559.00 frames. ], tot_loss[loss=0.1933, simple_loss=0.2773, pruned_loss=0.05468, over 1608569.52 frames. ], batch size: 34, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:26:07,751 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=241409.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 04:26:17,960 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.728e+02 2.441e+02 2.932e+02 3.651e+02 7.920e+02, threshold=5.865e+02, percent-clipped=3.0 2023-02-09 04:26:18,126 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=241424.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:26:30,068 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.8597, 1.6180, 2.1748, 1.4300, 1.5682, 2.1548, 1.1769, 1.7183], device='cuda:0'), covar=tensor([0.1342, 0.1036, 0.0385, 0.0890, 0.1728, 0.0492, 0.1475, 0.1178], device='cuda:0'), in_proj_covar=tensor([0.0202, 0.0206, 0.0137, 0.0222, 0.0276, 0.0149, 0.0174, 0.0200], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 04:26:40,283 INFO [train.py:901] (0/4) Epoch 30, batch 7050, loss[loss=0.1517, simple_loss=0.2394, pruned_loss=0.03198, over 5477.00 frames. ], tot_loss[loss=0.194, simple_loss=0.2783, pruned_loss=0.05486, over 1609369.77 frames. ], batch size: 12, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:26:46,805 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=241463.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:26:53,092 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.81 vs. limit=2.0 2023-02-09 04:27:16,742 INFO [train.py:901] (0/4) Epoch 30, batch 7100, loss[loss=0.1956, simple_loss=0.2747, pruned_loss=0.05825, over 8130.00 frames. ], tot_loss[loss=0.194, simple_loss=0.2785, pruned_loss=0.05476, over 1611838.17 frames. ], batch size: 22, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:27:30,724 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.720e+02 2.381e+02 2.857e+02 3.660e+02 8.579e+02, threshold=5.714e+02, percent-clipped=3.0 2023-02-09 04:27:32,322 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.5212, 1.8899, 2.5191, 1.4509, 1.9141, 1.8314, 1.6482, 1.9635], device='cuda:0'), covar=tensor([0.2040, 0.2685, 0.1093, 0.4777, 0.2110, 0.3548, 0.2552, 0.2359], device='cuda:0'), in_proj_covar=tensor([0.0545, 0.0648, 0.0566, 0.0675, 0.0670, 0.0620, 0.0573, 0.0649], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:0') 2023-02-09 04:27:51,599 INFO [train.py:901] (0/4) Epoch 30, batch 7150, loss[loss=0.22, simple_loss=0.3201, pruned_loss=0.05998, over 8248.00 frames. ], tot_loss[loss=0.1939, simple_loss=0.2784, pruned_loss=0.05473, over 1611311.47 frames. ], batch size: 24, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:28:15,816 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.6129, 2.0167, 3.0178, 1.4143, 2.3147, 1.8747, 1.7275, 2.3476], device='cuda:0'), covar=tensor([0.2269, 0.3070, 0.1055, 0.5424, 0.2275, 0.4150, 0.2992, 0.2603], device='cuda:0'), in_proj_covar=tensor([0.0548, 0.0652, 0.0569, 0.0680, 0.0674, 0.0625, 0.0578, 0.0654], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:0') 2023-02-09 04:28:28,711 INFO [train.py:901] (0/4) Epoch 30, batch 7200, loss[loss=0.2531, simple_loss=0.3286, pruned_loss=0.08877, over 8515.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.2796, pruned_loss=0.05565, over 1614372.07 frames. ], batch size: 26, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:28:36,103 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=192, metric=1.53 vs. limit=2.0 2023-02-09 04:28:43,490 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.677e+02 2.259e+02 2.765e+02 3.853e+02 1.030e+03, threshold=5.530e+02, percent-clipped=3.0 2023-02-09 04:28:53,993 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=241639.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:29:03,966 INFO [train.py:901] (0/4) Epoch 30, batch 7250, loss[loss=0.2058, simple_loss=0.2965, pruned_loss=0.05752, over 8254.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2803, pruned_loss=0.0561, over 1614799.12 frames. ], batch size: 24, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:29:11,988 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=241665.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 04:29:30,402 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=241690.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 04:29:40,036 INFO [train.py:901] (0/4) Epoch 30, batch 7300, loss[loss=0.201, simple_loss=0.2797, pruned_loss=0.06111, over 8133.