diff --git "a/log/log-train-2023-02-08-23-42-53-2" "b/log/log-train-2023-02-08-23-42-53-2" new file mode 100644--- /dev/null +++ "b/log/log-train-2023-02-08-23-42-53-2" @@ -0,0 +1,2801 @@ +2023-02-08 23:42:53,783 INFO [train.py:973] (2/4) Training started +2023-02-08 23:42:53,784 INFO [train.py:983] (2/4) Device: cuda:2 +2023-02-08 23:42:53,850 INFO [train.py:992] (2/4) {'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] (2/4) About to create model +2023-02-08 23:42:54,155 INFO [zipformer.py:402] (2/4) 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 23:42:54,167 INFO [train.py:998] (2/4) Number of model parameters: 20697573 +2023-02-08 23:42:54,168 INFO [checkpoint.py:112] (2/4) Loading checkpoint from pruned_transducer_stateless7_streaming/exp/v1/epoch-27.pt +2023-02-08 23:43:03,631 INFO [train.py:1013] (2/4) Using DDP +2023-02-08 23:43:03,870 INFO [train.py:1030] (2/4) Loading optimizer state dict +2023-02-08 23:43:04,090 INFO [train.py:1038] (2/4) Loading scheduler state dict +2023-02-08 23:43:04,090 INFO [asr_datamodule.py:420] (2/4) About to get the shuffled train-clean-100, train-clean-360 and train-other-500 cuts +2023-02-08 23:43:04,266 INFO [asr_datamodule.py:224] (2/4) Enable MUSAN +2023-02-08 23:43:04,267 INFO [asr_datamodule.py:225] (2/4) About to get Musan cuts +2023-02-08 23:43:05,857 INFO [asr_datamodule.py:249] (2/4) Enable SpecAugment +2023-02-08 23:43:05,857 INFO [asr_datamodule.py:250] (2/4) Time warp factor: 80 +2023-02-08 23:43:05,857 INFO [asr_datamodule.py:260] (2/4) Num frame mask: 10 +2023-02-08 23:43:05,857 INFO [asr_datamodule.py:273] (2/4) About to create train dataset +2023-02-08 23:43:05,858 INFO [asr_datamodule.py:300] (2/4) Using DynamicBucketingSampler. +2023-02-08 23:43:05,878 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-08 23:43:08,031 INFO [asr_datamodule.py:316] (2/4) About to create train dataloader +2023-02-08 23:43:08,032 INFO [asr_datamodule.py:430] (2/4) About to get dev-clean cuts +2023-02-08 23:43:08,033 INFO [asr_datamodule.py:437] (2/4) About to get dev-other cuts +2023-02-08 23:43:08,034 INFO [asr_datamodule.py:347] (2/4) About to create dev dataset +2023-02-08 23:43:08,393 INFO [asr_datamodule.py:364] (2/4) About to create dev dataloader +2023-02-08 23:43:08,394 INFO [train.py:1122] (2/4) Loading grad scaler state dict +2023-02-08 23:43:20,316 WARNING [train.py:1067] (2/4) 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] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 +2023-02-08 23:43:26,065 INFO [train.py:901] (2/4) Epoch 28, batch 0, loss[loss=0.2635, simple_loss=0.3266, pruned_loss=0.1002, over 7822.00 frames. ], tot_loss[loss=0.2635, simple_loss=0.3266, pruned_loss=0.1002, over 7822.00 frames. ], batch size: 20, lr: 2.71e-03, grad_scale: 8.0 +2023-02-08 23:43:26,065 INFO [train.py:926] (2/4) Computing validation loss +2023-02-08 23:43:38,188 INFO [train.py:935] (2/4) Epoch 28, validation: loss=0.1714, simple_loss=0.2712, pruned_loss=0.03579, over 944034.00 frames. +2023-02-08 23:43:38,189 INFO [train.py:936] (2/4) Maximum memory allocated so far is 5838MB +2023-02-08 23:43:48,595 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218250.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:43:59,673 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-08 23:43:59,745 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=218260.0, num_to_drop=1, layers_to_drop={0} +2023-02-08 23:44:16,006 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.41 vs. limit=2.0 +2023-02-08 23:44:20,162 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0747, 1.9676, 2.4129, 2.0784, 2.5189, 2.1905, 2.0254, 1.3193], + device='cuda:2'), covar=tensor([0.5907, 0.5106, 0.2178, 0.3912, 0.2506, 0.3281, 0.2050, 0.5704], + device='cuda:2'), in_proj_covar=tensor([0.0955, 0.1015, 0.0823, 0.0986, 0.1019, 0.0922, 0.0763, 0.0846], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-08 23:44:24,442 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.44 vs. limit=5.0 +2023-02-08 23:44:26,829 INFO [train.py:901] (2/4) Epoch 28, batch 50, loss[loss=0.1997, simple_loss=0.2883, pruned_loss=0.05557, over 8463.00 frames. ], tot_loss[loss=0.2049, simple_loss=0.2869, pruned_loss=0.06145, over 365673.98 frames. ], batch size: 25, lr: 2.71e-03, grad_scale: 8.0 +2023-02-08 23:44:44,696 WARNING [train.py:1067] (2/4) 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] (2/4) 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 23:44:58,760 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8713, 1.5754, 3.1736, 1.4389, 2.3785, 3.3934, 3.5531, 2.9075], + device='cuda:2'), covar=tensor([0.1193, 0.1724, 0.0332, 0.2177, 0.0963, 0.0262, 0.0610, 0.0592], + device='cuda:2'), in_proj_covar=tensor([0.0306, 0.0324, 0.0291, 0.0320, 0.0322, 0.0277, 0.0438, 0.0306], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:2') +2023-02-08 23:45:09,774 INFO [train.py:901] (2/4) Epoch 28, batch 100, loss[loss=0.1887, simple_loss=0.2729, pruned_loss=0.05221, over 8240.00 frames. ], tot_loss[loss=0.2018, simple_loss=0.2852, pruned_loss=0.05914, over 646175.08 frames. ], batch size: 22, lr: 2.71e-03, grad_scale: 8.0 +2023-02-08 23:45:12,261 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-08 23:45:23,499 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.87 vs. limit=5.0 +2023-02-08 23:45:42,216 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218375.0, num_to_drop=1, layers_to_drop={0} +2023-02-08 23:45:52,960 INFO [train.py:901] (2/4) Epoch 28, batch 150, loss[loss=0.2193, simple_loss=0.3072, pruned_loss=0.06568, over 8290.00 frames. ], tot_loss[loss=0.201, simple_loss=0.2849, pruned_loss=0.05852, over 860056.92 frames. ], batch size: 23, lr: 2.71e-03, grad_scale: 8.0 +2023-02-08 23:46:01,141 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218397.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:46:02,731 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.74 vs. limit=2.0 +2023-02-08 23:46:12,820 INFO [optim.py:369] (2/4) 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 23:46:19,735 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218422.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:46:21,826 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4735, 1.3436, 1.7566, 1.1345, 1.1462, 1.7312, 0.3269, 1.1473], + device='cuda:2'), covar=tensor([0.1602, 0.1214, 0.0421, 0.0954, 0.2356, 0.0507, 0.1753, 0.1308], + device='cuda:2'), in_proj_covar=tensor([0.0200, 0.0206, 0.0136, 0.0224, 0.0277, 0.0147, 0.0172, 0.0199], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-08 23:46:32,308 INFO [train.py:901] (2/4) Epoch 28, batch 200, loss[loss=0.1943, simple_loss=0.28, pruned_loss=0.05427, over 8241.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2842, pruned_loss=0.0582, over 1030304.90 frames. ], batch size: 24, lr: 2.71e-03, grad_scale: 8.0 +2023-02-08 23:46:50,628 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=218462.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:47:10,732 INFO [train.py:901] (2/4) Epoch 28, batch 250, loss[loss=0.1715, simple_loss=0.2499, pruned_loss=0.0465, over 8244.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2839, pruned_loss=0.05824, over 1160043.40 frames. ], batch size: 22, lr: 2.71e-03, grad_scale: 16.0 +2023-02-08 23:47:23,077 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-08 23:47:30,508 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-08 23:47:31,295 INFO [optim.py:369] (2/4) 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 23:47:33,443 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-08 23:47:41,362 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218527.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:47:48,879 INFO [train.py:901] (2/4) Epoch 28, batch 300, loss[loss=0.1924, simple_loss=0.2804, pruned_loss=0.05223, over 8143.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2836, pruned_loss=0.05814, over 1261330.38 frames. ], batch size: 22, lr: 2.71e-03, grad_scale: 16.0 +2023-02-08 23:47:53,405 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=218544.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:48:11,947 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.38 vs. limit=2.0 +2023-02-08 23:48:14,321 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=218572.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:48:18,151 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218577.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:48:18,211 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6711, 2.2286, 4.1426, 1.6249, 3.1813, 2.3419, 1.6833, 3.3090], + device='cuda:2'), covar=tensor([0.2203, 0.2813, 0.0829, 0.4877, 0.1785, 0.3288, 0.2772, 0.2095], + device='cuda:2'), in_proj_covar=tensor([0.0541, 0.0636, 0.0565, 0.0671, 0.0658, 0.0614, 0.0567, 0.0645], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-08 23:48:25,696 INFO [train.py:901] (2/4) Epoch 28, batch 350, loss[loss=0.1993, simple_loss=0.2818, pruned_loss=0.05839, over 7820.00 frames. ], tot_loss[loss=0.2006, simple_loss=0.2838, pruned_loss=0.05868, over 1338721.29 frames. ], batch size: 20, lr: 2.71e-03, grad_scale: 16.0 +2023-02-08 23:48:28,663 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218592.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:48:32,352 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8331, 1.4151, 1.7126, 1.3324, 1.0089, 1.4982, 1.6372, 1.3791], + device='cuda:2'), covar=tensor([0.0557, 0.1310, 0.1667, 0.1518, 0.0580, 0.1480, 0.0712, 0.0699], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0153, 0.0190, 0.0161, 0.0101, 0.0163, 0.0112, 0.0146], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-08 23:48:43,888 INFO [optim.py:369] (2/4) 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 23:48:44,821 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3592, 2.1468, 2.7650, 2.3662, 2.8273, 2.4041, 2.2432, 1.6073], + device='cuda:2'), covar=tensor([0.5949, 0.5276, 0.2069, 0.3875, 0.2510, 0.3152, 0.1865, 0.5629], + device='cuda:2'), in_proj_covar=tensor([0.0961, 0.1019, 0.0824, 0.0987, 0.1020, 0.0924, 0.0764, 0.0847], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-08 23:48:59,282 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218631.0, num_to_drop=1, layers_to_drop={0} +2023-02-08 23:49:02,080 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.0691, 4.0027, 3.6991, 2.2400, 3.5732, 3.6662, 3.6190, 3.5035], + device='cuda:2'), covar=tensor([0.0781, 0.0635, 0.1047, 0.3996, 0.0935, 0.1122, 0.1370, 0.0883], + device='cuda:2'), in_proj_covar=tensor([0.0539, 0.0461, 0.0446, 0.0556, 0.0444, 0.0463, 0.0439, 0.0406], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-08 23:49:04,807 INFO [train.py:901] (2/4) Epoch 28, batch 400, loss[loss=0.1504, simple_loss=0.2365, pruned_loss=0.03217, over 8252.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2836, pruned_loss=0.05908, over 1402197.13 frames. ], batch size: 22, lr: 2.71e-03, grad_scale: 16.0 +2023-02-08 23:49:09,174 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4502, 2.3768, 3.1229, 2.5834, 3.0777, 2.5130, 2.3477, 1.8429], + device='cuda:2'), covar=tensor([0.6009, 0.5255, 0.2239, 0.4061, 0.2771, 0.3260, 0.2060, 0.5945], + device='cuda:2'), in_proj_covar=tensor([0.0961, 0.1018, 0.0824, 0.0987, 0.1019, 0.0925, 0.0764, 0.0846], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-08 23:49:17,852 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218656.0, num_to_drop=1, layers_to_drop={0} +2023-02-08 23:49:19,967 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218659.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:49:40,568 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218687.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:49:41,121 INFO [train.py:901] (2/4) Epoch 28, batch 450, loss[loss=0.1549, simple_loss=0.2413, pruned_loss=0.03426, over 7796.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2827, pruned_loss=0.05885, over 1448765.14 frames. ], batch size: 19, lr: 2.71e-03, grad_scale: 16.0 +2023-02-08 23:49:59,789 INFO [optim.py:369] (2/4) 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 23:50:18,542 INFO [train.py:901] (2/4) Epoch 28, batch 500, loss[loss=0.1613, simple_loss=0.249, pruned_loss=0.03679, over 7539.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2813, pruned_loss=0.05789, over 1485993.51 frames. ], batch size: 18, lr: 2.71e-03, grad_scale: 16.0 +2023-02-08 23:50:37,341 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1509, 4.0630, 3.6793, 2.1175, 3.5538, 3.7770, 3.6208, 3.6920], + device='cuda:2'), covar=tensor([0.0745, 0.0578, 0.1010, 0.4287, 0.0944, 0.0947, 0.1337, 0.0697], + device='cuda:2'), in_proj_covar=tensor([0.0542, 0.0462, 0.0446, 0.0557, 0.0445, 0.0465, 0.0440, 0.0406], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-08 23:50:57,135 INFO [train.py:901] (2/4) Epoch 28, batch 550, loss[loss=0.2036, simple_loss=0.2755, pruned_loss=0.0658, over 8074.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2827, pruned_loss=0.05879, over 1517059.57 frames. ], batch size: 21, lr: 2.71e-03, grad_scale: 8.0 +2023-02-08 23:51:16,043 INFO [optim.py:369] (2/4) 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 23:51:30,169 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218833.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:51:33,478 INFO [train.py:901] (2/4) Epoch 28, batch 600, loss[loss=0.1787, simple_loss=0.2478, pruned_loss=0.0548, over 7539.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2816, pruned_loss=0.05821, over 1533321.52 frames. ], batch size: 18, lr: 2.71e-03, grad_scale: 8.0 +2023-02-08 23:51:53,195 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218858.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:51:56,594 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-08 23:51:59,013 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8156, 1.8353, 1.7045, 2.3518, 1.1245, 1.6052, 1.8483, 1.8676], + device='cuda:2'), covar=tensor([0.0747, 0.0802, 0.0842, 0.0371, 0.0984, 0.1199, 0.0654, 0.0704], + device='cuda:2'), in_proj_covar=tensor([0.0229, 0.0193, 0.0242, 0.0211, 0.0201, 0.0244, 0.0247, 0.0203], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-08 23:52:04,161 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=218871.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:52:18,548 INFO [train.py:901] (2/4) Epoch 28, batch 650, loss[loss=0.1541, simple_loss=0.2461, pruned_loss=0.0311, over 7702.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2796, pruned_loss=0.05718, over 1550122.00 frames. ], batch size: 18, lr: 2.71e-03, grad_scale: 8.0 +2023-02-08 23:52:40,031 INFO [optim.py:369] (2/4) 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 23:52:41,075 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218915.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:52:55,980 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=218936.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:52:57,376 INFO [train.py:901] (2/4) Epoch 28, batch 700, loss[loss=0.2091, simple_loss=0.2992, pruned_loss=0.05949, over 8504.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2816, pruned_loss=0.05783, over 1568759.29 frames. ], batch size: 26, lr: 2.71e-03, grad_scale: 8.0 +2023-02-08 23:52:59,039 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218940.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:53:01,202 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=218943.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:53:18,864 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=218968.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:53:31,058 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=218983.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:53:33,198 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=218986.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:53:34,532 INFO [train.py:901] (2/4) Epoch 28, batch 750, loss[loss=0.2096, simple_loss=0.2896, pruned_loss=0.06475, over 8255.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2821, pruned_loss=0.05839, over 1578459.95 frames. ], batch size: 24, lr: 2.71e-03, grad_scale: 8.0 +2023-02-08 23:53:46,822 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219002.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:53:55,127 INFO [optim.py:369] (2/4) 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 23:53:55,167 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-08 23:54:04,603 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-08 23:54:12,581 INFO [train.py:901] (2/4) Epoch 28, batch 800, loss[loss=0.2238, simple_loss=0.3091, pruned_loss=0.06923, over 8103.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2816, pruned_loss=0.05844, over 1583476.12 frames. ], batch size: 23, lr: 2.71e-03, grad_scale: 8.0 +2023-02-08 23:54:13,441 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219039.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:54:22,035 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 28, batch 850, loss[loss=0.191, simple_loss=0.2696, pruned_loss=0.05617, over 7926.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2815, pruned_loss=0.0579, over 1588019.05 frames. ], batch size: 20, lr: 2.71e-03, grad_scale: 8.0 +2023-02-08 23:54:50,777 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219090.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:55:10,262 INFO [optim.py:369] (2/4) 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 23:55:27,563 INFO [train.py:901] (2/4) Epoch 28, batch 900, loss[loss=0.192, simple_loss=0.2689, pruned_loss=0.05757, over 7421.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2813, pruned_loss=0.05792, over 1596406.60 frames. ], batch size: 17, lr: 2.71e-03, grad_scale: 8.0 +2023-02-08 23:55:34,942 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.63 vs. limit=5.0 +2023-02-08 23:56:03,865 INFO [train.py:901] (2/4) Epoch 28, batch 950, loss[loss=0.179, simple_loss=0.2566, pruned_loss=0.05066, over 7713.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2814, pruned_loss=0.05745, over 1601908.35 frames. ], batch size: 18, lr: 2.71e-03, grad_scale: 8.0 +2023-02-08 23:56:22,890 INFO [optim.py:369] (2/4) 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 23:56:29,706 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219221.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:56:34,858 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-08 23:56:43,600 INFO [train.py:901] (2/4) Epoch 28, batch 1000, loss[loss=0.1658, simple_loss=0.2452, pruned_loss=0.04319, over 7980.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2799, pruned_loss=0.05662, over 1603380.68 frames. ], batch size: 21, lr: 2.71e-03, grad_scale: 8.0 +2023-02-08 23:56:46,627 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219242.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:56:47,401 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2993, 2.2642, 3.0140, 2.3807, 2.9722, 2.4137, 2.2323, 1.8164], + device='cuda:2'), covar=tensor([0.6292, 0.5632, 0.2271, 0.4523, 0.2941, 0.3513, 0.2141, 0.6113], + device='cuda:2'), in_proj_covar=tensor([0.0967, 0.1026, 0.0832, 0.0994, 0.1027, 0.0932, 0.0774, 0.0852], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-08 23:57:04,456 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219267.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:57:11,558 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-08 23:57:19,441 INFO [train.py:901] (2/4) Epoch 28, batch 1050, loss[loss=0.1824, simple_loss=0.2564, pruned_loss=0.05425, over 7530.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2798, pruned_loss=0.05659, over 1602463.54 frames. ], batch size: 18, lr: 2.71e-03, grad_scale: 8.0 +2023-02-08 23:57:23,705 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-08 23:57:33,329 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219307.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:57:38,274 INFO [optim.py:369] (2/4) 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 23:57:47,817 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=219327.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:57:52,096 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219332.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:57:56,800 INFO [train.py:901] (2/4) Epoch 28, batch 1100, loss[loss=0.2268, simple_loss=0.311, pruned_loss=0.07135, over 8508.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2805, pruned_loss=0.05696, over 1601832.43 frames. ], batch size: 26, lr: 2.71e-03, grad_scale: 8.0 +2023-02-08 23:58:03,512 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=219346.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:58:12,850 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.7590, 1.9769, 5.8661, 2.1061, 5.3045, 4.8292, 5.4118, 5.2699], + device='cuda:2'), covar=tensor([0.0479, 0.4567, 0.0494, 0.4497, 0.1065, 0.0979, 0.0520, 0.0535], + device='cuda:2'), in_proj_covar=tensor([0.0679, 0.0662, 0.0731, 0.0657, 0.0744, 0.0635, 0.0639, 0.0714], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-08 23:58:13,593 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219360.0, num_to_drop=1, layers_to_drop={0} +2023-02-08 23:58:30,138 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=219383.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:58:33,477 INFO [train.py:901] (2/4) Epoch 28, batch 1150, loss[loss=0.2489, simple_loss=0.3118, pruned_loss=0.09298, over 7251.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2815, pruned_loss=0.05729, over 1597636.90 frames. ], batch size: 71, lr: 2.71e-03, grad_scale: 8.0 +2023-02-08 23:58:37,161 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-08 23:58:52,533 INFO [optim.py:369] (2/4) 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 23:59:02,425 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.17 vs. limit=5.0 +2023-02-08 23:59:07,140 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=219434.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:59:10,002 INFO [train.py:901] (2/4) Epoch 28, batch 1200, loss[loss=0.1793, simple_loss=0.2664, pruned_loss=0.04609, over 8027.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.282, pruned_loss=0.0573, over 1603418.65 frames. ], batch size: 22, lr: 2.71e-03, grad_scale: 8.0 +2023-02-08 23:59:13,047 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219442.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:59:27,991 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219461.0, num_to_drop=0, layers_to_drop=set() +2023-02-08 23:59:30,760 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5892, 2.8791, 2.4310, 3.9538, 1.8408, 2.2177, 2.7122, 2.7695], + device='cuda:2'), covar=tensor([0.0687, 0.0821, 0.0748, 0.0308, 0.1070, 0.1209, 0.0893, 0.0854], + device='cuda:2'), in_proj_covar=tensor([0.0227, 0.0192, 0.0241, 0.0209, 0.0200, 0.0243, 0.0246, 0.0202], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-08 23:59:47,602 INFO [train.py:901] (2/4) Epoch 28, batch 1250, loss[loss=0.1804, simple_loss=0.2727, pruned_loss=0.04406, over 7811.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2816, pruned_loss=0.05698, over 1607890.12 frames. ], batch size: 20, lr: 2.71e-03, grad_scale: 8.0 +2023-02-08 23:59:55,045 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219498.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:00:02,360 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219508.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:00:06,605 INFO [optim.py:369] (2/4) 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 00:00:20,247 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3929, 2.6987, 2.9361, 1.6697, 3.2752, 2.1245, 1.6339, 2.3047], + device='cuda:2'), covar=tensor([0.0926, 0.0441, 0.0336, 0.0925, 0.0543, 0.0920, 0.1050, 0.0646], + device='cuda:2'), in_proj_covar=tensor([0.0475, 0.0412, 0.0366, 0.0461, 0.0395, 0.0552, 0.0403, 0.0444], + device='cuda:2'), out_proj_covar=tensor([1.2574e-04, 1.0706e-04, 9.5227e-05, 1.2065e-04, 1.0346e-04, 1.5374e-04, + 1.0753e-04, 1.1646e-04], device='cuda:2') +2023-02-09 00:00:23,408 INFO [train.py:901] (2/4) Epoch 28, batch 1300, loss[loss=0.2095, simple_loss=0.3053, pruned_loss=0.05685, over 8468.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2817, pruned_loss=0.05736, over 1606368.50 frames. ], batch size: 25, lr: 2.71e-03, grad_scale: 8.0 +2023-02-09 00:00:31,528 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219549.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:00:42,938 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=219565.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:01:02,337 INFO [train.py:901] (2/4) Epoch 28, batch 1350, loss[loss=0.1887, simple_loss=0.2634, pruned_loss=0.05705, over 7940.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2815, pruned_loss=0.05703, over 1606961.30 frames. ], batch size: 20, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:01:22,006 INFO [optim.py:369] (2/4) 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 00:01:39,672 INFO [train.py:901] (2/4) Epoch 28, batch 1400, loss[loss=0.2064, simple_loss=0.2924, pruned_loss=0.06023, over 8465.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2812, pruned_loss=0.05667, over 1610024.88 frames. ], batch size: 49, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:01:47,197 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7358, 4.7313, 4.2370, 2.3327, 4.1866, 4.2991, 4.2153, 4.1477], + device='cuda:2'), covar=tensor([0.0638, 0.0471, 0.0951, 0.3865, 0.0900, 0.1059, 0.1130, 0.0696], + device='cuda:2'), in_proj_covar=tensor([0.0539, 0.0459, 0.0447, 0.0555, 0.0442, 0.0464, 0.0438, 0.0405], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 00:02:09,800 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219680.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:02:15,332 INFO [train.py:901] (2/4) Epoch 28, batch 1450, loss[loss=0.2335, simple_loss=0.3115, pruned_loss=0.07779, over 7210.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2808, pruned_loss=0.05658, over 1613804.92 frames. ], batch size: 71, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:02:22,723 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2437, 2.5251, 2.1514, 3.7971, 1.6476, 1.8709, 2.4330, 2.6917], + device='cuda:2'), covar=tensor([0.0831, 0.0860, 0.1024, 0.0284, 0.1092, 0.1387, 0.0891, 0.0831], + device='cuda:2'), in_proj_covar=tensor([0.0226, 0.0191, 0.0240, 0.0208, 0.0199, 0.0242, 0.0245, 0.0202], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 00:02:24,272 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219698.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:02:25,389 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-09 00:02:28,351 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=219704.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 00:02:33,734 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.52 vs. limit=5.0 +2023-02-09 00:02:36,733 INFO [optim.py:369] (2/4) 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 00:02:39,220 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219717.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:02:43,507 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219723.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:02:54,214 INFO [train.py:901] (2/4) Epoch 28, batch 1500, loss[loss=0.1848, simple_loss=0.2562, pruned_loss=0.05666, over 7644.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2812, pruned_loss=0.05673, over 1615298.17 frames. ], batch size: 19, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:02:55,522 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2023-02-09 00:02:57,282 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219742.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:03:00,791 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.17 vs. limit=2.0 +2023-02-09 00:03:05,714 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219754.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:03:23,845 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219779.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:03:25,277 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219781.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:03:29,883 INFO [train.py:901] (2/4) Epoch 28, batch 1550, loss[loss=0.1906, simple_loss=0.2643, pruned_loss=0.05843, over 7247.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2824, pruned_loss=0.05755, over 1615400.98 frames. ], batch size: 16, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:03:42,604 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=219805.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:03:49,436 INFO [optim.py:369] (2/4) 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 00:03:54,630 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219819.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 00:04:03,146 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219830.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:04:08,593 INFO [train.py:901] (2/4) Epoch 28, batch 1600, loss[loss=0.1835, simple_loss=0.2727, pruned_loss=0.04719, over 8229.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2815, pruned_loss=0.05694, over 1617169.30 frames. ], batch size: 22, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:04:18,710 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=219852.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:04:35,327 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=219875.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:04:44,470 INFO [train.py:901] (2/4) Epoch 28, batch 1650, loss[loss=0.2327, simple_loss=0.2998, pruned_loss=0.08279, over 7214.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2814, pruned_loss=0.05713, over 1612516.57 frames. ], batch size: 16, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:05:02,696 INFO [optim.py:369] (2/4) 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 00:05:06,525 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0622, 1.8213, 2.3493, 2.0323, 2.3349, 2.1793, 1.9544, 1.1932], + device='cuda:2'), covar=tensor([0.6152, 0.5302, 0.2051, 0.3865, 0.2601, 0.3317, 0.2063, 0.5637], + device='cuda:2'), in_proj_covar=tensor([0.0967, 0.1026, 0.0832, 0.0998, 0.1026, 0.0933, 0.0774, 0.0853], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 00:05:09,983 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3391, 2.1298, 2.6953, 2.3028, 2.6782, 2.3553, 2.2362, 1.7915], + device='cuda:2'), covar=tensor([0.5124, 0.4606, 0.2068, 0.3576, 0.2357, 0.3132, 0.1773, 0.4812], + device='cuda:2'), in_proj_covar=tensor([0.0968, 0.1026, 0.0832, 0.0998, 0.1026, 0.0933, 0.0774, 0.0853], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 00:05:20,265 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 28, batch 1700, loss[loss=0.2099, simple_loss=0.2784, pruned_loss=0.07072, over 8243.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2813, pruned_loss=0.05689, over 1615003.14 frames. ], batch size: 22, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:05:39,122 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=219961.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:05:43,291 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=219967.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:05:56,599 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3101, 2.1409, 1.7440, 1.9853, 1.7187, 1.5918, 1.7240, 1.7516], + device='cuda:2'), covar=tensor([0.1289, 0.0484, 0.1278, 0.0526, 0.0833, 0.1522, 0.0960, 0.0911], + device='cuda:2'), in_proj_covar=tensor([0.0360, 0.0245, 0.0344, 0.0316, 0.0307, 0.0350, 0.0351, 0.0326], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 00:05:57,778 INFO [train.py:901] (2/4) Epoch 28, batch 1750, loss[loss=0.2009, simple_loss=0.2932, pruned_loss=0.05428, over 8468.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2808, pruned_loss=0.05652, over 1616327.69 frames. ], batch size: 25, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:06:17,596 INFO [optim.py:369] (2/4) 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 00:06:34,451 INFO [train.py:901] (2/4) Epoch 28, batch 1800, loss[loss=0.1867, simple_loss=0.2715, pruned_loss=0.0509, over 7687.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2809, pruned_loss=0.05696, over 1617416.00 frames. ], batch size: 18, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:06:39,240 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1652, 1.4412, 4.3716, 1.6057, 3.9261, 3.6999, 3.9901, 3.8659], + device='cuda:2'), covar=tensor([0.0719, 0.4677, 0.0574, 0.4393, 0.1140, 0.0958, 0.0635, 0.0736], + device='cuda:2'), in_proj_covar=tensor([0.0681, 0.0668, 0.0738, 0.0661, 0.0750, 0.0638, 0.0643, 0.0720], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 00:07:02,982 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220075.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 00:07:11,958 INFO [train.py:901] (2/4) Epoch 28, batch 1850, loss[loss=0.2384, simple_loss=0.3099, pruned_loss=0.08342, over 6970.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.282, pruned_loss=0.05747, over 1615844.59 frames. ], batch size: 71, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:07:20,524 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220100.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 00:07:23,911 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220105.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:07:30,313 INFO [optim.py:369] (2/4) 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 00:07:31,761 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220116.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:07:38,070 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 28, batch 1900, loss[loss=0.189, simple_loss=0.2788, pruned_loss=0.04966, over 8188.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2817, pruned_loss=0.05709, over 1621159.52 frames. ], batch size: 23, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:08:19,370 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-09 00:08:25,770 INFO [train.py:901] (2/4) Epoch 28, batch 1950, loss[loss=0.1864, simple_loss=0.2684, pruned_loss=0.05221, over 7977.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.283, pruned_loss=0.05798, over 1618438.41 frames. ], batch size: 21, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:08:32,997 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-09 00:08:44,725 INFO [optim.py:369] (2/4) 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 00:08:48,265 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=220219.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:08:51,058 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220223.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:08:53,663 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-09 00:09:01,379 INFO [train.py:901] (2/4) Epoch 28, batch 2000, loss[loss=0.1948, simple_loss=0.2643, pruned_loss=0.06261, over 7242.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2828, pruned_loss=0.05749, over 1620265.95 frames. ], batch size: 16, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:09:02,181 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4653, 1.3539, 2.6893, 1.2221, 2.1691, 2.9162, 3.1884, 2.1610], + device='cuda:2'), covar=tensor([0.1739, 0.2110, 0.0570, 0.2864, 0.1118, 0.0459, 0.0763, 0.1122], + device='cuda:2'), in_proj_covar=tensor([0.0309, 0.0327, 0.0293, 0.0320, 0.0322, 0.0277, 0.0441, 0.0309], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:2') +2023-02-09 00:09:02,868 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220240.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:09:07,807 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220247.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:09:08,577 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220248.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:09:29,131 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220276.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:09:37,451 INFO [train.py:901] (2/4) Epoch 28, batch 2050, loss[loss=0.1849, simple_loss=0.2854, pruned_loss=0.0422, over 8244.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2824, pruned_loss=0.05719, over 1620625.74 frames. ], batch size: 24, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:09:58,200 INFO [optim.py:369] (2/4) 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 00:10:12,783 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220334.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:10:15,428 INFO [train.py:901] (2/4) Epoch 28, batch 2100, loss[loss=0.1959, simple_loss=0.2835, pruned_loss=0.05419, over 8191.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2834, pruned_loss=0.0579, over 1624215.84 frames. ], batch size: 23, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:10:51,250 INFO [train.py:901] (2/4) Epoch 28, batch 2150, loss[loss=0.2212, simple_loss=0.2965, pruned_loss=0.07293, over 8140.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2838, pruned_loss=0.05839, over 1621612.89 frames. ], batch size: 22, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:11:11,476 INFO [optim.py:369] (2/4) 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 00:11:28,321 INFO [train.py:901] (2/4) Epoch 28, batch 2200, loss[loss=0.2285, simple_loss=0.3105, pruned_loss=0.07328, over 8080.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2825, pruned_loss=0.05819, over 1618039.85 frames. ], batch size: 21, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:11:36,282 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=220449.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:11:44,020 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=220460.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:11:48,267 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6055, 1.8589, 1.9747, 1.3070, 2.0832, 1.4573, 0.5349, 1.9078], + device='cuda:2'), covar=tensor([0.0636, 0.0412, 0.0334, 0.0692, 0.0447, 0.0994, 0.1020, 0.0330], + device='cuda:2'), in_proj_covar=tensor([0.0477, 0.0415, 0.0366, 0.0464, 0.0399, 0.0554, 0.0404, 0.0446], + device='cuda:2'), out_proj_covar=tensor([1.2618e-04, 1.0759e-04, 9.5326e-05, 1.2143e-04, 1.0427e-04, 1.5443e-04, + 1.0803e-04, 1.1695e-04], device='cuda:2') +2023-02-09 00:11:51,008 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6878, 1.8809, 1.9840, 1.4286, 2.1242, 1.5175, 0.7317, 1.9959], + device='cuda:2'), covar=tensor([0.0715, 0.0447, 0.0344, 0.0715, 0.0405, 0.0963, 0.1106, 0.0359], + device='cuda:2'), in_proj_covar=tensor([0.0477, 0.0415, 0.0366, 0.0465, 0.0399, 0.0554, 0.0405, 0.0447], + device='cuda:2'), out_proj_covar=tensor([1.2627e-04, 1.0766e-04, 9.5398e-05, 1.2151e-04, 1.0434e-04, 1.5453e-04, + 1.0809e-04, 1.1701e-04], device='cuda:2') +2023-02-09 00:12:03,395 INFO [train.py:901] (2/4) Epoch 28, batch 2250, loss[loss=0.2269, simple_loss=0.3063, pruned_loss=0.07376, over 7827.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.283, pruned_loss=0.05787, over 1618483.61 frames. ], batch size: 20, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:12:09,250 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220496.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:12:22,281 INFO [optim.py:369] (2/4) 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 00:12:24,961 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-09 00:12:27,561 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220521.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:12:41,520 INFO [train.py:901] (2/4) Epoch 28, batch 2300, loss[loss=0.1923, simple_loss=0.2896, pruned_loss=0.04754, over 8451.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2824, pruned_loss=0.05757, over 1617834.21 frames. ], batch size: 29, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:12:43,141 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.16 vs. limit=5.0 +2023-02-09 00:12:59,726 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220564.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:13:06,020 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220573.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:13:07,462 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220575.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:13:13,909 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7953, 1.6573, 1.8571, 1.6862, 1.0713, 1.6150, 2.3974, 2.0934], + device='cuda:2'), covar=tensor([0.0451, 0.1257, 0.1731, 0.1426, 0.0612, 0.1462, 0.0595, 0.0595], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0153, 0.0189, 0.0161, 0.0101, 0.0162, 0.0112, 0.0145], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-09 00:13:16,606 INFO [train.py:901] (2/4) Epoch 28, batch 2350, loss[loss=0.2506, simple_loss=0.3298, pruned_loss=0.08565, over 7243.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2837, pruned_loss=0.05856, over 1612780.69 frames. ], batch size: 71, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:13:18,226 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220590.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:13:18,828 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=220591.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:13:35,684 INFO [optim.py:369] (2/4) 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 00:13:36,666 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220615.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:13:40,247 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=220620.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:13:53,607 INFO [train.py:901] (2/4) Epoch 28, batch 2400, loss[loss=0.1821, simple_loss=0.267, pruned_loss=0.04865, over 8087.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2828, pruned_loss=0.05796, over 1613259.73 frames. ], batch size: 21, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:14:01,101 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.32 vs. limit=2.0 +2023-02-09 00:14:16,686 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220669.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:14:29,715 INFO [train.py:901] (2/4) Epoch 28, batch 2450, loss[loss=0.1773, simple_loss=0.2562, pruned_loss=0.04923, over 7420.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2819, pruned_loss=0.05767, over 1612186.10 frames. ], batch size: 17, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:14:42,731 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220706.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:14:48,791 INFO [optim.py:369] (2/4) 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 00:15:03,094 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=220735.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:15:05,105 INFO [train.py:901] (2/4) Epoch 28, batch 2500, loss[loss=0.2104, simple_loss=0.3016, pruned_loss=0.05953, over 8202.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2806, pruned_loss=0.05672, over 1609894.58 frames. ], batch size: 23, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:15:15,089 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9384, 1.4507, 1.5897, 1.3703, 1.0203, 1.3807, 1.7554, 1.3954], + device='cuda:2'), covar=tensor([0.0558, 0.1332, 0.1772, 0.1549, 0.0603, 0.1575, 0.0705, 0.0718], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0153, 0.0189, 0.0161, 0.0101, 0.0163, 0.0113, 0.0146], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-09 00:15:42,750 INFO [train.py:901] (2/4) Epoch 28, batch 2550, loss[loss=0.2096, simple_loss=0.2821, pruned_loss=0.06857, over 7928.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2809, pruned_loss=0.05639, over 1615411.39 frames. ], batch size: 20, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:16:02,765 INFO [optim.py:369] (2/4) 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 00:16:06,706 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220820.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:16:14,501 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220831.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:16:19,226 INFO [train.py:901] (2/4) Epoch 28, batch 2600, loss[loss=0.182, simple_loss=0.2536, pruned_loss=0.0552, over 7651.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2805, pruned_loss=0.05671, over 1616423.70 frames. ], batch size: 19, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:16:20,647 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220840.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:16:24,289 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220845.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:16:32,229 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220856.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:16:57,368 INFO [train.py:901] (2/4) Epoch 28, batch 2650, loss[loss=0.1831, simple_loss=0.2745, pruned_loss=0.04589, over 8340.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2806, pruned_loss=0.05651, over 1613835.26 frames. ], batch size: 25, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:17:01,285 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.76 vs. limit=5.0 +2023-02-09 00:17:16,289 INFO [optim.py:369] (2/4) 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] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=220917.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:17:21,290 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=220922.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:17:32,909 INFO [train.py:901] (2/4) Epoch 28, batch 2700, loss[loss=0.1999, simple_loss=0.2839, pruned_loss=0.05795, over 8181.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2822, pruned_loss=0.05765, over 1611543.14 frames. ], batch size: 23, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:17:50,478 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220962.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:18:09,047 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=220987.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:18:09,538 INFO [train.py:901] (2/4) Epoch 28, batch 2750, loss[loss=0.1833, simple_loss=0.2692, pruned_loss=0.04867, over 7544.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2812, pruned_loss=0.05765, over 1612263.05 frames. ], batch size: 18, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:18:11,782 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=220991.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:18:29,677 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=221013.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:18:31,037 INFO [optim.py:369] (2/4) 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 00:18:31,973 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221016.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:18:35,000 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-09 00:18:43,192 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.9000, 3.8341, 3.5310, 1.8400, 3.4705, 3.5392, 3.3920, 3.3822], + device='cuda:2'), covar=tensor([0.0752, 0.0597, 0.0965, 0.3952, 0.0875, 0.1086, 0.1275, 0.0942], + device='cuda:2'), in_proj_covar=tensor([0.0536, 0.0453, 0.0442, 0.0552, 0.0438, 0.0460, 0.0434, 0.0403], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 00:18:43,281 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221032.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:18:47,245 INFO [train.py:901] (2/4) Epoch 28, batch 2800, loss[loss=0.1487, simple_loss=0.2272, pruned_loss=0.03512, over 7430.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2796, pruned_loss=0.05674, over 1610519.38 frames. ], batch size: 17, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:19:22,678 INFO [train.py:901] (2/4) Epoch 28, batch 2850, loss[loss=0.1686, simple_loss=0.2609, pruned_loss=0.03814, over 8288.