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.2799, pruned_loss=0.05595, over 1614346.47 frames. ], batch size: 22, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:29:50,490 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=241719.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:29:53,738 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.652e+02 2.349e+02 2.997e+02 3.899e+02 6.597e+02, threshold=5.994e+02, percent-clipped=5.0 2023-02-09 04:29:57,751 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.8154, 1.7114, 2.3189, 1.4897, 1.4181, 2.2802, 0.5452, 1.4200], device='cuda:0'), covar=tensor([0.1420, 0.1004, 0.0323, 0.0961, 0.1989, 0.0413, 0.1493, 0.1195], device='cuda:0'), in_proj_covar=tensor([0.0202, 0.0205, 0.0137, 0.0222, 0.0275, 0.0149, 0.0173, 0.0199], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], device='cuda:0') 2023-02-09 04:30:08,895 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=241744.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:30:15,574 INFO [train.py:901] (0/4) Epoch 30, batch 7350, loss[loss=0.2143, simple_loss=0.2999, pruned_loss=0.06441, over 8361.00 frames. ], tot_loss[loss=0.1954, simple_loss=0.2792, pruned_loss=0.0558, over 1612450.95 frames. ], batch size: 24, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:30:25,404 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=241768.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:30:39,744 WARNING [train.py:1067] (0/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 2023-02-09 04:30:51,563 INFO [train.py:901] (0/4) Epoch 30, batch 7400, loss[loss=0.2203, simple_loss=0.3065, pruned_loss=0.06708, over 8596.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.2796, pruned_loss=0.05605, over 1611780.99 frames. ], batch size: 39, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:30:51,676 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([3.8442, 1.4195, 4.1725, 1.8043, 3.2977, 3.2204, 3.6642, 3.6728], device='cuda:0'), covar=tensor([0.1390, 0.6801, 0.1510, 0.5688, 0.2550, 0.2233, 0.1261, 0.1135], device='cuda:0'), in_proj_covar=tensor([0.0703, 0.0675, 0.0762, 0.0679, 0.0764, 0.0652, 0.0663, 0.0739], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 04:30:59,775 WARNING [train.py:1067] (0/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 2023-02-09 04:31:04,083 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=241821.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:31:05,935 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.776e+02 2.497e+02 3.037e+02 3.880e+02 5.984e+02, threshold=6.074e+02, percent-clipped=0.0 2023-02-09 04:31:23,879 INFO [scaling.py:679] (0/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 2023-02-09 04:31:24,335 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=241849.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 04:31:27,626 INFO [train.py:901] (0/4) Epoch 30, batch 7450, loss[loss=0.1567, simple_loss=0.2566, pruned_loss=0.02843, over 7984.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2804, pruned_loss=0.05627, over 1611896.96 frames. ], batch size: 21, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:31:27,785 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([5.9045, 1.6115, 6.0880, 2.2883, 5.5277, 5.1245, 5.5892, 5.5409], device='cuda:0'), covar=tensor([0.0576, 0.5228, 0.0373, 0.4178, 0.0948, 0.0911, 0.0550, 0.0570], device='cuda:0'), in_proj_covar=tensor([0.0701, 0.0674, 0.0760, 0.0677, 0.0762, 0.0650, 0.0661, 0.0737], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], device='cuda:0') 2023-02-09 04:31:40,211 WARNING [train.py:1067] (0/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 2023-02-09 04:31:48,172 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=241883.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:31:54,445 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.7274, 2.2556, 3.4727, 1.4686, 2.7185, 2.0147, 1.8845, 2.5370], device='cuda:0'), covar=tensor([0.2482, 0.3480, 0.1305, 0.6111, 0.2504, 0.4633, 0.3148, 0.3378], device='cuda:0'), in_proj_covar=tensor([0.0547, 0.0651, 0.0568, 0.0680, 0.0672, 0.0622, 0.0575, 0.0651], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:0') 2023-02-09 04:32:02,789 INFO [train.py:901] (0/4) Epoch 30, batch 7500, loss[loss=0.1578, simple_loss=0.2259, pruned_loss=0.04481, over 7232.