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2805, pruned_loss=0.05686, over 1612968.79 frames. ], batch size: 23, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:19:36,290 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4658, 1.2762, 2.3995, 1.3612, 2.2346, 2.5020, 2.7109, 2.1620], + device='cuda:2'), covar=tensor([0.1172, 0.1485, 0.0446, 0.2046, 0.0770, 0.0402, 0.0764, 0.0635], + device='cuda:2'), in_proj_covar=tensor([0.0309, 0.0329, 0.0295, 0.0322, 0.0325, 0.0278, 0.0443, 0.0310], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:2') +2023-02-09 00:19:43,226 INFO [optim.py:369] (2/4) 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 00:19:44,152 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8124, 1.5132, 1.8487, 1.4904, 1.0519, 1.5485, 2.1185, 1.8273], + device='cuda:2'), covar=tensor([0.0443, 0.1318, 0.1707, 0.1516, 0.0601, 0.1490, 0.0632, 0.0680], + device='cuda:2'), in_proj_covar=tensor([0.0098, 0.0152, 0.0188, 0.0161, 0.0101, 0.0162, 0.0113, 0.0146], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-09 00:19:45,663 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3917, 2.7688, 2.9380, 1.8567, 3.1878, 2.0215, 1.8269, 2.4002], + device='cuda:2'), covar=tensor([0.0997, 0.0423, 0.0348, 0.0985, 0.0523, 0.1026, 0.1033, 0.0574], + device='cuda:2'), in_proj_covar=tensor([0.0479, 0.0415, 0.0368, 0.0464, 0.0399, 0.0555, 0.0406, 0.0447], + device='cuda:2'), out_proj_covar=tensor([1.2668e-04, 1.0748e-04, 9.5821e-05, 1.2132e-04, 1.0425e-04, 1.5453e-04, + 1.0846e-04, 1.1720e-04], device='cuda:2') +2023-02-09 00:19:53,961 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221128.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:20:00,662 INFO [train.py:901] (2/4) Epoch 28, batch 2900, loss[loss=0.2034, simple_loss=0.2939, pruned_loss=0.05643, over 8495.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2819, pruned_loss=0.05749, over 1608909.03 frames. ], batch size: 28, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:20:14,797 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.6507, 5.6749, 5.0550, 2.3394, 5.1407, 5.3875, 5.2084, 5.2078], + device='cuda:2'), covar=tensor([0.0489, 0.0357, 0.0821, 0.4420, 0.0702, 0.0744, 0.1037, 0.0624], + device='cuda:2'), in_proj_covar=tensor([0.0539, 0.0456, 0.0445, 0.0556, 0.0440, 0.0463, 0.0437, 0.0404], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 00:20:32,280 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-09 00:20:33,725 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=221184.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:20:36,493 INFO [train.py:901] (2/4) Epoch 28, batch 2950, loss[loss=0.1687, simple_loss=0.2489, pruned_loss=0.04424, over 7806.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2814, pruned_loss=0.05729, over 1607630.45 frames. ], batch size: 20, lr: 2.70e-03, grad_scale: 8.0 +2023-02-09 00:20:55,456 INFO [optim.py:369] (2/4) 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 00:21:13,540 INFO [train.py:901] (2/4) Epoch 28, batch 3000, loss[loss=0.2281, simple_loss=0.3065, pruned_loss=0.07485, over 8321.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2807, pruned_loss=0.05676, over 1611284.44 frames. ], batch size: 26, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:21:13,541 INFO [train.py:926] (2/4) Computing validation loss +2023-02-09 00:21:31,975 INFO [train.py:935] (2/4) Epoch 28, validation: loss=0.1712, simple_loss=0.2708, pruned_loss=0.03578, over 944034.00 frames. +2023-02-09 00:21:31,976 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6510MB +2023-02-09 00:21:54,457 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=221266.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:22:01,998 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7522, 1.7392, 1.6383, 2.1061, 1.0557, 1.5030, 1.6533, 1.8220], + device='cuda:2'), covar=tensor([0.0735, 0.0778, 0.0879, 0.0546, 0.1054, 0.1179, 0.0711, 0.0663], + device='cuda:2'), in_proj_covar=tensor([0.0230, 0.0194, 0.0244, 0.0212, 0.0202, 0.0245, 0.0249, 0.0204], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 00:22:10,137 INFO [train.py:901] (2/4) Epoch 28, batch 3050, loss[loss=0.2035, simple_loss=0.2853, pruned_loss=0.06087, over 6987.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.283, pruned_loss=0.05785, over 1613862.82 frames. ], batch size: 71, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:22:10,366 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221288.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:22:18,080 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221299.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:22:28,218 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221313.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:22:29,357 INFO [optim.py:369] (2/4) 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 00:22:45,361 INFO [train.py:901] (2/4) Epoch 28, batch 3100, loss[loss=0.2137, simple_loss=0.2974, pruned_loss=0.06501, over 8484.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2827, pruned_loss=0.05766, over 1616575.90 frames. ], batch size: 28, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:23:18,459 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221381.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:23:20,619 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221384.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:23:21,943 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0382, 1.3670, 3.1799, 1.5703, 2.2751, 3.3991, 3.5504, 2.9291], + device='cuda:2'), covar=tensor([0.1028, 0.1883, 0.0314, 0.2029, 0.1026, 0.0243, 0.0453, 0.0516], + device='cuda:2'), in_proj_covar=tensor([0.0309, 0.0330, 0.0294, 0.0323, 0.0326, 0.0278, 0.0443, 0.0309], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:2') +2023-02-09 00:23:23,254 INFO [train.py:901] (2/4) Epoch 28, batch 3150, loss[loss=0.2397, simple_loss=0.3194, pruned_loss=0.07997, over 8499.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2822, pruned_loss=0.05738, over 1617280.27 frames. ], batch size: 26, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:23:27,178 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3609, 1.5762, 2.0058, 1.2751, 1.4591, 1.5650, 1.4320, 1.3772], + device='cuda:2'), covar=tensor([0.2111, 0.2695, 0.1154, 0.5035, 0.2240, 0.3747, 0.2746, 0.2618], + device='cuda:2'), in_proj_covar=tensor([0.0537, 0.0636, 0.0563, 0.0669, 0.0659, 0.0608, 0.0562, 0.0645], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 00:23:38,989 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221409.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:23:43,020 INFO [optim.py:369] (2/4) 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 00:24:00,284 INFO [train.py:901] (2/4) Epoch 28, batch 3200, loss[loss=0.2135, simple_loss=0.2849, pruned_loss=0.07111, over 7979.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2831, pruned_loss=0.05782, over 1617519.62 frames. ], batch size: 21, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:24:20,317 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2941, 2.5559, 2.8377, 1.6528, 3.1946, 1.9329, 1.5424, 2.2745], + device='cuda:2'), covar=tensor([0.0882, 0.0459, 0.0336, 0.0894, 0.0475, 0.0957, 0.1142, 0.0661], + device='cuda:2'), in_proj_covar=tensor([0.0477, 0.0413, 0.0366, 0.0463, 0.0397, 0.0553, 0.0405, 0.0446], + device='cuda:2'), out_proj_covar=tensor([1.2618e-04, 1.0701e-04, 9.5382e-05, 1.2113e-04, 1.0373e-04, 1.5417e-04, + 1.0822e-04, 1.1671e-04], device='cuda:2') +2023-02-09 00:24:36,745 INFO [train.py:901] (2/4) Epoch 28, batch 3250, loss[loss=0.1867, simple_loss=0.2707, pruned_loss=0.05131, over 7525.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2825, pruned_loss=0.05698, over 1614784.49 frames. ], batch size: 18, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:24:56,686 INFO [optim.py:369] (2/4) 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 00:25:04,061 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221525.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:25:12,963 INFO [train.py:901] (2/4) Epoch 28, batch 3300, loss[loss=0.2082, simple_loss=0.2945, pruned_loss=0.06092, over 8365.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2833, pruned_loss=0.05742, over 1616501.92 frames. ], batch size: 24, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:25:23,906 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.31 vs. limit=2.0 +2023-02-09 00:25:25,043 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221555.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:25:42,947 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221580.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:25:48,454 INFO [train.py:901] (2/4) Epoch 28, batch 3350, loss[loss=0.2246, simple_loss=0.3077, pruned_loss=0.07078, over 8596.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2838, pruned_loss=0.05748, over 1619063.80 frames. ], batch size: 50, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:26:09,958 INFO [optim.py:369] (2/4) 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 00:26:23,943 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4784, 1.4154, 1.8290, 1.2299, 1.1552, 1.7979, 0.2241, 1.1481], + device='cuda:2'), covar=tensor([0.1530, 0.1140, 0.0374, 0.0820, 0.2272, 0.0447, 0.1789, 0.1113], + device='cuda:2'), in_proj_covar=tensor([0.0203, 0.0208, 0.0138, 0.0226, 0.0281, 0.0148, 0.0176, 0.0201], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 00:26:26,129 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=221637.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:26:26,629 INFO [train.py:901] (2/4) Epoch 28, batch 3400, loss[loss=0.1935, simple_loss=0.2847, pruned_loss=0.05116, over 8463.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2833, pruned_loss=0.0575, over 1618110.24 frames. ], batch size: 25, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:26:43,876 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=221662.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:27:02,342 INFO [train.py:901] (2/4) Epoch 28, batch 3450, loss[loss=0.2389, simple_loss=0.3171, pruned_loss=0.08033, over 8444.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2835, pruned_loss=0.05745, over 1620447.52 frames. ], batch size: 27, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:27:21,423 INFO [optim.py:369] (2/4) 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 00:27:39,496 INFO [train.py:901] (2/4) Epoch 28, batch 3500, loss[loss=0.1745, simple_loss=0.2564, pruned_loss=0.04633, over 7808.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2839, pruned_loss=0.05793, over 1616109.87 frames. ], batch size: 20, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:28:03,514 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-09 00:28:07,244 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221776.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:28:15,811 INFO [train.py:901] (2/4) Epoch 28, batch 3550, loss[loss=0.1585, simple_loss=0.2499, pruned_loss=0.03356, over 8514.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2825, pruned_loss=0.05753, over 1614205.57 frames. ], batch size: 28, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:28:35,273 INFO [optim.py:369] (2/4) 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 00:28:52,586 INFO [train.py:901] (2/4) Epoch 28, batch 3600, loss[loss=0.1704, simple_loss=0.2504, pruned_loss=0.04518, over 7651.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2823, pruned_loss=0.05724, over 1610055.06 frames. ], batch size: 19, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:29:15,453 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=221869.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:29:20,482 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.7203, 5.8670, 5.1273, 2.5728, 5.1723, 5.5467, 5.3092, 5.3161], + device='cuda:2'), covar=tensor([0.0497, 0.0309, 0.0804, 0.4166, 0.0704, 0.0803, 0.1034, 0.0570], + device='cuda:2'), in_proj_covar=tensor([0.0541, 0.0456, 0.0447, 0.0559, 0.0443, 0.0464, 0.0440, 0.0406], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 00:29:28,481 INFO [train.py:901] (2/4) Epoch 28, batch 3650, loss[loss=0.1561, simple_loss=0.2444, pruned_loss=0.0339, over 7822.00 frames. ], tot_loss[loss=0.2003, simple_loss=0.2838, pruned_loss=0.05837, over 1614627.59 frames. ], batch size: 20, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:29:42,348 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=221908.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:29:47,070 INFO [optim.py:369] (2/4) 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 00:30:02,963 INFO [train.py:901] (2/4) Epoch 28, batch 3700, loss[loss=0.1848, simple_loss=0.2699, pruned_loss=0.04987, over 8028.00 frames. ], tot_loss[loss=0.2009, simple_loss=0.2844, pruned_loss=0.05876, over 1614586.13 frames. ], batch size: 22, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:30:05,069 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-09 00:30:29,849 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 +2023-02-09 00:30:38,684 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=221984.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:30:41,396 INFO [train.py:901] (2/4) Epoch 28, batch 3750, loss[loss=0.228, simple_loss=0.3016, pruned_loss=0.07718, over 6805.00 frames. ], tot_loss[loss=0.2005, simple_loss=0.2844, pruned_loss=0.05829, over 1613357.10 frames. ], batch size: 71, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:30:51,146 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.45 vs. limit=2.0 +2023-02-09 00:31:01,419 INFO [optim.py:369] (2/4) 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 00:31:10,204 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222027.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:31:17,963 INFO [train.py:901] (2/4) Epoch 28, batch 3800, loss[loss=0.2259, simple_loss=0.3045, pruned_loss=0.07361, over 8632.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2841, pruned_loss=0.05833, over 1610893.11 frames. ], batch size: 31, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:31:55,278 INFO [train.py:901] (2/4) Epoch 28, batch 3850, loss[loss=0.2128, simple_loss=0.2998, pruned_loss=0.06293, over 8321.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2831, pruned_loss=0.05773, over 1613449.80 frames. ], batch size: 26, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:32:06,549 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-09 00:32:11,779 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-09 00:32:13,833 INFO [optim.py:369] (2/4) 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 00:32:17,527 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=222120.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:32:30,066 INFO [train.py:901] (2/4) Epoch 28, batch 3900, loss[loss=0.1529, simple_loss=0.236, pruned_loss=0.03487, over 8241.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2829, pruned_loss=0.05757, over 1612281.75 frames. ], batch size: 22, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:33:01,202 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2873, 1.4491, 3.4014, 1.1987, 3.0342, 2.8788, 3.0946, 3.0424], + device='cuda:2'), covar=tensor([0.0800, 0.3931, 0.0749, 0.4133, 0.1269, 0.1052, 0.0793, 0.0856], + device='cuda:2'), in_proj_covar=tensor([0.0682, 0.0669, 0.0734, 0.0662, 0.0746, 0.0634, 0.0642, 0.0717], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 00:33:06,447 INFO [train.py:901] (2/4) Epoch 28, batch 3950, loss[loss=0.1823, simple_loss=0.2775, pruned_loss=0.04354, over 8238.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2826, pruned_loss=0.05765, over 1611262.72 frames. ], batch size: 24, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:33:17,880 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222203.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:33:26,088 INFO [optim.py:369] (2/4) 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 00:33:31,798 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222223.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:33:40,239 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222235.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:33:42,144 INFO [train.py:901] (2/4) Epoch 28, batch 4000, loss[loss=0.1806, simple_loss=0.2575, pruned_loss=0.05179, over 7271.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2818, pruned_loss=0.05719, over 1612693.64 frames. ], batch size: 16, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:33:43,721 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222240.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:33:51,747 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=222252.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:34:01,385 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222265.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:34:02,822 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7763, 1.7010, 2.5079, 2.0080, 2.2339, 1.8808, 1.6434, 1.1309], + device='cuda:2'), covar=tensor([0.7167, 0.6145, 0.2260, 0.3956, 0.3277, 0.4629, 0.2822, 0.5614], + device='cuda:2'), in_proj_covar=tensor([0.0965, 0.1027, 0.0832, 0.0993, 0.1021, 0.0934, 0.0772, 0.0851], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 00:34:08,192 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6929, 2.0689, 3.1813, 1.5597, 2.5351, 2.0749, 1.8097, 2.4845], + device='cuda:2'), covar=tensor([0.1953, 0.2585, 0.0911, 0.4544, 0.1705, 0.3316, 0.2442, 0.2101], + device='cuda:2'), in_proj_covar=tensor([0.0539, 0.0638, 0.0565, 0.0673, 0.0664, 0.0611, 0.0566, 0.0645], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 00:34:09,544 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3066, 2.0625, 1.6375, 1.8789, 1.6425, 1.3848, 1.5785, 1.6901], + device='cuda:2'), covar=tensor([0.1515, 0.0517, 0.1379, 0.0661, 0.0957, 0.1862, 0.1166, 0.0985], + device='cuda:2'), in_proj_covar=tensor([0.0359, 0.0245, 0.0343, 0.0314, 0.0305, 0.0348, 0.0350, 0.0323], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 00:34:17,514 INFO [train.py:901] (2/4) Epoch 28, batch 4050, loss[loss=0.2357, simple_loss=0.3125, pruned_loss=0.0794, over 6636.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2824, pruned_loss=0.05815, over 1610965.41 frames. ], batch size: 74, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:34:38,321 INFO [optim.py:369] (2/4) 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 00:34:54,250 INFO [train.py:901] (2/4) Epoch 28, batch 4100, loss[loss=0.1901, simple_loss=0.2751, pruned_loss=0.05251, over 8532.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2818, pruned_loss=0.05785, over 1606830.69 frames. ], batch size: 28, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:35:14,877 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222367.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:35:17,646 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=222371.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:35:29,381 INFO [train.py:901] (2/4) Epoch 28, batch 4150, loss[loss=0.1389, simple_loss=0.2204, pruned_loss=0.02867, over 7446.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2813, pruned_loss=0.05715, over 1607815.47 frames. ], batch size: 17, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:35:49,108 INFO [optim.py:369] (2/4) 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 00:35:50,176 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.30 vs. limit=5.0 +2023-02-09 00:36:07,144 INFO [train.py:901] (2/4) Epoch 28, batch 4200, loss[loss=0.2025, simple_loss=0.2785, pruned_loss=0.06327, over 7800.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2809, pruned_loss=0.05672, over 1611174.44 frames. ], batch size: 19, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:36:14,728 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-09 00:36:37,817 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-09 00:36:41,346 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 28, batch 4250, loss[loss=0.1757, simple_loss=0.252, pruned_loss=0.04975, over 7715.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2806, pruned_loss=0.05666, over 1610349.09 frames. ], batch size: 18, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:36:44,931 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222491.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:37:00,852 INFO [optim.py:369] (2/4) 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 00:37:01,766 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222516.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:37:18,019 INFO [train.py:901] (2/4) Epoch 28, batch 4300, loss[loss=0.1764, simple_loss=0.2667, pruned_loss=0.04306, over 8466.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2805, pruned_loss=0.05643, over 1611972.68 frames. ], batch size: 28, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:37:20,182 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222541.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:37:25,031 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=222547.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:37:28,747 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-09 00:37:29,476 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-09 00:37:39,465 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=222567.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:37:54,079 INFO [train.py:901] (2/4) Epoch 28, batch 4350, loss[loss=0.1981, simple_loss=0.2931, pruned_loss=0.05159, over 8462.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.281, pruned_loss=0.05693, over 1610592.07 frames. ], batch size: 25, lr: 2.69e-03, grad_scale: 8.0 +2023-02-09 00:37:54,222 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222588.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:37:59,284 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0976, 1.6514, 1.9131, 1.6104, 1.1771, 1.7402, 1.9083, 1.8171], + device='cuda:2'), covar=tensor([0.0563, 0.1103, 0.1512, 0.1359, 0.0574, 0.1306, 0.0632, 0.0606], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0153, 0.0190, 0.0162, 0.0101, 0.0163, 0.0113, 0.0146], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-09 00:38:05,223 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.62 vs. limit=2.0 +2023-02-09 00:38:11,699 WARNING [train.py:1067] (2/4) 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] (2/4) 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 00:38:18,809 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222623.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:38:29,009 INFO [train.py:901] (2/4) Epoch 28, batch 4400, loss[loss=0.1904, simple_loss=0.2659, pruned_loss=0.05745, over 7545.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2807, pruned_loss=0.05681, over 1607107.08 frames. ], batch size: 18, lr: 2.69e-03, grad_scale: 16.0 +2023-02-09 00:38:36,663 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222648.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:38:47,084 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222662.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:38:54,364 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-09 00:39:01,974 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=222682.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:39:05,978 INFO [train.py:901] (2/4) Epoch 28, batch 4450, loss[loss=0.2203, simple_loss=0.3022, pruned_loss=0.06918, over 8459.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2824, pruned_loss=0.0574, over 1612042.34 frames. ], batch size: 27, lr: 2.69e-03, grad_scale: 16.0 +2023-02-09 00:39:24,976 INFO [optim.py:369] (2/4) 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 00:39:41,230 INFO [train.py:901] (2/4) Epoch 28, batch 4500, loss[loss=0.195, simple_loss=0.282, pruned_loss=0.05403, over 8232.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.283, pruned_loss=0.05761, over 1616720.48 frames. ], batch size: 22, lr: 2.69e-03, grad_scale: 16.0 +2023-02-09 00:39:44,228 INFO [zipformer.py:1185] (2/4) 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] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-09 00:39:53,950 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.89 vs. limit=2.0 +2023-02-09 00:40:02,730 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222767.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:40:18,345 INFO [train.py:901] (2/4) Epoch 28, batch 4550, loss[loss=0.2163, simple_loss=0.2867, pruned_loss=0.0729, over 8091.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2823, pruned_loss=0.05758, over 1616509.33 frames. ], batch size: 21, lr: 2.69e-03, grad_scale: 16.0 +2023-02-09 00:40:22,053 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6435, 2.5259, 1.8776, 2.3071, 2.1598, 1.6833, 2.1190, 2.2022], + device='cuda:2'), covar=tensor([0.1698, 0.0485, 0.1250, 0.0671, 0.0834, 0.1555, 0.1110, 0.1105], + device='cuda:2'), in_proj_covar=tensor([0.0360, 0.0245, 0.0342, 0.0315, 0.0304, 0.0349, 0.0349, 0.0322], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 00:40:37,191 INFO [optim.py:369] (2/4) 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 00:40:40,226 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-09 00:40:53,695 INFO [train.py:901] (2/4) Epoch 28, batch 4600, loss[loss=0.1818, simple_loss=0.2758, pruned_loss=0.04388, over 8626.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2814, pruned_loss=0.0569, over 1615729.17 frames. ], batch size: 34, lr: 2.69e-03, grad_scale: 16.0 +2023-02-09 00:41:27,918 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=222885.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:41:30,003 INFO [train.py:901] (2/4) Epoch 28, batch 4650, loss[loss=0.1835, simple_loss=0.2661, pruned_loss=0.05049, over 8351.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2824, pruned_loss=0.05744, over 1618274.60 frames. ], batch size: 26, lr: 2.68e-03, grad_scale: 16.0 +2023-02-09 00:41:50,685 INFO [optim.py:369] (2/4) 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 00:41:53,069 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222918.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:42:02,696 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=222932.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:42:06,747 INFO [train.py:901] (2/4) Epoch 28, batch 4700, loss[loss=0.1607, simple_loss=0.2517, pruned_loss=0.03482, over 8198.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2826, pruned_loss=0.05755, over 1619270.26 frames. ], batch size: 23, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:42:06,959 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=222938.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:42:10,454 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222943.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:42:20,635 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222958.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:42:24,272 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=222963.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:42:40,849 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=222987.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:42:41,323 INFO [train.py:901] (2/4) Epoch 28, batch 4750, loss[loss=0.2023, simple_loss=0.2826, pruned_loss=0.06102, over 8242.00 frames. ], tot_loss[loss=0.2004, simple_loss=0.2837, pruned_loss=0.0586, over 1615480.15 frames. ], batch size: 22, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:42:50,004 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223000.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:42:52,051 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223002.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:42:54,707 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-09 00:42:58,161 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-09 00:43:02,875 INFO [optim.py:369] (2/4) 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 00:43:18,663 INFO [train.py:901] (2/4) Epoch 28, batch 4800, loss[loss=0.1869, simple_loss=0.2687, pruned_loss=0.05257, over 8242.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2827, pruned_loss=0.05787, over 1620063.31 frames. ], batch size: 22, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:43:25,201 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223047.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:43:48,815 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-09 00:43:49,658 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223082.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:43:53,653 INFO [train.py:901] (2/4) Epoch 28, batch 4850, loss[loss=0.1967, simple_loss=0.2821, pruned_loss=0.05568, over 8138.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2826, pruned_loss=0.05785, over 1616510.96 frames. ], batch size: 22, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:44:13,745 INFO [optim.py:369] (2/4) 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 00:44:18,768 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9317, 1.7748, 2.5227, 1.6317, 1.4921, 2.5336, 0.5955, 1.5696], + device='cuda:2'), covar=tensor([0.1234, 0.0985, 0.0321, 0.0974, 0.1884, 0.0324, 0.1648, 0.1106], + device='cuda:2'), in_proj_covar=tensor([0.0200, 0.0205, 0.0137, 0.0223, 0.0277, 0.0147, 0.0174, 0.0198], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 00:44:21,502 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6875, 1.6574, 1.9187, 1.6838, 1.2611, 1.7676, 2.3381, 2.2037], + device='cuda:2'), covar=tensor([0.0504, 0.1229, 0.1658, 0.1463, 0.0630, 0.1461, 0.0635, 0.0604], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0153, 0.0190, 0.0161, 0.0101, 0.0163, 0.0113, 0.0146], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-09 00:44:31,160 INFO [train.py:901] (2/4) Epoch 28, batch 4900, loss[loss=0.1829, simple_loss=0.2608, pruned_loss=0.05254, over 7694.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2823, pruned_loss=0.05787, over 1616497.45 frames. ], batch size: 18, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:45:07,006 INFO [train.py:901] (2/4) Epoch 28, batch 4950, loss[loss=0.2069, simple_loss=0.298, pruned_loss=0.05791, over 8358.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2824, pruned_loss=0.05839, over 1612101.42 frames. ], batch size: 24, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:45:09,279 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223191.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 00:45:26,418 INFO [optim.py:369] (2/4) 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 00:45:42,295 INFO [train.py:901] (2/4) Epoch 28, batch 5000, loss[loss=0.2367, simple_loss=0.3168, pruned_loss=0.07824, over 8356.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2822, pruned_loss=0.05787, over 1613034.02 frames. ], batch size: 26, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:45:56,141 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223256.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:46:14,352 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223281.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:46:18,935 INFO [train.py:901] (2/4) Epoch 28, batch 5050, loss[loss=0.2248, simple_loss=0.3143, pruned_loss=0.06765, over 8316.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2812, pruned_loss=0.05761, over 1605560.07 frames. ], batch size: 25, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:46:28,802 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=223302.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:46:29,636 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223303.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:46:32,888 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-09 00:46:35,804 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1341, 1.6625, 1.4858, 1.6023, 1.3216, 1.3355, 1.3600, 1.3402], + device='cuda:2'), covar=tensor([0.1128, 0.0524, 0.1339, 0.0584, 0.0812, 0.1489, 0.0888, 0.0825], + device='cuda:2'), in_proj_covar=tensor([0.0357, 0.0243, 0.0339, 0.0314, 0.0303, 0.0345, 0.0347, 0.0321], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 00:46:38,384 INFO [optim.py:369] (2/4) 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 00:46:46,947 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223328.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:46:48,963 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=223331.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:46:53,548 INFO [train.py:901] (2/4) Epoch 28, batch 5100, loss[loss=0.1988, simple_loss=0.28, pruned_loss=0.05884, over 8106.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2811, pruned_loss=0.05748, over 1605413.82 frames. ], batch size: 23, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:46:59,469 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=223346.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:47:13,977 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1277, 2.1816, 1.9343, 2.3656, 1.8403, 1.9536, 2.1345, 2.2867], + device='cuda:2'), covar=tensor([0.0618, 0.0689, 0.0739, 0.0559, 0.0794, 0.0918, 0.0544, 0.0573], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0194, 0.0244, 0.0213, 0.0203, 0.0246, 0.0250, 0.0205], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 00:47:31,191 INFO [train.py:901] (2/4) Epoch 28, batch 5150, loss[loss=0.1634, simple_loss=0.2451, pruned_loss=0.04086, over 8143.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2827, pruned_loss=0.05799, over 1611635.79 frames. ], batch size: 22, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:47:37,583 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0500, 1.2226, 1.1790, 0.7642, 1.1943, 0.9894, 0.1226, 1.1953], + device='cuda:2'), covar=tensor([0.0472, 0.0419, 0.0399, 0.0620, 0.0466, 0.1077, 0.0957, 0.0366], + device='cuda:2'), in_proj_covar=tensor([0.0469, 0.0409, 0.0362, 0.0456, 0.0394, 0.0549, 0.0400, 0.0438], + device='cuda:2'), out_proj_covar=tensor([1.2407e-04, 1.0587e-04, 9.4242e-05, 1.1912e-04, 1.0306e-04, 1.5314e-04, + 1.0684e-04, 1.1481e-04], device='cuda:2') +2023-02-09 00:47:38,171 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4234, 4.4026, 4.0601, 2.0759, 3.9367, 3.9835, 4.0305, 3.8404], + device='cuda:2'), covar=tensor([0.0747, 0.0511, 0.1022, 0.4403, 0.0861, 0.1158, 0.1272, 0.0777], + device='cuda:2'), in_proj_covar=tensor([0.0539, 0.0454, 0.0449, 0.0554, 0.0443, 0.0463, 0.0441, 0.0407], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 00:47:43,143 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.39 vs. limit=5.0 +2023-02-09 00:47:50,491 INFO [optim.py:369] (2/4) 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 00:47:51,402 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223417.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:47:57,706 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=223426.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:47:58,571 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6863, 1.5710, 2.2121, 1.5200, 1.2551, 2.1699, 0.5416, 1.3520], + device='cuda:2'), covar=tensor([0.1376, 0.1281, 0.0368, 0.0958, 0.2324, 0.0453, 0.1817, 0.1234], + device='cuda:2'), in_proj_covar=tensor([0.0200, 0.0206, 0.0137, 0.0224, 0.0279, 0.0148, 0.0175, 0.0200], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 00:48:05,981 INFO [train.py:901] (2/4) Epoch 28, batch 5200, loss[loss=0.2115, simple_loss=0.2973, pruned_loss=0.0629, over 8335.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2831, pruned_loss=0.05807, over 1612381.97 frames. ], batch size: 26, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:48:11,622 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223446.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:48:22,805 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223461.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:48:31,226 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-09 00:48:44,069 INFO [train.py:901] (2/4) Epoch 28, batch 5250, loss[loss=0.2056, simple_loss=0.2871, pruned_loss=0.06206, over 8082.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2826, pruned_loss=0.05759, over 1614006.20 frames. ], batch size: 21, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:49:03,763 INFO [optim.py:369] (2/4) 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 00:49:17,137 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 28, batch 5300, loss[loss=0.2054, simple_loss=0.2975, pruned_loss=0.05662, over 8514.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2819, pruned_loss=0.05689, over 1617585.67 frames. ], batch size: 26, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:49:21,384 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=223541.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:49:22,686 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4858, 2.3534, 3.0415, 2.4709, 2.9336, 2.5077, 2.4356, 1.9912], + device='cuda:2'), covar=tensor([0.5781, 0.5409, 0.2346, 0.4249, 0.2997, 0.3311, 0.1822, 0.5875], + device='cuda:2'), in_proj_covar=tensor([0.0966, 0.1030, 0.0835, 0.0999, 0.1026, 0.0936, 0.0773, 0.0855], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 00:49:55,534 INFO [train.py:901] (2/4) Epoch 28, batch 5350, loss[loss=0.2086, simple_loss=0.2903, pruned_loss=0.06342, over 8106.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2815, pruned_loss=0.05666, over 1612471.31 frames. ], batch size: 23, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:50:01,801 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=223596.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:50:04,676 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8558, 1.3815, 3.9884, 1.5163, 3.6004, 3.3596, 3.6363, 3.5540], + device='cuda:2'), covar=tensor([0.0649, 0.4787, 0.0648, 0.4049, 0.1106, 0.0982, 0.0656, 0.0733], + device='cuda:2'), in_proj_covar=tensor([0.0679, 0.0663, 0.0734, 0.0656, 0.0737, 0.0632, 0.0640, 0.0712], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 00:50:15,667 INFO [optim.py:369] (2/4) 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 00:50:30,415 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2953, 2.1691, 2.7466, 2.3284, 2.7037, 2.4173, 2.2411, 1.6247], + device='cuda:2'), covar=tensor([0.5447, 0.5068, 0.2105, 0.4120, 0.2603, 0.3143, 0.1986, 0.5690], + device='cuda:2'), in_proj_covar=tensor([0.0967, 0.1031, 0.0837, 0.1000, 0.1027, 0.0938, 0.0774, 0.0856], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 00:50:30,879 INFO [train.py:901] (2/4) Epoch 28, batch 5400, loss[loss=0.2167, simple_loss=0.2971, pruned_loss=0.06814, over 8455.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2817, pruned_loss=0.05693, over 1610939.15 frames. ], batch size: 29, lr: 2.68e-03, grad_scale: 4.0 +2023-02-09 00:50:39,707 INFO [zipformer.py:1185] (2/4) 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] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223673.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:51:03,757 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7536, 2.6024, 1.8743, 2.4145, 2.1466, 1.5257, 2.1857, 2.3734], + device='cuda:2'), covar=tensor([0.1597, 0.0459, 0.1353, 0.0704, 0.0865, 0.1806, 0.1133, 0.0990], + device='cuda:2'), in_proj_covar=tensor([0.0359, 0.0244, 0.0342, 0.0315, 0.0303, 0.0347, 0.0350, 0.0323], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 00:51:06,370 INFO [train.py:901] (2/4) Epoch 28, batch 5450, loss[loss=0.1811, simple_loss=0.2759, pruned_loss=0.04321, over 8087.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2809, pruned_loss=0.05675, over 1609262.18 frames. ], batch size: 21, lr: 2.68e-03, grad_scale: 4.0 +2023-02-09 00:51:13,798 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223698.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:51:16,756 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223702.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:51:28,420 INFO [optim.py:369] (2/4) 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 00:51:28,659 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223717.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:51:29,454 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3530, 2.3193, 3.1246, 2.4956, 3.1266, 2.5084, 2.2948, 1.9194], + device='cuda:2'), covar=tensor([0.6263, 0.5260, 0.2195, 0.4335, 0.2672, 0.3287, 0.2022, 0.5955], + device='cuda:2'), in_proj_covar=tensor([0.0966, 0.1029, 0.0837, 0.0998, 0.1025, 0.0934, 0.0772, 0.0855], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 00:51:31,411 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-09 00:51:35,317 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 +2023-02-09 00:51:36,526 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223727.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:51:44,560 INFO [train.py:901] (2/4) Epoch 28, batch 5500, loss[loss=0.1538, simple_loss=0.2423, pruned_loss=0.03267, over 7928.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2805, pruned_loss=0.0562, over 1611431.85 frames. ], batch size: 20, lr: 2.68e-03, grad_scale: 4.0 +2023-02-09 00:51:47,330 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223742.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:52:18,970 INFO [train.py:901] (2/4) Epoch 28, batch 5550, loss[loss=0.1697, simple_loss=0.2666, pruned_loss=0.03642, over 8251.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2812, pruned_loss=0.05684, over 1610023.15 frames. ], batch size: 24, lr: 2.68e-03, grad_scale: 4.0 +2023-02-09 00:52:20,252 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 +2023-02-09 00:52:25,546 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223797.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:52:39,665 INFO [optim.py:369] (2/4) 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 00:52:43,463 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223822.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:52:55,584 INFO [train.py:901] (2/4) Epoch 28, batch 5600, loss[loss=0.1952, simple_loss=0.2938, pruned_loss=0.04834, over 8246.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2802, pruned_loss=0.05654, over 1607076.04 frames. ], batch size: 24, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:53:05,096 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4676, 1.3705, 1.7672, 1.2292, 1.0872, 1.7756, 0.2056, 1.0917], + device='cuda:2'), covar=tensor([0.1397, 0.1114, 0.0426, 0.0824, 0.2443, 0.0427, 0.1821, 0.1138], + device='cuda:2'), in_proj_covar=tensor([0.0201, 0.0205, 0.0137, 0.0223, 0.0278, 0.0147, 0.0174, 0.0199], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 00:53:31,691 INFO [train.py:901] (2/4) Epoch 28, batch 5650, loss[loss=0.2042, simple_loss=0.2808, pruned_loss=0.06384, over 7442.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2807, pruned_loss=0.05681, over 1609139.97 frames. ], batch size: 73, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:53:41,201 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-09 00:53:44,202 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=223906.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 00:53:44,787 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8185, 2.0371, 1.6911, 2.5978, 1.2806, 1.5707, 1.9492, 1.9823], + device='cuda:2'), covar=tensor([0.0769, 0.0705, 0.0866, 0.0384, 0.0984, 0.1265, 0.0693, 0.0739], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0196, 0.0246, 0.0215, 0.0205, 0.0249, 0.0252, 0.0206], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 00:53:51,308 INFO [optim.py:369] (2/4) 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 00:54:01,598 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=223931.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 00:54:06,155 INFO [train.py:901] (2/4) Epoch 28, batch 5700, loss[loss=0.1805, simple_loss=0.2543, pruned_loss=0.05338, over 7439.00 frames. ], tot_loss[loss=0.1953, simple_loss=0.279, pruned_loss=0.0558, over 1607151.44 frames. ], batch size: 17, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:54:07,517 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=223940.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:54:10,615 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.55 vs. limit=2.0 +2023-02-09 00:54:16,420 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-09 00:54:43,234 INFO [train.py:901] (2/4) Epoch 28, batch 5750, loss[loss=0.1812, simple_loss=0.2688, pruned_loss=0.04675, over 7820.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.2798, pruned_loss=0.05623, over 1610243.06 frames. ], batch size: 20, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:54:48,703 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-09 00:55:04,150 INFO [optim.py:369] (2/4) 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 00:55:18,773 INFO [train.py:901] (2/4) Epoch 28, batch 5800, loss[loss=0.2128, simple_loss=0.3061, pruned_loss=0.05973, over 8330.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2804, pruned_loss=0.05635, over 1614053.38 frames. ], batch size: 25, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:55:22,322 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8245, 1.3272, 4.0142, 1.4441, 3.5592, 3.3761, 3.6786, 3.5671], + device='cuda:2'), covar=tensor([0.0728, 0.5003, 0.0727, 0.4545, 0.1296, 0.1154, 0.0640, 0.0814], + device='cuda:2'), in_proj_covar=tensor([0.0683, 0.0665, 0.0737, 0.0661, 0.0742, 0.0636, 0.0643, 0.0718], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 00:55:30,595 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=224055.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:55:31,246 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.9024, 1.2176, 3.2653, 1.1659, 2.5403, 2.5793, 2.9429, 2.9314], + device='cuda:2'), covar=tensor([0.1997, 0.6292, 0.1605, 0.5584, 0.3095, 0.2297, 0.1445, 0.1589], + device='cuda:2'), in_proj_covar=tensor([0.0683, 0.0664, 0.0737, 0.0661, 0.0742, 0.0636, 0.0643, 0.0718], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 00:55:43,050 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5707, 2.0415, 3.2062, 1.4409, 2.3217, 2.0331, 1.7090, 2.5338], + device='cuda:2'), covar=tensor([0.2018, 0.2663, 0.0885, 0.4796, 0.2097, 0.3356, 0.2531, 0.2282], + device='cuda:2'), in_proj_covar=tensor([0.0541, 0.0636, 0.0564, 0.0671, 0.0663, 0.0612, 0.0563, 0.0647], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 00:55:47,089 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1008, 3.5491, 2.2909, 2.7533, 2.5942, 2.0025, 2.6284, 2.9399], + device='cuda:2'), covar=tensor([0.1726, 0.0393, 0.1115, 0.0816, 0.0853, 0.1499, 0.1077, 0.1094], + device='cuda:2'), in_proj_covar=tensor([0.0359, 0.0243, 0.0341, 0.0314, 0.0303, 0.0346, 0.0350, 0.0322], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 00:55:55,795 INFO [train.py:901] (2/4) Epoch 28, batch 5850, loss[loss=0.1945, simple_loss=0.2764, pruned_loss=0.05626, over 7822.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2816, pruned_loss=0.05666, over 1620214.89 frames. ], batch size: 20, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:56:15,667 INFO [optim.py:369] (2/4) 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 00:56:30,204 INFO [train.py:901] (2/4) Epoch 28, batch 5900, loss[loss=0.2651, simple_loss=0.3288, pruned_loss=0.1008, over 6843.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2822, pruned_loss=0.05669, over 1621763.95 frames. ], batch size: 71, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:57:06,188 INFO [train.py:901] (2/4) Epoch 28, batch 5950, loss[loss=0.188, simple_loss=0.2751, pruned_loss=0.05047, over 8251.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2839, pruned_loss=0.05773, over 1620030.04 frames. ], batch size: 22, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:57:25,382 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 +2023-02-09 00:57:28,300 INFO [optim.py:369] (2/4) 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 00:57:37,732 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=224230.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:57:43,109 INFO [train.py:901] (2/4) Epoch 28, batch 6000, loss[loss=0.1925, simple_loss=0.2833, pruned_loss=0.05087, over 8469.00 frames. ], tot_loss[loss=0.2, simple_loss=0.284, pruned_loss=0.058, over 1622477.31 frames. ], batch size: 27, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:57:43,110 INFO [train.py:926] (2/4) Computing validation loss +2023-02-09 00:57:56,800 INFO [train.py:935] (2/4) Epoch 28, validation: loss=0.1714, simple_loss=0.2708, pruned_loss=0.03603, over 944034.00 frames. +2023-02-09 00:57:56,801 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6724MB +2023-02-09 00:57:59,129 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5185, 2.3920, 3.0804, 2.5031, 2.9313, 2.5671, 2.4108, 2.0320], + device='cuda:2'), covar=tensor([0.5476, 0.5206, 0.2279, 0.4177, 0.2850, 0.3263, 0.1883, 0.5693], + device='cuda:2'), in_proj_covar=tensor([0.0973, 0.1035, 0.0841, 0.1004, 0.1028, 0.0940, 0.0776, 0.0859], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 00:58:07,091 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5370, 4.4987, 4.1451, 2.2485, 4.0916, 4.0985, 4.0877, 3.9364], + device='cuda:2'), covar=tensor([0.0641, 0.0459, 0.0851, 0.4029, 0.0756, 0.1021, 0.1141, 0.0812], + device='cuda:2'), in_proj_covar=tensor([0.0545, 0.0457, 0.0451, 0.0559, 0.0445, 0.0466, 0.0442, 0.0408], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 00:58:33,310 INFO [train.py:901] (2/4) Epoch 28, batch 6050, loss[loss=0.1761, simple_loss=0.2618, pruned_loss=0.04523, over 7651.