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2806, pruned_loss=0.05626, over 1609661.98 frames. ], batch size: 16, lr: 2.49e-03, grad_scale: 16.0 2023-02-09 04:32:18,963 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.524e+02 2.491e+02 2.852e+02 3.531e+02 9.058e+02, threshold=5.704e+02, percent-clipped=2.0 2023-02-09 04:32:35,936 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.1020, 2.3110, 1.9909, 2.9153, 1.3697, 1.7598, 2.3116, 2.2704], device='cuda:0'), covar=tensor([0.0741, 0.0840, 0.0832, 0.0336, 0.1104, 0.1315, 0.0754, 0.0826], device='cuda:0'), in_proj_covar=tensor([0.0233, 0.0195, 0.0245, 0.0215, 0.0203, 0.0248, 0.0250, 0.0205], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], device='cuda:0') 2023-02-09 04:32:39,986 INFO [train.py:901] (0/4) Epoch 30, batch 7550, loss[loss=0.2069, simple_loss=0.2942, pruned_loss=0.05978, over 8294.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.28, pruned_loss=0.05579, over 1610689.72 frames. ], batch size: 23, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:32:58,713 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.2510, 2.6974, 2.0918, 3.5277, 1.7191, 1.8834, 2.4517, 2.6250], device='cuda:0'), covar=tensor([0.0722, 0.0747, 0.0840, 0.0342, 0.1021, 0.1197, 0.0804, 0.0732], device='cuda:0'), in_proj_covar=tensor([0.0232, 0.0194, 0.0244, 0.0214, 0.0203, 0.0247, 0.0249, 0.0204], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], device='cuda:0') 2023-02-09 04:33:01,505 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=241983.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:33:13,525 INFO [checkpoint.py:75] (0/4) Saving checkpoint to pruned_transducer_stateless7_streaming/exp/v1/checkpoint-242000.pt 2023-02-09 04:33:17,223 INFO [train.py:901] (0/4) Epoch 30, batch 7600, loss[loss=0.1883, simple_loss=0.282, pruned_loss=0.0473, over 8122.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.28, pruned_loss=0.0557, over 1615687.58 frames. ], batch size: 22, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:33:32,888 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.614e+02 2.414e+02 3.070e+02 3.745e+02 6.631e+02, threshold=6.140e+02, percent-clipped=3.0 2023-02-09 04:33:54,463 INFO [train.py:901] (0/4) Epoch 30, batch 7650, loss[loss=0.2587, simple_loss=0.337, pruned_loss=0.09022, over 8248.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2823, pruned_loss=0.05719, over 1614151.89 frames. ], batch size: 24, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:34:07,141 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([1.4090, 1.5177, 1.4411, 1.7525, 0.7306, 1.3210, 1.4029, 1.4890], device='cuda:0'), covar=tensor([0.0870, 0.0777, 0.0896, 0.0537, 0.1101, 0.1323, 0.0669, 0.0709], device='cuda:0'), in_proj_covar=tensor([0.0233, 0.0195, 0.0245, 0.0215, 0.0203, 0.0248, 0.0250, 0.0204], device='cuda:0'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], device='cuda:0') 2023-02-09 04:34:25,725 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242098.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:34:29,807 INFO [train.py:901] (0/4) Epoch 30, batch 7700, loss[loss=0.1897, simple_loss=0.2896, pruned_loss=0.04493, over 8321.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2813, pruned_loss=0.0563, over 1617746.45 frames. ], batch size: 25, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:34:44,314 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.689e+02 2.431e+02 3.028e+02 3.722e+02 6.918e+02, threshold=6.057e+02, percent-clipped=1.0 2023-02-09 04:34:54,349 WARNING [train.py:1067] (0/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 2023-02-09 04:34:55,305 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=242139.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:35:05,990 INFO [train.py:901] (0/4) Epoch 30, batch 7750, loss[loss=0.2333, simple_loss=0.3193, pruned_loss=0.07362, over 8200.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2802, pruned_loss=0.05627, over 1611284.84 frames. ], batch size: 23, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:35:13,206 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=242164.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:35:13,754 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=242165.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:35:18,585 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242171.