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2835, pruned_loss=0.05764, over 1619469.71 frames. ], batch size: 19, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:58:43,552 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6525, 2.0646, 3.2527, 1.4238, 2.4758, 2.1279, 1.7219, 2.6089], + device='cuda:2'), covar=tensor([0.1916, 0.2807, 0.0844, 0.4849, 0.1989, 0.3252, 0.2523, 0.2279], + device='cuda:2'), in_proj_covar=tensor([0.0541, 0.0637, 0.0565, 0.0671, 0.0665, 0.0614, 0.0565, 0.0649], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 00:58:48,811 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.77 vs. limit=2.0 +2023-02-09 00:58:49,931 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=224311.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:58:52,994 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-09 00:58:54,045 INFO [optim.py:369] (2/4) 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 00:59:08,857 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=224336.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 00:59:10,070 INFO [train.py:901] (2/4) Epoch 28, batch 6100, loss[loss=0.1805, simple_loss=0.261, pruned_loss=0.04997, over 7326.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2827, pruned_loss=0.05726, over 1615813.26 frames. ], batch size: 16, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:59:26,304 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-09 00:59:46,741 INFO [train.py:901] (2/4) Epoch 28, batch 6150, loss[loss=0.1945, simple_loss=0.2809, pruned_loss=0.05405, over 8461.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2828, pruned_loss=0.05737, over 1619520.32 frames. ], batch size: 29, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 00:59:54,611 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5274, 1.9468, 2.0713, 1.2091, 2.1326, 1.5453, 0.6053, 1.9010], + device='cuda:2'), covar=tensor([0.0767, 0.0423, 0.0293, 0.0744, 0.0478, 0.1060, 0.1004, 0.0367], + device='cuda:2'), in_proj_covar=tensor([0.0471, 0.0409, 0.0363, 0.0458, 0.0393, 0.0550, 0.0402, 0.0440], + device='cuda:2'), 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:2') +2023-02-09 01:00:06,960 INFO [optim.py:369] (2/4) 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 01:00:21,346 INFO [train.py:901] (2/4) Epoch 28, batch 6200, loss[loss=0.1949, simple_loss=0.29, pruned_loss=0.04994, over 8322.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2826, pruned_loss=0.05753, over 1615691.98 frames. ], batch size: 25, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 01:00:58,234 INFO [train.py:901] (2/4) Epoch 28, batch 6250, loss[loss=0.1702, simple_loss=0.2441, pruned_loss=0.04819, over 7550.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2822, pruned_loss=0.0577, over 1605664.57 frames. ], batch size: 18, lr: 2.68e-03, grad_scale: 8.0 +2023-02-09 01:00:59,402 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.54 vs. limit=5.0 +2023-02-09 01:01:18,589 INFO [optim.py:369] (2/4) 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 01:01:33,268 INFO [train.py:901] (2/4) Epoch 28, batch 6300, loss[loss=0.1572, simple_loss=0.2386, pruned_loss=0.03792, over 6799.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2819, pruned_loss=0.05724, over 1606961.78 frames. ], batch size: 15, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:02:00,129 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=224574.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:02:04,363 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=224580.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:02:10,460 INFO [train.py:901] (2/4) Epoch 28, batch 6350, loss[loss=0.182, simple_loss=0.2679, pruned_loss=0.0481, over 7817.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2835, pruned_loss=0.05787, over 1612240.34 frames. ], batch size: 20, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:02:30,928 INFO [optim.py:369] (2/4) 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 01:02:45,884 INFO [train.py:901] (2/4) Epoch 28, batch 6400, loss[loss=0.2331, simple_loss=0.3191, pruned_loss=0.0736, over 8470.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2826, pruned_loss=0.05768, over 1611876.79 frames. ], batch size: 25, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:03:16,422 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-09 01:03:21,510 INFO [train.py:901] (2/4) Epoch 28, batch 6450, loss[loss=0.1882, simple_loss=0.2812, pruned_loss=0.04763, over 8187.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.282, pruned_loss=0.05788, over 1610991.59 frames. ], batch size: 23, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:03:22,394 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=224689.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:03:43,003 INFO [optim.py:369] (2/4) 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 01:03:44,832 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-09 01:03:56,389 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1694, 2.4029, 2.5030, 1.6603, 2.7036, 1.8585, 1.6176, 2.1539], + device='cuda:2'), covar=tensor([0.0929, 0.0451, 0.0346, 0.0834, 0.0531, 0.0870, 0.1093, 0.0594], + device='cuda:2'), in_proj_covar=tensor([0.0475, 0.0412, 0.0366, 0.0461, 0.0396, 0.0555, 0.0406, 0.0444], + device='cuda:2'), out_proj_covar=tensor([1.2572e-04, 1.0680e-04, 9.5128e-05, 1.2062e-04, 1.0360e-04, 1.5473e-04, + 1.0836e-04, 1.1623e-04], device='cuda:2') +2023-02-09 01:03:57,598 INFO [train.py:901] (2/4) Epoch 28, batch 6500, loss[loss=0.1596, simple_loss=0.2365, pruned_loss=0.04136, over 7699.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2814, pruned_loss=0.0573, over 1611520.85 frames. ], batch size: 18, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:04:02,103 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=224744.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:04:11,987 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=224758.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:04:32,256 INFO [train.py:901] (2/4) Epoch 28, batch 6550, loss[loss=0.2419, simple_loss=0.2947, pruned_loss=0.09452, over 7804.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2824, pruned_loss=0.05795, over 1615194.98 frames. ], batch size: 19, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:04:47,798 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-09 01:04:53,991 INFO [optim.py:369] (2/4) 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 01:05:08,030 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-09 01:05:09,349 INFO [train.py:901] (2/4) Epoch 28, batch 6600, loss[loss=0.2189, simple_loss=0.2842, pruned_loss=0.0768, over 7655.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2825, pruned_loss=0.05831, over 1609985.26 frames. ], batch size: 19, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:05:17,204 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1702, 1.8765, 2.5274, 1.6443, 1.6316, 2.5025, 1.3729, 2.0599], + device='cuda:2'), covar=tensor([0.1533, 0.1018, 0.0292, 0.1034, 0.1829, 0.0369, 0.1478, 0.1121], + device='cuda:2'), in_proj_covar=tensor([0.0204, 0.0209, 0.0138, 0.0226, 0.0280, 0.0149, 0.0176, 0.0201], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 01:05:44,396 INFO [train.py:901] (2/4) Epoch 28, batch 6650, loss[loss=0.1533, simple_loss=0.2417, pruned_loss=0.03245, over 7433.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2818, pruned_loss=0.05785, over 1612256.80 frames. ], batch size: 17, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:06:04,781 INFO [optim.py:369] (2/4) 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 01:06:10,966 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=224924.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:06:11,639 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.2118, 4.2120, 3.7742, 1.9960, 3.7221, 3.8509, 3.8392, 3.6960], + device='cuda:2'), covar=tensor([0.0751, 0.0546, 0.1040, 0.4455, 0.0934, 0.1190, 0.1199, 0.0898], + device='cuda:2'), in_proj_covar=tensor([0.0550, 0.0462, 0.0456, 0.0566, 0.0448, 0.0471, 0.0448, 0.0414], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 01:06:17,465 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.91 vs. limit=2.0 +2023-02-09 01:06:21,167 INFO [train.py:901] (2/4) Epoch 28, batch 6700, loss[loss=0.1621, simple_loss=0.2532, pruned_loss=0.03548, over 7975.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2813, pruned_loss=0.05749, over 1611929.07 frames. ], batch size: 21, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:06:26,235 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=224945.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:06:30,480 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0196, 1.6339, 1.8353, 1.5700, 0.9385, 1.6735, 1.7976, 1.7082], + device='cuda:2'), covar=tensor([0.0539, 0.1193, 0.1590, 0.1394, 0.0602, 0.1405, 0.0696, 0.0640], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0154, 0.0190, 0.0161, 0.0102, 0.0164, 0.0113, 0.0147], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-09 01:06:44,447 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 28, batch 6750, loss[loss=0.1876, simple_loss=0.2812, pruned_loss=0.04701, over 8337.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.2794, pruned_loss=0.05621, over 1607215.69 frames. ], batch size: 25, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:06:57,468 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.72 vs. limit=2.0 +2023-02-09 01:06:58,657 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0267, 1.6271, 1.3885, 1.5424, 1.3128, 1.2543, 1.2919, 1.2562], + device='cuda:2'), covar=tensor([0.1282, 0.0536, 0.1453, 0.0669, 0.0862, 0.1661, 0.0984, 0.0984], + device='cuda:2'), in_proj_covar=tensor([0.0362, 0.0247, 0.0345, 0.0317, 0.0304, 0.0350, 0.0354, 0.0327], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 01:07:01,383 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3568, 2.6288, 2.9662, 1.7769, 3.1724, 2.0052, 1.6130, 2.3302], + device='cuda:2'), covar=tensor([0.0890, 0.0413, 0.0340, 0.0899, 0.0530, 0.0891, 0.1072, 0.0640], + device='cuda:2'), in_proj_covar=tensor([0.0473, 0.0409, 0.0364, 0.0459, 0.0396, 0.0552, 0.0403, 0.0442], + device='cuda:2'), 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:2') +2023-02-09 01:07:17,010 INFO [optim.py:369] (2/4) 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 01:07:28,015 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-09 01:07:32,093 INFO [train.py:901] (2/4) Epoch 28, batch 6800, loss[loss=0.1805, simple_loss=0.2637, pruned_loss=0.04859, over 7424.00 frames. ], tot_loss[loss=0.1954, simple_loss=0.2793, pruned_loss=0.05578, over 1606362.12 frames. ], batch size: 17, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:07:32,951 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225039.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:07:57,290 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7104, 1.4744, 1.8180, 1.4815, 0.8476, 1.6171, 1.6019, 1.5257], + device='cuda:2'), covar=tensor([0.0591, 0.1298, 0.1601, 0.1465, 0.0609, 0.1421, 0.0725, 0.0655], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0153, 0.0189, 0.0161, 0.0102, 0.0163, 0.0113, 0.0147], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-09 01:08:08,382 INFO [train.py:901] (2/4) Epoch 28, batch 6850, loss[loss=0.1663, simple_loss=0.2542, pruned_loss=0.03916, over 8242.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2802, pruned_loss=0.05634, over 1611848.17 frames. ], batch size: 22, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:08:08,458 INFO [zipformer.py:1185] (2/4) 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] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=225102.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:08:18,711 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-09 01:08:28,500 INFO [optim.py:369] (2/4) 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 01:08:31,368 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 28, batch 6900, loss[loss=0.2272, simple_loss=0.3064, pruned_loss=0.07401, over 7816.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2805, pruned_loss=0.05684, over 1610354.88 frames. ], batch size: 20, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:08:47,491 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.69 vs. limit=2.0 +2023-02-09 01:08:58,663 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225160.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:09:18,984 INFO [train.py:901] (2/4) Epoch 28, batch 6950, loss[loss=0.2077, simple_loss=0.2838, pruned_loss=0.06583, over 7430.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2805, pruned_loss=0.05703, over 1608127.67 frames. ], batch size: 17, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:09:30,330 INFO [zipformer.py:1185] (2/4) 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] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-09 01:09:40,023 INFO [optim.py:369] (2/4) 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 01:09:40,234 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225217.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:09:54,781 INFO [train.py:901] (2/4) Epoch 28, batch 7000, loss[loss=0.2061, simple_loss=0.2886, pruned_loss=0.06185, over 8554.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2798, pruned_loss=0.05634, over 1609553.91 frames. ], batch size: 34, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:10:31,221 INFO [train.py:901] (2/4) Epoch 28, batch 7050, loss[loss=0.2236, simple_loss=0.3085, pruned_loss=0.06937, over 8528.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2811, pruned_loss=0.05688, over 1609482.60 frames. ], batch size: 48, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:10:36,315 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225295.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:10:52,614 INFO [optim.py:369] (2/4) 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 01:10:55,649 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225320.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:11:08,202 INFO [train.py:901] (2/4) Epoch 28, batch 7100, loss[loss=0.1865, simple_loss=0.2826, pruned_loss=0.04524, over 8343.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2813, pruned_loss=0.05632, over 1614161.65 frames. ], batch size: 24, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:11:43,049 INFO [train.py:901] (2/4) Epoch 28, batch 7150, loss[loss=0.2163, simple_loss=0.3053, pruned_loss=0.06366, over 8360.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2807, pruned_loss=0.05618, over 1613436.19 frames. ], batch size: 24, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:12:05,401 INFO [optim.py:369] (2/4) 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 01:12:21,601 INFO [train.py:901] (2/4) Epoch 28, batch 7200, loss[loss=0.1982, simple_loss=0.284, pruned_loss=0.05622, over 8545.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2819, pruned_loss=0.05657, over 1614425.68 frames. ], batch size: 28, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:12:35,169 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225457.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:12:36,585 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225459.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:12:40,503 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=225465.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:12:41,417 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2653, 2.0354, 2.6158, 2.2520, 2.5578, 2.3323, 2.1563, 1.4690], + device='cuda:2'), covar=tensor([0.5662, 0.5186, 0.2121, 0.3815, 0.2577, 0.3256, 0.1950, 0.5483], + device='cuda:2'), in_proj_covar=tensor([0.0969, 0.1033, 0.0841, 0.1000, 0.1028, 0.0938, 0.0775, 0.0855], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 01:12:44,884 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2582, 3.6790, 2.4474, 2.9153, 2.9080, 2.1478, 2.8554, 3.2205], + device='cuda:2'), covar=tensor([0.1532, 0.0372, 0.1055, 0.0735, 0.0771, 0.1416, 0.1017, 0.1032], + device='cuda:2'), in_proj_covar=tensor([0.0362, 0.0247, 0.0345, 0.0317, 0.0304, 0.0350, 0.0353, 0.0325], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 01:12:46,185 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225473.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:12:47,076 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-09 01:12:51,112 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225480.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:12:53,922 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225484.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:12:56,557 INFO [train.py:901] (2/4) Epoch 28, batch 7250, loss[loss=0.186, simple_loss=0.265, pruned_loss=0.05354, over 7272.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2824, pruned_loss=0.05722, over 1617612.07 frames. ], batch size: 16, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:13:03,612 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225498.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:13:07,603 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=225504.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:13:16,360 INFO [optim.py:369] (2/4) 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 01:13:21,986 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225523.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:13:32,777 INFO [train.py:901] (2/4) Epoch 28, batch 7300, loss[loss=0.1863, simple_loss=0.2833, pruned_loss=0.04469, over 8323.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2819, pruned_loss=0.05702, over 1617277.04 frames. ], batch size: 25, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:13:34,184 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225540.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:14:00,395 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225577.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 01:14:02,421 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225580.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:14:07,681 INFO [train.py:901] (2/4) Epoch 28, batch 7350, loss[loss=0.1577, simple_loss=0.2421, pruned_loss=0.03663, over 8092.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2812, pruned_loss=0.05681, over 1611998.23 frames. ], batch size: 21, lr: 2.67e-03, grad_scale: 8.0 +2023-02-09 01:14:24,565 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-09 01:14:27,903 INFO [optim.py:369] (2/4) 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 01:14:29,528 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225619.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:14:42,945 INFO [train.py:901] (2/4) Epoch 28, batch 7400, loss[loss=0.192, simple_loss=0.2846, pruned_loss=0.04967, over 8475.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2825, pruned_loss=0.05742, over 1615287.15 frames. ], batch size: 27, lr: 2.67e-03, grad_scale: 16.0 +2023-02-09 01:14:42,955 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-09 01:14:48,689 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2144, 1.4495, 4.2300, 1.9238, 2.4715, 4.7651, 4.8802, 4.1441], + device='cuda:2'), covar=tensor([0.1321, 0.2100, 0.0277, 0.2129, 0.1215, 0.0190, 0.0430, 0.0519], + device='cuda:2'), in_proj_covar=tensor([0.0309, 0.0327, 0.0295, 0.0325, 0.0326, 0.0277, 0.0443, 0.0309], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:2') +2023-02-09 01:15:18,724 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.3162, 1.1787, 3.3836, 1.0766, 3.0315, 2.8110, 3.1095, 3.0042], + device='cuda:2'), covar=tensor([0.0738, 0.4493, 0.0825, 0.4473, 0.1235, 0.1135, 0.0741, 0.0877], + device='cuda:2'), in_proj_covar=tensor([0.0683, 0.0664, 0.0735, 0.0660, 0.0747, 0.0636, 0.0645, 0.0717], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 01:15:19,316 INFO [train.py:901] (2/4) Epoch 28, batch 7450, loss[loss=0.1873, simple_loss=0.2766, pruned_loss=0.04894, over 8194.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2829, pruned_loss=0.05764, over 1617703.65 frames. ], batch size: 23, lr: 2.67e-03, grad_scale: 16.0 +2023-02-09 01:15:23,689 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8750, 1.6884, 2.1775, 1.7206, 1.2913, 1.9182, 2.4222, 2.2922], + device='cuda:2'), covar=tensor([0.0446, 0.1209, 0.1516, 0.1383, 0.0553, 0.1343, 0.0573, 0.0576], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0153, 0.0189, 0.0160, 0.0101, 0.0162, 0.0113, 0.0146], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-09 01:15:25,033 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-09 01:15:40,013 INFO [optim.py:369] (2/4) 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 01:15:54,482 INFO [train.py:901] (2/4) Epoch 28, batch 7500, loss[loss=0.1797, simple_loss=0.2704, pruned_loss=0.04451, over 8361.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2826, pruned_loss=0.05716, over 1618944.34 frames. ], batch size: 24, lr: 2.67e-03, grad_scale: 16.0 +2023-02-09 01:16:32,496 INFO [train.py:901] (2/4) Epoch 28, batch 7550, loss[loss=0.1936, simple_loss=0.2815, pruned_loss=0.05287, over 8457.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2828, pruned_loss=0.05761, over 1616876.00 frames. ], batch size: 25, lr: 2.67e-03, grad_scale: 16.0 +2023-02-09 01:16:39,078 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4821, 2.5078, 1.8701, 2.1404, 2.0995, 1.6074, 2.0145, 2.1305], + device='cuda:2'), covar=tensor([0.1656, 0.0455, 0.1292, 0.0719, 0.0830, 0.1576, 0.1100, 0.1066], + device='cuda:2'), in_proj_covar=tensor([0.0356, 0.0244, 0.0341, 0.0312, 0.0300, 0.0345, 0.0348, 0.0320], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 01:16:41,875 INFO [zipformer.py:1185] (2/4) 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] (2/4) 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 01:16:57,651 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=225824.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:17:05,948 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225836.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:17:07,133 INFO [train.py:901] (2/4) Epoch 28, batch 7600, loss[loss=0.2088, simple_loss=0.2978, pruned_loss=0.05997, over 8460.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2822, pruned_loss=0.05735, over 1616163.33 frames. ], batch size: 29, lr: 2.67e-03, grad_scale: 16.0 +2023-02-09 01:17:23,278 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225861.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:17:27,277 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=225867.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:17:34,157 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=225875.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:17:37,631 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225880.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:17:40,887 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 28, batch 7650, loss[loss=0.2363, simple_loss=0.3148, pruned_loss=0.07893, over 8599.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2825, pruned_loss=0.05742, over 1613835.07 frames. ], batch size: 39, lr: 2.67e-03, grad_scale: 16.0 +2023-02-09 01:17:51,600 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=225900.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:18:01,184 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=225913.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:18:03,348 INFO [zipformer.py:1185] (2/4) 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] (2/4) 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] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=225921.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 01:18:18,658 INFO [train.py:901] (2/4) Epoch 28, batch 7700, loss[loss=0.2505, simple_loss=0.3272, pruned_loss=0.08689, over 8372.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2821, pruned_loss=0.05812, over 1613033.77 frames. ], batch size: 24, lr: 2.67e-03, grad_scale: 16.0 +2023-02-09 01:18:19,536 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225939.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:18:31,340 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-09 01:18:44,244 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-09 01:18:44,500 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4357, 2.3794, 2.9974, 2.4641, 3.0263, 2.4984, 2.4470, 1.9285], + device='cuda:2'), covar=tensor([0.5783, 0.5388, 0.2307, 0.4196, 0.2788, 0.3527, 0.1934, 0.6016], + device='cuda:2'), in_proj_covar=tensor([0.0964, 0.1031, 0.0836, 0.0999, 0.1025, 0.0936, 0.0772, 0.0855], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 01:18:49,255 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225982.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:18:53,157 INFO [train.py:901] (2/4) Epoch 28, batch 7750, loss[loss=0.2138, simple_loss=0.2956, pruned_loss=0.06603, over 8367.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2807, pruned_loss=0.05708, over 1612316.69 frames. ], batch size: 24, lr: 2.67e-03, grad_scale: 16.0 +2023-02-09 01:19:01,896 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=225999.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:19:15,716 INFO [optim.py:369] (2/4) 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 01:19:29,668 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226036.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 01:19:30,737 INFO [train.py:901] (2/4) Epoch 28, batch 7800, loss[loss=0.2565, simple_loss=0.3301, pruned_loss=0.0915, over 8242.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2805, pruned_loss=0.05685, over 1614978.10 frames. ], batch size: 24, lr: 2.67e-03, grad_scale: 16.0 +2023-02-09 01:19:45,517 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226059.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:20:05,554 INFO [train.py:901] (2/4) Epoch 28, batch 7850, loss[loss=0.2002, simple_loss=0.2858, pruned_loss=0.05733, over 8245.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2803, pruned_loss=0.05698, over 1608779.18 frames. ], batch size: 22, lr: 2.67e-03, grad_scale: 16.0 +2023-02-09 01:20:25,274 INFO [optim.py:369] (2/4) 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 01:20:39,675 INFO [train.py:901] (2/4) Epoch 28, batch 7900, loss[loss=0.217, simple_loss=0.304, pruned_loss=0.065, over 8450.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2805, pruned_loss=0.05693, over 1608149.97 frames. ], batch size: 27, lr: 2.67e-03, grad_scale: 16.0 +2023-02-09 01:21:02,887 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226172.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:21:13,378 INFO [train.py:901] (2/4) Epoch 28, batch 7950, loss[loss=0.1839, simple_loss=0.2483, pruned_loss=0.0598, over 7677.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2804, pruned_loss=0.05732, over 1607338.75 frames. ], batch size: 18, lr: 2.67e-03, grad_scale: 16.0 +2023-02-09 01:21:18,309 INFO [zipformer.py:1185] (2/4) 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] (2/4) 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] (2/4) 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 01:21:35,312 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226220.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:21:36,606 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0449, 1.3951, 1.6572, 1.3155, 0.8969, 1.4264, 1.6309, 1.5062], + device='cuda:2'), covar=tensor([0.0580, 0.1319, 0.1746, 0.1567, 0.0675, 0.1531, 0.0770, 0.0706], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0153, 0.0189, 0.0161, 0.0102, 0.0163, 0.0113, 0.0146], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-09 01:21:37,938 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 28, batch 8000, loss[loss=0.1965, simple_loss=0.2921, pruned_loss=0.05044, over 8337.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.281, pruned_loss=0.05714, over 1613601.75 frames. ], batch size: 25, lr: 2.66e-03, grad_scale: 16.0 +2023-02-09 01:21:47,896 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226238.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:21:52,826 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5194, 1.5988, 4.7396, 1.8109, 4.1352, 3.9312, 4.2882, 4.1619], + device='cuda:2'), covar=tensor([0.0601, 0.4871, 0.0537, 0.4363, 0.1128, 0.1044, 0.0602, 0.0685], + device='cuda:2'), in_proj_covar=tensor([0.0688, 0.0668, 0.0738, 0.0664, 0.0753, 0.0641, 0.0648, 0.0721], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 01:21:57,012 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=226251.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:21:59,704 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226255.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:22:00,950 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=226257.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:22:05,094 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226263.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:22:17,276 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226280.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:22:22,560 INFO [train.py:901] (2/4) Epoch 28, batch 8050, loss[loss=0.1741, simple_loss=0.2547, pruned_loss=0.04672, over 7551.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2795, pruned_loss=0.05706, over 1599012.39 frames. ], batch size: 18, lr: 2.66e-03, grad_scale: 16.0 +2023-02-09 01:22:25,396 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226292.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 01:22:30,116 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.3583, 1.6217, 4.6187, 1.8371, 4.0655, 3.8899, 4.1854, 4.0557], + device='cuda:2'), covar=tensor([0.0681, 0.4464, 0.0547, 0.4190, 0.1144, 0.0978, 0.0586, 0.0719], + device='cuda:2'), in_proj_covar=tensor([0.0685, 0.0665, 0.0735, 0.0661, 0.0749, 0.0638, 0.0644, 0.0718], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 01:22:37,472 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-09 01:22:42,673 INFO [optim.py:369] (2/4) 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 01:22:42,888 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226317.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 01:22:57,909 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-09 01:23:01,703 INFO [train.py:901] (2/4) Epoch 29, batch 0, loss[loss=0.2081, simple_loss=0.2945, pruned_loss=0.0608, over 8577.00 frames. ], tot_loss[loss=0.2081, simple_loss=0.2945, pruned_loss=0.0608, over 8577.00 frames. ], batch size: 31, lr: 2.62e-03, grad_scale: 16.0 +2023-02-09 01:23:01,703 INFO [train.py:926] (2/4) Computing validation loss +2023-02-09 01:23:13,271 INFO [train.py:935] (2/4) Epoch 29, validation: loss=0.1705, simple_loss=0.2705, pruned_loss=0.03528, over 944034.00 frames. +2023-02-09 01:23:13,272 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6724MB +2023-02-09 01:23:26,221 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226339.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:23:29,656 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-09 01:23:39,591 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.81 vs. limit=5.0 +2023-02-09 01:23:49,914 INFO [train.py:901] (2/4) Epoch 29, batch 50, loss[loss=0.2077, simple_loss=0.2976, pruned_loss=0.05894, over 8343.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2836, pruned_loss=0.0559, over 369403.14 frames. ], batch size: 25, lr: 2.62e-03, grad_scale: 16.0 +2023-02-09 01:23:50,828 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226372.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:23:53,662 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4468, 1.7442, 1.6877, 1.1210, 1.7523, 1.4578, 0.3220, 1.6395], + device='cuda:2'), covar=tensor([0.0542, 0.0433, 0.0366, 0.0557, 0.0501, 0.0938, 0.1000, 0.0321], + device='cuda:2'), in_proj_covar=tensor([0.0478, 0.0414, 0.0368, 0.0462, 0.0398, 0.0555, 0.0406, 0.0443], + device='cuda:2'), out_proj_covar=tensor([1.2657e-04, 1.0724e-04, 9.5925e-05, 1.2073e-04, 1.0402e-04, 1.5462e-04, + 1.0826e-04, 1.1606e-04], device='cuda:2') +2023-02-09 01:24:06,020 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-09 01:24:12,517 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=226403.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:24:22,926 INFO [optim.py:369] (2/4) 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] (2/4) Epoch 29, batch 100, loss[loss=0.1641, simple_loss=0.2391, pruned_loss=0.04452, over 7799.00 frames. ], tot_loss[loss=0.196, simple_loss=0.28, pruned_loss=0.05599, over 643795.97 frames. ], batch size: 19, lr: 2.62e-03, grad_scale: 16.0 +2023-02-09 01:24:30,604 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-09 01:24:33,770 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1790, 1.9889, 2.4592, 2.1453, 2.4405, 2.2875, 2.1389, 1.4472], + device='cuda:2'), covar=tensor([0.5886, 0.5097, 0.2276, 0.3901, 0.2614, 0.3238, 0.1934, 0.5408], + device='cuda:2'), in_proj_covar=tensor([0.0969, 0.1034, 0.0840, 0.1003, 0.1030, 0.0940, 0.0776, 0.0858], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 01:25:02,686 INFO [train.py:901] (2/4) Epoch 29, batch 150, loss[loss=0.2131, simple_loss=0.3018, pruned_loss=0.0622, over 8106.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2816, pruned_loss=0.0569, over 857782.18 frames. ], batch size: 23, lr: 2.62e-03, grad_scale: 16.0 +2023-02-09 01:25:34,579 INFO [optim.py:369] (2/4) 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 01:25:35,509 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 29, batch 200, loss[loss=0.1809, simple_loss=0.2678, pruned_loss=0.04704, over 8444.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2813, pruned_loss=0.057, over 1024342.56 frames. ], batch size: 27, lr: 2.62e-03, grad_scale: 16.0 +2023-02-09 01:26:12,528 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8383, 1.6968, 2.2928, 1.7209, 1.0727, 1.8245, 2.3151, 2.4818], + device='cuda:2'), covar=tensor([0.0488, 0.1217, 0.1541, 0.1393, 0.0610, 0.1404, 0.0588, 0.0543], + device='cuda:2'), in_proj_covar=tensor([0.0099, 0.0153, 0.0189, 0.0161, 0.0102, 0.0163, 0.0113, 0.0147], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-09 01:26:12,613 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5988, 2.5297, 3.2262, 2.5801, 3.1061, 2.7123, 2.6048, 2.1961], + device='cuda:2'), covar=tensor([0.5575, 0.5163, 0.2189, 0.4515, 0.2883, 0.3357, 0.1777, 0.5690], + device='cuda:2'), in_proj_covar=tensor([0.0965, 0.1030, 0.0837, 0.0999, 0.1025, 0.0935, 0.0773, 0.0852], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 01:26:13,673 INFO [train.py:901] (2/4) Epoch 29, batch 250, loss[loss=0.248, simple_loss=0.3163, pruned_loss=0.08984, over 6709.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2822, pruned_loss=0.05714, over 1153565.74 frames. ], batch size: 71, lr: 2.62e-03, grad_scale: 16.0 +2023-02-09 01:26:26,049 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-09 01:26:31,141 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=226595.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:26:31,292 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226595.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:26:33,828 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-09 01:26:46,401 INFO [optim.py:369] (2/4) 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 01:26:48,780 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226620.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:26:49,256 INFO [train.py:901] (2/4) Epoch 29, batch 300, loss[loss=0.1948, simple_loss=0.276, pruned_loss=0.05684, over 7804.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2834, pruned_loss=0.05787, over 1257046.00 frames. ], batch size: 20, lr: 2.62e-03, grad_scale: 16.0 +2023-02-09 01:26:54,305 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226628.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:27:12,768 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226653.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:27:25,798 INFO [train.py:901] (2/4) Epoch 29, batch 350, loss[loss=0.1858, simple_loss=0.2709, pruned_loss=0.05031, over 8122.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.283, pruned_loss=0.05757, over 1336590.78 frames. ], batch size: 22, lr: 2.62e-03, grad_scale: 16.0 +2023-02-09 01:27:45,007 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0676, 1.2591, 1.1968, 0.7047, 1.2123, 1.0239, 0.0840, 1.2436], + device='cuda:2'), covar=tensor([0.0512, 0.0449, 0.0427, 0.0698, 0.0527, 0.1151, 0.1034, 0.0408], + device='cuda:2'), in_proj_covar=tensor([0.0483, 0.0418, 0.0372, 0.0466, 0.0401, 0.0559, 0.0408, 0.0447], + device='cuda:2'), out_proj_covar=tensor([1.2785e-04, 1.0831e-04, 9.7018e-05, 1.2160e-04, 1.0475e-04, 1.5584e-04, + 1.0894e-04, 1.1705e-04], device='cuda:2') +2023-02-09 01:27:46,741 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4838, 1.9207, 2.6519, 1.4085, 1.9370, 1.8551, 1.6475, 1.9488], + device='cuda:2'), covar=tensor([0.1981, 0.2499, 0.0861, 0.4686, 0.2000, 0.3358, 0.2450, 0.2273], + device='cuda:2'), in_proj_covar=tensor([0.0540, 0.0637, 0.0562, 0.0671, 0.0664, 0.0614, 0.0564, 0.0646], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 01:27:54,190 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=226710.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:27:58,932 INFO [optim.py:369] (2/4) 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 01:28:01,712 INFO [train.py:901] (2/4) Epoch 29, batch 400, loss[loss=0.1722, simple_loss=0.2493, pruned_loss=0.04759, over 7708.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2836, pruned_loss=0.05751, over 1398681.69 frames. ], batch size: 18, lr: 2.61e-03, grad_scale: 16.0 +2023-02-09 01:28:37,571 INFO [train.py:901] (2/4) Epoch 29, batch 450, loss[loss=0.1765, simple_loss=0.247, pruned_loss=0.05304, over 7706.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2819, pruned_loss=0.05643, over 1448358.11 frames. ], batch size: 18, lr: 2.61e-03, grad_scale: 16.0 +2023-02-09 01:28:38,565 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.07 vs. limit=5.0 +2023-02-09 01:28:39,871 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226774.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:28:53,563 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7529, 1.6833, 2.3417, 1.4782, 1.3593, 2.3020, 0.5301, 1.5120], + device='cuda:2'), covar=tensor([0.1481, 0.1149, 0.0305, 0.0973, 0.2342, 0.0352, 0.1769, 0.1171], + device='cuda:2'), in_proj_covar=tensor([0.0202, 0.0208, 0.0138, 0.0224, 0.0280, 0.0148, 0.0174, 0.0200], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 01:28:56,603 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.30 vs. limit=2.0 +2023-02-09 01:28:58,203 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226799.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:29:11,132 INFO [optim.py:369] (2/4) 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] (2/4) Epoch 29, batch 500, loss[loss=0.1818, simple_loss=0.277, pruned_loss=0.0433, over 8658.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2813, pruned_loss=0.05668, over 1480664.78 frames. ], batch size: 39, lr: 2.61e-03, grad_scale: 16.0 +2023-02-09 01:29:48,217 INFO [train.py:901] (2/4) Epoch 29, batch 550, loss[loss=0.1579, simple_loss=0.2431, pruned_loss=0.03632, over 7658.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2819, pruned_loss=0.05708, over 1513196.46 frames. ], batch size: 19, lr: 2.61e-03, grad_scale: 16.0 +2023-02-09 01:29:51,498 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-09 01:30:21,915 INFO [optim.py:369] (2/4) 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 01:30:24,633 INFO [train.py:901] (2/4) Epoch 29, batch 600, loss[loss=0.1993, simple_loss=0.2778, pruned_loss=0.06037, over 8095.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2825, pruned_loss=0.0579, over 1532245.38 frames. ], batch size: 21, lr: 2.61e-03, grad_scale: 16.0 +2023-02-09 01:30:28,546 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.94 vs. limit=5.0 +2023-02-09 01:30:35,119 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9305, 1.5239, 2.9655, 1.6117, 2.3327, 3.1585, 3.2830, 2.7599], + device='cuda:2'), covar=tensor([0.1018, 0.1581, 0.0330, 0.1815, 0.0819, 0.0290, 0.0692, 0.0520], + device='cuda:2'), in_proj_covar=tensor([0.0310, 0.0327, 0.0295, 0.0325, 0.0327, 0.0278, 0.0444, 0.0309], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:2') +2023-02-09 01:30:38,693 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3980, 2.3602, 3.0018, 2.5400, 2.9241, 2.5719, 2.4015, 1.8250], + device='cuda:2'), covar=tensor([0.5700, 0.5274, 0.2400, 0.4214, 0.2771, 0.3210, 0.1859, 0.5897], + device='cuda:2'), in_proj_covar=tensor([0.0965, 0.1033, 0.0838, 0.0999, 0.1026, 0.0936, 0.0772, 0.0852], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 01:30:43,060 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-09 01:30:56,506 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=226966.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:30:59,759 INFO [train.py:901] (2/4) Epoch 29, batch 650, loss[loss=0.1843, simple_loss=0.2663, pruned_loss=0.05113, over 8488.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2823, pruned_loss=0.05795, over 1551756.29 frames. ], batch size: 28, lr: 2.61e-03, grad_scale: 16.0 +2023-02-09 01:31:13,843 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=226991.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:31:25,428 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=227007.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:31:32,763 INFO [optim.py:369] (2/4) 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 01:31:36,262 INFO [train.py:901] (2/4) Epoch 29, batch 700, loss[loss=0.1786, simple_loss=0.2657, pruned_loss=0.04575, over 7546.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2819, pruned_loss=0.05758, over 1563060.05 frames. ], batch size: 18, lr: 2.61e-03, grad_scale: 16.0 +2023-02-09 01:31:56,836 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8107, 2.0766, 1.6764, 2.8157, 1.3253, 1.5601, 2.0102, 2.1714], + device='cuda:2'), covar=tensor([0.0906, 0.0951, 0.1056, 0.0361, 0.1125, 0.1432, 0.0849, 0.0806], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0196, 0.0245, 0.0214, 0.0204, 0.0249, 0.0251, 0.0206], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 01:32:12,690 INFO [train.py:901] (2/4) Epoch 29, batch 750, loss[loss=0.2077, simple_loss=0.2839, pruned_loss=0.06577, over 8579.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2815, pruned_loss=0.05751, over 1571949.29 frames. ], batch size: 34, lr: 2.61e-03, grad_scale: 16.0 +2023-02-09 01:32:31,395 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-09 01:32:40,242 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-09 01:32:44,349 INFO [optim.py:369] (2/4) 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] (2/4) Epoch 29, batch 800, loss[loss=0.1869, simple_loss=0.2828, pruned_loss=0.04544, over 8619.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2814, pruned_loss=0.05688, over 1586508.96 frames. ], batch size: 31, lr: 2.61e-03, grad_scale: 16.0 +2023-02-09 01:33:24,591 INFO [train.py:901] (2/4) Epoch 29, batch 850, loss[loss=0.1505, simple_loss=0.2263, pruned_loss=0.03733, over 7701.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2828, pruned_loss=0.05787, over 1590846.07 frames. ], batch size: 18, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:33:57,033 INFO [optim.py:369] (2/4) 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 01:33:59,149 INFO [train.py:901] (2/4) Epoch 29, batch 900, loss[loss=0.1895, simple_loss=0.2783, pruned_loss=0.05035, over 8319.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2826, pruned_loss=0.0575, over 1599168.21 frames. ], batch size: 25, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:34:35,613 INFO [train.py:901] (2/4) Epoch 29, batch 950, loss[loss=0.2137, simple_loss=0.2781, pruned_loss=0.07467, over 7791.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2829, pruned_loss=0.05781, over 1602385.21 frames. ], batch size: 19, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:35:04,861 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp1.1 from training. Duration: 25.3818125 +2023-02-09 01:35:09,005 INFO [optim.py:369] (2/4) 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 01:35:11,178 INFO [train.py:901] (2/4) Epoch 29, batch 1000, loss[loss=0.2197, simple_loss=0.3027, pruned_loss=0.06834, over 8508.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2819, pruned_loss=0.05702, over 1606468.65 frames. ], batch size: 28, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:35:32,311 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=227351.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:35:39,828 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-09 01:35:47,252 INFO [train.py:901] (2/4) Epoch 29, batch 1050, loss[loss=0.2315, simple_loss=0.3262, pruned_loss=0.06842, over 8703.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2823, pruned_loss=0.05702, over 1612338.37 frames. ], batch size: 34, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:35:52,713 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-09 01:36:22,044 INFO [optim.py:369] (2/4) 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 01:36:23,012 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8052, 1.6663, 2.4827, 1.6115, 1.3177, 2.3884, 0.5918, 1.5397], + device='cuda:2'), covar=tensor([0.1396, 0.1248, 0.0291, 0.1032, 0.2321, 0.0330, 0.1781, 0.1204], + device='cuda:2'), in_proj_covar=tensor([0.0202, 0.0207, 0.0137, 0.0224, 0.0278, 0.0147, 0.0173, 0.0199], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 01:36:24,319 INFO [train.py:901] (2/4) Epoch 29, batch 1100, loss[loss=0.1867, simple_loss=0.2753, pruned_loss=0.04911, over 8493.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2828, pruned_loss=0.05719, over 1613946.40 frames. ], batch size: 26, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:36:30,382 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=227429.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:36:56,358 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=227466.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:36:59,075 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8551, 1.4661, 1.6923, 1.3372, 0.9679, 1.4605, 1.5948, 1.6368], + device='cuda:2'), covar=tensor([0.0566, 0.1243, 0.1713, 0.1548, 0.0605, 0.1510, 0.0765, 0.0675], + device='cuda:2'), in_proj_covar=tensor([0.0100, 0.0154, 0.0191, 0.0163, 0.0102, 0.0164, 0.0114, 0.0147], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-09 01:36:59,552 INFO [train.py:901] (2/4) Epoch 29, batch 1150, loss[loss=0.2169, simple_loss=0.2959, pruned_loss=0.06896, over 7811.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2825, pruned_loss=0.05701, over 1614208.34 frames. ], batch size: 20, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:37:06,449 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-09 01:37:34,264 INFO [optim.py:369] (2/4) 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 01:37:36,419 INFO [train.py:901] (2/4) Epoch 29, batch 1200, loss[loss=0.1718, simple_loss=0.2635, pruned_loss=0.04007, over 8129.