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:35:34,895 INFO [zipformer.py:1185] (0/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=242193.0, num_to_drop=1, layers_to_drop={0} 2023-02-09 04:35:42,660 INFO [train.py:901] (0/4) Epoch 30, batch 7800, loss[loss=0.1963, simple_loss=0.2648, pruned_loss=0.0639, over 7186.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.2795, pruned_loss=0.05599, over 1608636.03 frames. ], batch size: 16, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:35:51,706 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([2.1357, 1.8584, 2.2485, 1.9612, 2.2590, 2.1818, 2.0335, 1.2226], device='cuda:0'), covar=tensor([0.5652, 0.5103, 0.2281, 0.3968, 0.2728, 0.3508, 0.1974, 0.5645], device='cuda:0'), in_proj_covar=tensor([0.0974, 0.1039, 0.0856, 0.1017, 0.1036, 0.0951, 0.0782, 0.0863], device='cuda:0'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], device='cuda:0') 2023-02-09 04:35:57,874 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.684e+02 2.396e+02 3.014e+02 3.960e+02 8.063e+02, threshold=6.029e+02, percent-clipped=4.0 2023-02-09 04:36:11,187 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242244.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:36:17,995 INFO [train.py:901] (0/4) Epoch 30, batch 7850, loss[loss=0.1708, simple_loss=0.2543, pruned_loss=0.04366, over 7780.00 frames. ], tot_loss[loss=0.1952, simple_loss=0.2791, pruned_loss=0.05569, over 1608453.92 frames. ], batch size: 19, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:36:36,022 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242280.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:36:52,303 INFO [train.py:901] (0/4) Epoch 30, batch 7900, loss[loss=0.1967, simple_loss=0.2762, pruned_loss=0.05864, over 7919.00 frames. ], tot_loss[loss=0.1944, simple_loss=0.2787, pruned_loss=0.05501, over 1605762.04 frames. ], batch size: 20, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:36:54,993 INFO [zipformer.py:1185] (0/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242308.0, num_to_drop=1, layers_to_drop={1} 2023-02-09 04:37:06,477 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.670e+02 2.359e+02 2.894e+02 3.889e+02 1.272e+03, threshold=5.788e+02, percent-clipped=10.0 2023-02-09 04:37:08,017 INFO [zipformer.py:1185] (0/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242327.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:37:26,300 INFO [train.py:901] (0/4) Epoch 30, batch 7950, loss[loss=0.1998, simple_loss=0.2874, pruned_loss=0.05612, over 8331.00 frames. ], tot_loss[loss=0.1951, simple_loss=0.2797, pruned_loss=0.05528, over 1609491.47 frames. ], batch size: 25, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:37:26,518 INFO [zipformer.py:1185] (0/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=242354.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:37:43,564 INFO [zipformer.py:1185] (0/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=242379.0, num_to_drop=0, layers_to_drop=set() 2023-02-09 04:37:52,887 INFO [zipformer.py:2431] (0/4) attn_weights_entropy = tensor([3.8609, 3.7839, 3.5400, 1.6979, 3.4169, 3.4573, 3.3236, 3.3466], device='cuda:0'), covar=tensor([0.0859, 0.0619, 0.0999, 0.4684, 0.0947, 0.1241, 0.1428, 0.0926], device='cuda:0'), in_proj_covar=tensor([0.0560, 0.0467, 0.0458, 0.0571, 0.0453, 0.0483, 0.0454, 0.0421], device='cuda:0'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], device='cuda:0') 2023-02-09 04:38:00,592 INFO [train.py:901] (0/4) Epoch 30, batch 8000, loss[loss=0.1605, simple_loss=0.2426, pruned_loss=0.03921, over 7706.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2809, pruned_loss=0.05638, over 1607380.98 frames. ], batch size: 18, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:38:14,874 INFO [optim.py:369] (0/4) Clipping_scale=2.0, grad-norm quartiles 1.710e+02 2.478e+02 2.968e+02 3.707e+02 1.083e+03, threshold=5.936e+02, percent-clipped=5.0 2023-02-09 04:38:35,236 INFO [train.py:901] (0/4) Epoch 30, batch 8050, loss[loss=0.2099, simple_loss=0.2922, pruned_loss=0.06385, over 7218.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2804, pruned_loss=0.05697, over 1593580.48 frames. ], batch size: 72, lr: 2.49e-03, grad_scale: 8.0 2023-02-09 04:38:58,267 INFO [checkpoint.py:75] (0/4) Saving checkpoint to pruned_transducer_stateless7_streaming/exp/v1/epoch-30.pt 2023-02-09 04:38:59,103 INFO [train.py:1165] (0/4) Done!