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2822, pruned_loss=0.05685, over 1611927.87 frames. ], batch size: 22, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:37:53,823 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9370, 2.0329, 1.7212, 2.4675, 1.1819, 1.5645, 1.7407, 2.0691], + device='cuda:2'), covar=tensor([0.0713, 0.0760, 0.0875, 0.0439, 0.1063, 0.1323, 0.0836, 0.0742], + device='cuda:2'), in_proj_covar=tensor([0.0230, 0.0195, 0.0244, 0.0213, 0.0203, 0.0247, 0.0250, 0.0205], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 01:38:09,052 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0038, 1.5544, 1.7780, 1.4315, 1.0973, 1.5504, 1.8409, 1.6201], + device='cuda:2'), covar=tensor([0.0549, 0.1289, 0.1674, 0.1467, 0.0591, 0.1492, 0.0706, 0.0661], + device='cuda:2'), in_proj_covar=tensor([0.0100, 0.0154, 0.0190, 0.0162, 0.0102, 0.0163, 0.0113, 0.0147], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-09 01:38:11,530 INFO [train.py:901] (2/4) Epoch 29, batch 1250, loss[loss=0.2271, simple_loss=0.3059, pruned_loss=0.07415, over 8452.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2814, pruned_loss=0.05625, over 1613721.59 frames. ], batch size: 27, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:38:13,776 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1345, 1.6755, 1.5109, 1.6715, 1.4666, 1.4176, 1.4295, 1.3710], + device='cuda:2'), covar=tensor([0.1216, 0.0549, 0.1322, 0.0547, 0.0762, 0.1573, 0.0902, 0.0868], + device='cuda:2'), in_proj_covar=tensor([0.0356, 0.0246, 0.0344, 0.0314, 0.0303, 0.0349, 0.0351, 0.0323], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 01:38:19,535 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5590, 2.0375, 3.1237, 1.4832, 2.3062, 2.0677, 1.6943, 2.3050], + device='cuda:2'), covar=tensor([0.1931, 0.2742, 0.0815, 0.4790, 0.2019, 0.3305, 0.2511, 0.2353], + device='cuda:2'), in_proj_covar=tensor([0.0539, 0.0639, 0.0564, 0.0672, 0.0667, 0.0617, 0.0565, 0.0648], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 01:38:46,455 INFO [optim.py:369] (2/4) 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 01:38:48,671 INFO [train.py:901] (2/4) Epoch 29, batch 1300, loss[loss=0.1994, simple_loss=0.2919, pruned_loss=0.05346, over 8490.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2805, pruned_loss=0.05593, over 1610031.25 frames. ], batch size: 28, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:39:24,554 INFO [train.py:901] (2/4) Epoch 29, batch 1350, loss[loss=0.2008, simple_loss=0.2847, pruned_loss=0.05844, over 8376.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.282, pruned_loss=0.05655, over 1614434.47 frames. ], batch size: 49, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:39:52,816 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=227711.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:39:58,210 INFO [optim.py:369] (2/4) 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 01:40:00,377 INFO [train.py:901] (2/4) Epoch 29, batch 1400, loss[loss=0.2189, simple_loss=0.3063, pruned_loss=0.06574, over 8537.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2818, pruned_loss=0.05635, over 1614935.14 frames. ], batch size: 31, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:40:01,308 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=227722.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:40:20,601 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=227747.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:40:38,114 INFO [train.py:901] (2/4) Epoch 29, batch 1450, loss[loss=0.2224, simple_loss=0.2987, pruned_loss=0.07307, over 8498.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2827, pruned_loss=0.05676, over 1617694.96 frames. ], batch size: 26, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:40:39,492 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=227773.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:40:48,019 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-09 01:41:11,911 INFO [optim.py:369] (2/4) 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 01:41:14,105 INFO [train.py:901] (2/4) Epoch 29, batch 1500, loss[loss=0.1942, simple_loss=0.2854, pruned_loss=0.05155, over 8232.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2821, pruned_loss=0.05681, over 1612947.05 frames. ], batch size: 24, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:41:32,538 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=227845.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:41:40,143 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2795, 3.1885, 2.9193, 1.6092, 2.8443, 2.9523, 2.9300, 2.8431], + device='cuda:2'), covar=tensor([0.1060, 0.0778, 0.1431, 0.4406, 0.1225, 0.1212, 0.1441, 0.1062], + device='cuda:2'), in_proj_covar=tensor([0.0545, 0.0458, 0.0456, 0.0563, 0.0446, 0.0470, 0.0442, 0.0414], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 01:41:51,485 INFO [train.py:901] (2/4) Epoch 29, batch 1550, loss[loss=0.176, simple_loss=0.2599, pruned_loss=0.04606, over 7791.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2816, pruned_loss=0.0572, over 1610958.32 frames. ], batch size: 19, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:42:04,415 INFO [zipformer.py:1185] (2/4) 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] (2/4) 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 01:42:27,699 INFO [train.py:901] (2/4) Epoch 29, batch 1600, loss[loss=0.1886, simple_loss=0.2759, pruned_loss=0.0507, over 7982.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2812, pruned_loss=0.05696, over 1612483.49 frames. ], batch size: 21, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:42:27,873 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2754, 2.6483, 1.9813, 3.9833, 1.5847, 1.8566, 2.5088, 2.6938], + device='cuda:2'), covar=tensor([0.0885, 0.0942, 0.1119, 0.0293, 0.1212, 0.1520, 0.0897, 0.0962], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0195, 0.0245, 0.0213, 0.0204, 0.0247, 0.0251, 0.0206], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 01:43:04,956 INFO [train.py:901] (2/4) Epoch 29, batch 1650, loss[loss=0.1866, simple_loss=0.2751, pruned_loss=0.04902, over 8024.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2824, pruned_loss=0.05721, over 1620933.42 frames. ], batch size: 22, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:43:39,888 INFO [optim.py:369] (2/4) 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] (2/4) Epoch 29, batch 1700, loss[loss=0.1767, simple_loss=0.2688, pruned_loss=0.04232, over 8517.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.2827, pruned_loss=0.05808, over 1617123.43 frames. ], batch size: 26, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:43:42,948 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8662, 3.8399, 3.4979, 1.9440, 3.4237, 3.5393, 3.4999, 3.3674], + device='cuda:2'), covar=tensor([0.0842, 0.0602, 0.1093, 0.4213, 0.0954, 0.1066, 0.1290, 0.0979], + device='cuda:2'), in_proj_covar=tensor([0.0548, 0.0461, 0.0457, 0.0567, 0.0447, 0.0472, 0.0445, 0.0415], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 01:43:44,493 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8079, 1.3989, 3.2845, 1.6207, 2.3208, 3.5427, 3.7358, 3.1118], + device='cuda:2'), covar=tensor([0.1273, 0.1929, 0.0339, 0.1990, 0.1069, 0.0268, 0.0551, 0.0514], + device='cuda:2'), in_proj_covar=tensor([0.0312, 0.0329, 0.0298, 0.0327, 0.0329, 0.0280, 0.0450, 0.0312], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:2') +2023-02-09 01:43:45,195 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228025.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:44:06,275 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=228055.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:44:18,096 INFO [train.py:901] (2/4) Epoch 29, batch 1750, loss[loss=0.1522, simple_loss=0.2264, pruned_loss=0.03898, over 7233.00 frames. ], tot_loss[loss=0.2002, simple_loss=0.2834, pruned_loss=0.05849, over 1618719.38 frames. ], batch size: 16, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:44:28,128 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2610, 1.9963, 2.4460, 2.1487, 2.4748, 2.3511, 2.2537, 1.4152], + device='cuda:2'), covar=tensor([0.6336, 0.5331, 0.2383, 0.3823, 0.2612, 0.3505, 0.2015, 0.5819], + device='cuda:2'), in_proj_covar=tensor([0.0974, 0.1038, 0.0842, 0.1005, 0.1030, 0.0942, 0.0776, 0.0857], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 01:44:38,632 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228098.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:44:38,668 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8041, 1.7655, 2.0789, 1.6708, 1.1069, 1.7268, 2.2710, 2.0803], + device='cuda:2'), covar=tensor([0.0459, 0.1244, 0.1580, 0.1373, 0.0578, 0.1409, 0.0599, 0.0606], + device='cuda:2'), in_proj_covar=tensor([0.0100, 0.0154, 0.0190, 0.0162, 0.0102, 0.0164, 0.0114, 0.0147], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0009, 0.0007, 0.0008], + device='cuda:2') +2023-02-09 01:44:52,744 INFO [optim.py:369] (2/4) 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 01:44:55,484 INFO [train.py:901] (2/4) Epoch 29, batch 1800, loss[loss=0.1931, simple_loss=0.2878, pruned_loss=0.04919, over 8502.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2827, pruned_loss=0.05811, over 1614939.02 frames. ], batch size: 28, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:45:11,245 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228144.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:45:12,123 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.83 vs. limit=2.0 +2023-02-09 01:45:28,612 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228169.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:45:29,270 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 29, batch 1850, loss[loss=0.1849, simple_loss=0.2694, pruned_loss=0.05017, over 8252.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2809, pruned_loss=0.05718, over 1614212.37 frames. ], batch size: 22, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:45:42,839 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=228189.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:46:03,761 INFO [optim.py:369] (2/4) 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 01:46:05,868 INFO [train.py:901] (2/4) Epoch 29, batch 1900, loss[loss=0.1925, simple_loss=0.2779, pruned_loss=0.05356, over 7652.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2812, pruned_loss=0.05716, over 1617874.51 frames. ], batch size: 19, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:46:29,136 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.37 vs. limit=5.0 +2023-02-09 01:46:32,893 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2023-02-09 01:46:38,539 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-09 01:46:41,416 INFO [train.py:901] (2/4) Epoch 29, batch 1950, loss[loss=0.18, simple_loss=0.2621, pruned_loss=0.04896, over 8314.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2794, pruned_loss=0.05661, over 1614395.53 frames. ], batch size: 25, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:46:50,986 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-09 01:47:05,322 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228304.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:47:10,140 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-09 01:47:16,244 INFO [optim.py:369] (2/4) 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 01:47:18,252 INFO [train.py:901] (2/4) Epoch 29, batch 2000, loss[loss=0.1874, simple_loss=0.2772, pruned_loss=0.0488, over 8103.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2808, pruned_loss=0.05708, over 1619098.31 frames. ], batch size: 23, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:47:36,284 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228347.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:47:39,106 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228351.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:47:44,567 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7555, 1.9827, 2.1034, 1.4228, 2.2758, 1.5751, 0.7704, 1.9871], + device='cuda:2'), covar=tensor([0.0736, 0.0445, 0.0337, 0.0694, 0.0440, 0.0938, 0.1065, 0.0405], + device='cuda:2'), in_proj_covar=tensor([0.0476, 0.0414, 0.0369, 0.0460, 0.0397, 0.0552, 0.0403, 0.0443], + device='cuda:2'), 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:2') +2023-02-09 01:47:52,203 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=228369.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:47:53,545 INFO [train.py:901] (2/4) Epoch 29, batch 2050, loss[loss=0.2009, simple_loss=0.2829, pruned_loss=0.0594, over 8541.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2818, pruned_loss=0.05739, over 1619454.75 frames. ], batch size: 39, lr: 2.61e-03, grad_scale: 8.0 +2023-02-09 01:48:25,962 INFO [optim.py:369] (2/4) 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 01:48:28,119 INFO [train.py:901] (2/4) Epoch 29, batch 2100, loss[loss=0.1971, simple_loss=0.2954, pruned_loss=0.04941, over 8472.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2821, pruned_loss=0.05703, over 1617634.97 frames. ], batch size: 25, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 01:48:32,346 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228426.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:48:44,591 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=228442.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:48:48,016 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228447.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 01:48:50,742 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228451.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:48:59,114 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5033, 1.9898, 3.0507, 1.4381, 2.3553, 1.9448, 1.6706, 2.4112], + device='cuda:2'), covar=tensor([0.2221, 0.2913, 0.1091, 0.5151, 0.2113, 0.3664, 0.2762, 0.2583], + device='cuda:2'), in_proj_covar=tensor([0.0543, 0.0642, 0.0568, 0.0675, 0.0667, 0.0617, 0.0566, 0.0650], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 01:49:04,307 INFO [train.py:901] (2/4) Epoch 29, batch 2150, loss[loss=0.1741, simple_loss=0.2557, pruned_loss=0.04629, over 7914.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2812, pruned_loss=0.05698, over 1615392.48 frames. ], batch size: 20, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 01:49:05,161 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1460, 1.3109, 4.3350, 1.5743, 3.8431, 3.6451, 3.9125, 3.7922], + device='cuda:2'), covar=tensor([0.0716, 0.5074, 0.0657, 0.4524, 0.1204, 0.1019, 0.0694, 0.0779], + device='cuda:2'), in_proj_covar=tensor([0.0686, 0.0669, 0.0744, 0.0664, 0.0750, 0.0639, 0.0647, 0.0721], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 01:49:14,150 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228484.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:49:37,698 INFO [optim.py:369] (2/4) 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 01:49:39,911 INFO [train.py:901] (2/4) Epoch 29, batch 2200, loss[loss=0.2313, simple_loss=0.3143, pruned_loss=0.07409, over 8503.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2813, pruned_loss=0.05751, over 1610286.55 frames. ], batch size: 28, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 01:49:47,022 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6773, 1.4042, 4.9122, 1.7918, 4.3140, 4.0788, 4.4351, 4.3103], + device='cuda:2'), covar=tensor([0.0622, 0.4698, 0.0474, 0.4199, 0.1027, 0.0901, 0.0512, 0.0632], + device='cuda:2'), in_proj_covar=tensor([0.0684, 0.0666, 0.0742, 0.0662, 0.0747, 0.0637, 0.0644, 0.0719], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 01:50:06,278 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228557.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:50:08,439 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228560.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:50:16,508 INFO [train.py:901] (2/4) Epoch 29, batch 2250, loss[loss=0.2016, simple_loss=0.2865, pruned_loss=0.05842, over 8354.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2811, pruned_loss=0.05729, over 1607303.59 frames. ], batch size: 49, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 01:50:26,599 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228585.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:50:44,246 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7843, 1.7199, 4.2000, 1.5637, 2.5835, 4.6449, 5.1171, 3.6350], + device='cuda:2'), covar=tensor([0.2007, 0.2433, 0.0420, 0.2882, 0.1436, 0.0360, 0.0459, 0.1020], + device='cuda:2'), in_proj_covar=tensor([0.0311, 0.0327, 0.0297, 0.0325, 0.0329, 0.0279, 0.0447, 0.0312], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:2') +2023-02-09 01:50:50,350 INFO [optim.py:369] (2/4) 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 01:50:52,578 INFO [train.py:901] (2/4) Epoch 29, batch 2300, loss[loss=0.174, simple_loss=0.2546, pruned_loss=0.04671, over 7242.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2814, pruned_loss=0.05713, over 1608227.19 frames. ], batch size: 16, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 01:50:53,463 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4665, 1.3946, 1.8049, 1.1952, 1.1198, 1.7806, 0.2927, 1.1643], + device='cuda:2'), covar=tensor([0.1581, 0.1183, 0.0380, 0.0825, 0.2317, 0.0506, 0.1801, 0.1107], + device='cuda:2'), in_proj_covar=tensor([0.0202, 0.0206, 0.0138, 0.0223, 0.0278, 0.0149, 0.0173, 0.0199], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 01:51:06,293 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.73 vs. limit=2.0 +2023-02-09 01:51:22,795 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8420, 1.5850, 4.0234, 1.4677, 3.5753, 3.3958, 3.6417, 3.5036], + device='cuda:2'), covar=tensor([0.0699, 0.4130, 0.0648, 0.4094, 0.1198, 0.0981, 0.0638, 0.0800], + device='cuda:2'), in_proj_covar=tensor([0.0686, 0.0669, 0.0746, 0.0666, 0.0751, 0.0640, 0.0647, 0.0722], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 01:51:28,772 INFO [train.py:901] (2/4) Epoch 29, batch 2350, loss[loss=0.2061, simple_loss=0.2984, pruned_loss=0.05694, over 8505.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2816, pruned_loss=0.0572, over 1607532.61 frames. ], batch size: 26, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 01:51:43,128 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=228691.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:51:45,885 INFO [zipformer.py:1185] (2/4) 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] (2/4) 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 01:52:04,276 INFO [train.py:901] (2/4) Epoch 29, batch 2400, loss[loss=0.1829, simple_loss=0.2663, pruned_loss=0.04975, over 8083.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2806, pruned_loss=0.05636, over 1610654.72 frames. ], batch size: 21, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 01:52:17,432 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228740.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:52:34,956 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228765.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:52:39,604 INFO [train.py:901] (2/4) Epoch 29, batch 2450, loss[loss=0.2019, simple_loss=0.295, pruned_loss=0.0544, over 8517.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2807, pruned_loss=0.05633, over 1615150.22 frames. ], batch size: 26, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 01:52:55,038 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=228791.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 01:53:05,521 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228806.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:53:08,255 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228810.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:53:10,297 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=228813.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:53:13,499 INFO [optim.py:369] (2/4) 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 01:53:16,189 INFO [train.py:901] (2/4) Epoch 29, batch 2500, loss[loss=0.2005, simple_loss=0.2961, pruned_loss=0.05244, over 8141.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2816, pruned_loss=0.0569, over 1617033.68 frames. ], batch size: 22, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 01:53:28,187 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=228838.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:53:49,890 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-09 01:53:51,003 INFO [train.py:901] (2/4) Epoch 29, batch 2550, loss[loss=0.1791, simple_loss=0.2568, pruned_loss=0.05066, over 7544.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2825, pruned_loss=0.05736, over 1618333.75 frames. ], batch size: 18, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 01:54:17,016 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=228906.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 01:54:25,696 INFO [optim.py:369] (2/4) 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 01:54:27,394 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0366, 2.3079, 3.1119, 1.9601, 2.6496, 2.3668, 2.1103, 2.6689], + device='cuda:2'), covar=tensor([0.1556, 0.2120, 0.0762, 0.3368, 0.1414, 0.2408, 0.1859, 0.1807], + device='cuda:2'), in_proj_covar=tensor([0.0546, 0.0644, 0.0569, 0.0679, 0.0670, 0.0619, 0.0568, 0.0652], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 01:54:27,832 INFO [train.py:901] (2/4) Epoch 29, batch 2600, loss[loss=0.2629, simple_loss=0.3266, pruned_loss=0.09961, over 6947.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2822, pruned_loss=0.05738, over 1612206.72 frames. ], batch size: 71, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 01:54:51,312 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=228953.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:55:03,867 INFO [train.py:901] (2/4) Epoch 29, batch 2650, loss[loss=0.1697, simple_loss=0.245, pruned_loss=0.04722, over 7433.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2813, pruned_loss=0.05683, over 1611252.85 frames. ], batch size: 17, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 01:55:27,203 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2578, 1.4688, 3.4199, 1.0185, 3.0039, 2.8614, 3.1219, 3.0279], + device='cuda:2'), covar=tensor([0.0767, 0.3820, 0.0780, 0.4308, 0.1308, 0.1051, 0.0697, 0.0841], + device='cuda:2'), in_proj_covar=tensor([0.0681, 0.0663, 0.0741, 0.0662, 0.0743, 0.0635, 0.0640, 0.0715], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 01:55:37,352 INFO [optim.py:369] (2/4) 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 01:55:39,858 INFO [train.py:901] (2/4) Epoch 29, batch 2700, loss[loss=0.2078, simple_loss=0.2886, pruned_loss=0.06347, over 8021.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2818, pruned_loss=0.05685, over 1615349.92 frames. ], batch size: 22, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 01:56:11,094 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229062.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:56:14,607 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229066.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:56:17,757 INFO [train.py:901] (2/4) Epoch 29, batch 2750, loss[loss=0.1787, simple_loss=0.2466, pruned_loss=0.05537, over 7535.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2812, pruned_loss=0.05663, over 1611531.66 frames. ], batch size: 18, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 01:56:29,426 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229087.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:56:32,344 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229091.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:56:51,298 INFO [optim.py:369] (2/4) 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] (2/4) Epoch 29, batch 2800, loss[loss=0.212, simple_loss=0.301, pruned_loss=0.06155, over 8493.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2818, pruned_loss=0.05716, over 1615015.76 frames. ], batch size: 28, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 01:56:54,325 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0903, 1.5717, 3.5218, 1.6208, 2.4717, 3.9158, 3.9510, 3.3583], + device='cuda:2'), covar=tensor([0.1168, 0.1876, 0.0340, 0.2139, 0.1065, 0.0224, 0.0603, 0.0538], + device='cuda:2'), in_proj_covar=tensor([0.0313, 0.0329, 0.0298, 0.0326, 0.0331, 0.0281, 0.0450, 0.0313], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 01:57:24,941 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229162.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 01:57:31,020 INFO [train.py:901] (2/4) Epoch 29, batch 2850, loss[loss=0.1882, simple_loss=0.2636, pruned_loss=0.05645, over 7453.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2817, pruned_loss=0.05717, over 1613845.88 frames. ], batch size: 17, lr: 2.60e-03, grad_scale: 16.0 +2023-02-09 01:57:43,221 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229187.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 01:58:03,980 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5060, 2.1772, 3.4283, 2.1957, 2.9546, 3.8857, 3.8732, 3.5317], + device='cuda:2'), covar=tensor([0.0962, 0.1567, 0.0583, 0.1753, 0.1611, 0.0214, 0.0718, 0.0441], + device='cuda:2'), in_proj_covar=tensor([0.0314, 0.0330, 0.0299, 0.0327, 0.0331, 0.0282, 0.0451, 0.0314], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 01:58:05,219 INFO [optim.py:369] (2/4) 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 01:58:07,399 INFO [train.py:901] (2/4) Epoch 29, batch 2900, loss[loss=0.1735, simple_loss=0.2581, pruned_loss=0.04445, over 7460.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2807, pruned_loss=0.05658, over 1611215.76 frames. ], batch size: 17, lr: 2.60e-03, grad_scale: 16.0 +2023-02-09 01:58:21,866 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-09 01:58:23,777 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6554, 1.4358, 1.5891, 1.4013, 0.9586, 1.4256, 1.5175, 1.5665], + device='cuda:2'), covar=tensor([0.0632, 0.1275, 0.1735, 0.1454, 0.0625, 0.1514, 0.0769, 0.0641], + device='cuda:2'), in_proj_covar=tensor([0.0101, 0.0155, 0.0191, 0.0163, 0.0103, 0.0165, 0.0114, 0.0148], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0007, 0.0008], + device='cuda:2') +2023-02-09 01:58:44,220 INFO [train.py:901] (2/4) Epoch 29, batch 2950, loss[loss=0.2007, simple_loss=0.2921, pruned_loss=0.05462, over 8557.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2807, pruned_loss=0.05667, over 1612630.45 frames. ], batch size: 31, lr: 2.60e-03, grad_scale: 16.0 +2023-02-09 01:58:47,089 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-09 01:59:02,249 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=229297.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:59:17,278 INFO [optim.py:369] (2/4) 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 01:59:17,404 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229318.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 01:59:19,291 INFO [train.py:901] (2/4) Epoch 29, batch 3000, loss[loss=0.1752, simple_loss=0.2702, pruned_loss=0.0401, over 8329.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.281, pruned_loss=0.05676, over 1610692.22 frames. ], batch size: 25, lr: 2.60e-03, grad_scale: 16.0 +2023-02-09 01:59:19,291 INFO [train.py:926] (2/4) Computing validation loss +2023-02-09 01:59:34,612 INFO [train.py:935] (2/4) Epoch 29, validation: loss=0.17, simple_loss=0.2699, pruned_loss=0.03504, over 944034.00 frames. +2023-02-09 01:59:34,613 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6724MB +2023-02-09 02:00:09,410 INFO [train.py:901] (2/4) Epoch 29, batch 3050, loss[loss=0.1545, simple_loss=0.2331, pruned_loss=0.0379, over 7429.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2806, pruned_loss=0.05705, over 1609378.59 frames. ], batch size: 17, lr: 2.60e-03, grad_scale: 16.0 +2023-02-09 02:00:28,259 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.6088, 4.6313, 4.1370, 2.1592, 4.0827, 4.1978, 4.0871, 4.1293], + device='cuda:2'), covar=tensor([0.0655, 0.0442, 0.0865, 0.4539, 0.0906, 0.1007, 0.1255, 0.0730], + device='cuda:2'), in_proj_covar=tensor([0.0551, 0.0465, 0.0458, 0.0567, 0.0446, 0.0471, 0.0446, 0.0416], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 02:00:40,616 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=229412.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:00:41,980 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7599, 1.8162, 1.6789, 2.3371, 1.0521, 1.5613, 1.8260, 1.8770], + device='cuda:2'), covar=tensor([0.0814, 0.0799, 0.0889, 0.0409, 0.1063, 0.1298, 0.0700, 0.0783], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0196, 0.0244, 0.0213, 0.0203, 0.0247, 0.0250, 0.0205], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 02:00:44,471 INFO [optim.py:369] (2/4) 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 02:00:46,515 INFO [train.py:901] (2/4) Epoch 29, batch 3100, loss[loss=0.1887, simple_loss=0.2788, pruned_loss=0.04932, over 8250.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2801, pruned_loss=0.05705, over 1607523.40 frames. ], batch size: 24, lr: 2.60e-03, grad_scale: 16.0 +2023-02-09 02:00:59,213 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8632, 2.1923, 3.6776, 1.7957, 1.8932, 3.6480, 0.6665, 2.1527], + device='cuda:2'), covar=tensor([0.1105, 0.1201, 0.0260, 0.1619, 0.2052, 0.0296, 0.2095, 0.1357], + device='cuda:2'), in_proj_covar=tensor([0.0203, 0.0208, 0.0138, 0.0224, 0.0280, 0.0149, 0.0173, 0.0201], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 02:01:15,192 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229461.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:01:21,900 INFO [train.py:901] (2/4) Epoch 29, batch 3150, loss[loss=0.1535, simple_loss=0.2372, pruned_loss=0.03491, over 7908.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2807, pruned_loss=0.05714, over 1610512.33 frames. ], batch size: 20, lr: 2.60e-03, grad_scale: 16.0 +2023-02-09 02:01:39,329 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229495.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:01:56,471 INFO [optim.py:369] (2/4) 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] (2/4) Epoch 29, batch 3200, loss[loss=0.1995, simple_loss=0.2961, pruned_loss=0.05146, over 8199.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2811, pruned_loss=0.05723, over 1607694.48 frames. ], batch size: 23, lr: 2.60e-03, grad_scale: 16.0 +2023-02-09 02:01:58,993 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.49 vs. limit=2.0 +2023-02-09 02:02:02,523 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.71 vs. limit=2.0 +2023-02-09 02:02:14,562 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.15 vs. limit=5.0 +2023-02-09 02:02:29,972 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7233, 2.5364, 1.8697, 2.3021, 2.2871, 1.5972, 2.2227, 2.1471], + device='cuda:2'), covar=tensor([0.1295, 0.0403, 0.1152, 0.0601, 0.0668, 0.1511, 0.0844, 0.0885], + device='cuda:2'), in_proj_covar=tensor([0.0353, 0.0243, 0.0342, 0.0314, 0.0300, 0.0349, 0.0349, 0.0319], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 02:02:35,461 INFO [train.py:901] (2/4) Epoch 29, batch 3250, loss[loss=0.1814, simple_loss=0.2663, pruned_loss=0.04827, over 7809.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2823, pruned_loss=0.05779, over 1607764.70 frames. ], batch size: 20, lr: 2.60e-03, grad_scale: 16.0 +2023-02-09 02:02:52,624 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229595.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:03:09,575 INFO [optim.py:369] (2/4) 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 02:03:11,692 INFO [train.py:901] (2/4) Epoch 29, batch 3300, loss[loss=0.2623, simple_loss=0.3411, pruned_loss=0.09172, over 8451.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2817, pruned_loss=0.05664, over 1614138.71 frames. ], batch size: 27, lr: 2.60e-03, grad_scale: 16.0 +2023-02-09 02:03:41,137 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=229662.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:03:45,538 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=229668.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:03:47,461 INFO [train.py:901] (2/4) Epoch 29, batch 3350, loss[loss=0.1681, simple_loss=0.248, pruned_loss=0.04412, over 7800.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2811, pruned_loss=0.05666, over 1614147.86 frames. ], batch size: 19, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 02:03:52,168 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229677.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 02:04:03,324 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=229693.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:04:20,884 INFO [optim.py:369] (2/4) 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 02:04:22,316 INFO [train.py:901] (2/4) Epoch 29, batch 3400, loss[loss=0.2086, simple_loss=0.2951, pruned_loss=0.06101, over 8195.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2812, pruned_loss=0.05697, over 1609990.78 frames. ], batch size: 23, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 02:04:25,972 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1294, 2.1892, 1.7322, 2.8945, 1.3909, 1.6653, 2.0602, 2.2079], + device='cuda:2'), covar=tensor([0.0646, 0.0737, 0.0939, 0.0323, 0.1084, 0.1281, 0.0856, 0.0815], + device='cuda:2'), in_proj_covar=tensor([0.0230, 0.0195, 0.0245, 0.0213, 0.0203, 0.0246, 0.0251, 0.0205], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 02:04:59,068 INFO [train.py:901] (2/4) Epoch 29, batch 3450, loss[loss=0.1692, simple_loss=0.2531, pruned_loss=0.04262, over 7544.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2801, pruned_loss=0.05649, over 1608864.43 frames. ], batch size: 18, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 02:05:03,516 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=229777.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:05:11,078 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5172, 1.8669, 1.8734, 1.2497, 1.9227, 1.4981, 0.4601, 1.8191], + device='cuda:2'), covar=tensor([0.0652, 0.0376, 0.0376, 0.0627, 0.0478, 0.1071, 0.1100, 0.0326], + device='cuda:2'), in_proj_covar=tensor([0.0484, 0.0416, 0.0375, 0.0468, 0.0403, 0.0559, 0.0411, 0.0450], + device='cuda:2'), out_proj_covar=tensor([1.2816e-04, 1.0766e-04, 9.7610e-05, 1.2232e-04, 1.0553e-04, 1.5573e-04, + 1.0969e-04, 1.1767e-04], device='cuda:2') +2023-02-09 02:05:23,715 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=229805.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:05:33,258 INFO [optim.py:369] (2/4) 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 02:05:34,611 INFO [train.py:901] (2/4) Epoch 29, batch 3500, loss[loss=0.2224, simple_loss=0.3051, pruned_loss=0.0698, over 8556.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2819, pruned_loss=0.0569, over 1616614.93 frames. ], batch size: 39, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 02:05:46,947 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=229839.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:05:49,139 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.2701, 3.1797, 2.9638, 1.6149, 2.9037, 2.9811, 2.8194, 2.8706], + device='cuda:2'), covar=tensor([0.1117, 0.0825, 0.1325, 0.4602, 0.1149, 0.1130, 0.1738, 0.1002], + device='cuda:2'), in_proj_covar=tensor([0.0547, 0.0462, 0.0455, 0.0567, 0.0446, 0.0470, 0.0445, 0.0413], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 02:05:58,499 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-09 02:06:12,025 INFO [train.py:901] (2/4) Epoch 29, batch 3550, loss[loss=0.2044, simple_loss=0.2885, pruned_loss=0.06015, over 8089.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2818, pruned_loss=0.05724, over 1613159.59 frames. ], batch size: 21, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 02:06:15,083 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4334, 1.4293, 1.3917, 1.8533, 0.8119, 1.2738, 1.4452, 1.4776], + device='cuda:2'), covar=tensor([0.0867, 0.0774, 0.0984, 0.0420, 0.1058, 0.1402, 0.0659, 0.0780], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0196, 0.0244, 0.0213, 0.0203, 0.0247, 0.0251, 0.0205], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 02:06:41,728 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=229912.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:06:46,512 INFO [optim.py:369] (2/4) 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 02:06:47,430 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=229920.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:06:47,947 INFO [train.py:901] (2/4) Epoch 29, batch 3600, loss[loss=0.16, simple_loss=0.2451, pruned_loss=0.03745, over 7805.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2819, pruned_loss=0.05726, over 1616147.14 frames. ], batch size: 20, lr: 2.60e-03, grad_scale: 8.0 +2023-02-09 02:07:00,995 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=229939.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:07:01,976 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-09 02:07:05,361 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9767, 2.0083, 1.7726, 2.6390, 1.2899, 1.7042, 2.0107, 2.0755], + device='cuda:2'), covar=tensor([0.0691, 0.0751, 0.0817, 0.0363, 0.1061, 0.1213, 0.0707, 0.0773], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0196, 0.0244, 0.0213, 0.0203, 0.0246, 0.0251, 0.0204], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 02:07:11,758 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 29, batch 3650, loss[loss=0.2126, simple_loss=0.2869, pruned_loss=0.06911, over 8846.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2797, pruned_loss=0.05632, over 1612782.54 frames. ], batch size: 35, lr: 2.60e-03, grad_scale: 4.0 +2023-02-09 02:07:37,192 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-09 02:07:38,963 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1825, 2.2792, 1.9212, 2.9871, 1.3796, 1.6995, 2.1718, 2.3070], + device='cuda:2'), covar=tensor([0.0692, 0.0791, 0.0875, 0.0334, 0.1147, 0.1363, 0.0867, 0.0830], + device='cuda:2'), in_proj_covar=tensor([0.0230, 0.0195, 0.0243, 0.0212, 0.0202, 0.0245, 0.0250, 0.0204], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 02:07:50,393 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230005.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:07:56,887 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-09 02:08:00,955 INFO [optim.py:369] (2/4) 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 02:08:01,702 INFO [train.py:901] (2/4) Epoch 29, batch 3700, loss[loss=0.1747, simple_loss=0.2618, pruned_loss=0.04384, over 8095.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.2798, pruned_loss=0.05589, over 1611766.61 frames. ], batch size: 21, lr: 2.60e-03, grad_scale: 4.0 +2023-02-09 02:08:02,483 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=230021.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 02:08:06,530 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-09 02:08:07,555 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.21 vs. limit=2.0 +2023-02-09 02:08:10,872 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230033.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:08:25,156 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230054.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:08:28,463 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230058.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:08:37,283 INFO [train.py:901] (2/4) Epoch 29, batch 3750, loss[loss=0.1972, simple_loss=0.2928, pruned_loss=0.05085, over 8454.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2816, pruned_loss=0.05666, over 1615967.73 frames. ], batch size: 27, lr: 2.60e-03, grad_scale: 4.0 +2023-02-09 02:09:13,042 INFO [optim.py:369] (2/4) 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 02:09:13,774 INFO [train.py:901] (2/4) Epoch 29, batch 3800, loss[loss=0.1633, simple_loss=0.235, pruned_loss=0.04575, over 7710.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2821, pruned_loss=0.0571, over 1613398.87 frames. ], batch size: 18, lr: 2.60e-03, grad_scale: 4.0 +2023-02-09 02:09:24,880 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230136.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 02:09:47,382 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230168.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:09:49,388 INFO [train.py:901] (2/4) Epoch 29, batch 3850, loss[loss=0.2093, simple_loss=0.2901, pruned_loss=0.0643, over 8134.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2828, pruned_loss=0.05747, over 1612507.23 frames. ], batch size: 22, lr: 2.60e-03, grad_scale: 4.0 +2023-02-09 02:09:53,095 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230176.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:09:56,411 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7925, 1.9852, 2.0710, 1.4215, 2.2147, 1.5555, 0.6898, 1.9832], + device='cuda:2'), covar=tensor([0.0786, 0.0403, 0.0386, 0.0722, 0.0601, 0.0993, 0.1061, 0.0355], + device='cuda:2'), in_proj_covar=tensor([0.0475, 0.0412, 0.0369, 0.0462, 0.0397, 0.0552, 0.0404, 0.0443], + device='cuda:2'), out_proj_covar=tensor([1.2572e-04, 1.0659e-04, 9.6022e-05, 1.2081e-04, 1.0388e-04, 1.5382e-04, + 1.0775e-04, 1.1608e-04], device='cuda:2') +2023-02-09 02:10:10,986 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230201.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:10:12,900 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-09 02:10:17,560 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230210.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:10:24,395 INFO [optim.py:369] (2/4) 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 02:10:25,130 INFO [train.py:901] (2/4) Epoch 29, batch 3900, loss[loss=0.2242, simple_loss=0.2915, pruned_loss=0.07844, over 8234.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2822, pruned_loss=0.05725, over 1610402.19 frames. ], batch size: 22, lr: 2.59e-03, grad_scale: 4.0 +2023-02-09 02:10:36,241 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230235.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:10:51,859 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=230256.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:10:54,117 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230259.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:11:02,669 INFO [train.py:901] (2/4) Epoch 29, batch 3950, loss[loss=0.2095, simple_loss=0.2983, pruned_loss=0.06035, over 8363.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2827, pruned_loss=0.05751, over 1607942.61 frames. ], batch size: 24, lr: 2.59e-03, grad_scale: 4.0 +2023-02-09 02:11:30,499 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230310.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:11:37,958 INFO [optim.py:369] (2/4) 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 02:11:38,719 INFO [train.py:901] (2/4) Epoch 29, batch 4000, loss[loss=0.2262, simple_loss=0.3, pruned_loss=0.07617, over 8498.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2819, pruned_loss=0.05749, over 1608923.09 frames. ], batch size: 28, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:11:50,444 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230335.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:11:59,727 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=230349.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:12:16,073 INFO [train.py:901] (2/4) Epoch 29, batch 4050, loss[loss=0.2071, simple_loss=0.2964, pruned_loss=0.05888, over 8469.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2827, pruned_loss=0.05753, over 1614135.22 frames. ], batch size: 27, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:12:16,258 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230371.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:12:31,110 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230392.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 02:12:48,565 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230417.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 02:12:50,324 INFO [optim.py:369] (2/4) 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 02:12:51,013 INFO [train.py:901] (2/4) Epoch 29, batch 4100, loss[loss=0.2351, simple_loss=0.3187, pruned_loss=0.07579, over 8454.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2813, pruned_loss=0.05691, over 1615563.26 frames. ], batch size: 29, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:13:22,441 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230464.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:13:26,984 INFO [train.py:901] (2/4) Epoch 29, batch 4150, loss[loss=0.1863, simple_loss=0.2729, pruned_loss=0.04991, over 8084.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2811, pruned_loss=0.05705, over 1615338.83 frames. ], batch size: 21, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:13:38,478 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230486.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:13:56,267 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=230512.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:14:01,810 INFO [optim.py:369] (2/4) 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 02:14:02,515 INFO [train.py:901] (2/4) Epoch 29, batch 4200, loss[loss=0.1732, simple_loss=0.2638, pruned_loss=0.04129, over 8143.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2824, pruned_loss=0.05724, over 1619780.83 frames. ], batch size: 22, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:14:16,944 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-09 02:14:17,779 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230543.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:14:24,143 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230551.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:14:38,389 INFO [train.py:901] (2/4) Epoch 29, batch 4250, loss[loss=0.1951, simple_loss=0.284, pruned_loss=0.05313, over 8363.00 frames. ], tot_loss[loss=0.1997, simple_loss=0.2834, pruned_loss=0.05796, over 1620966.62 frames. ], batch size: 24, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:14:41,195 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-09 02:15:01,587 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=230603.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:15:13,975 INFO [optim.py:369] (2/4) 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 02:15:14,735 INFO [train.py:901] (2/4) Epoch 29, batch 4300, loss[loss=0.1907, simple_loss=0.2746, pruned_loss=0.0534, over 8029.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.282, pruned_loss=0.05755, over 1615732.23 frames. ], batch size: 22, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:15:19,485 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230627.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:15:19,526 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230627.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:15:37,545 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230652.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:15:51,725 INFO [train.py:901] (2/4) Epoch 29, batch 4350, loss[loss=0.1846, simple_loss=0.2817, pruned_loss=0.04376, over 8651.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2811, pruned_loss=0.05679, over 1616790.64 frames. ], batch size: 34, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:15:55,703 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2337, 1.1303, 1.3362, 1.0333, 1.0145, 1.3440, 0.1069, 0.9429], + device='cuda:2'), covar=tensor([0.1336, 0.1116, 0.0460, 0.0620, 0.2241, 0.0482, 0.1782, 0.1062], + device='cuda:2'), in_proj_covar=tensor([0.0203, 0.0206, 0.0138, 0.0225, 0.0281, 0.0149, 0.0174, 0.0202], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 02:16:17,531 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-09 02:16:27,826 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230718.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:16:29,089 INFO [optim.py:369] (2/4) 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 02:16:29,341 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230720.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:16:29,846 INFO [train.py:901] (2/4) Epoch 29, batch 4400, loss[loss=0.1982, simple_loss=0.2747, pruned_loss=0.06081, over 8086.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2812, pruned_loss=0.05652, over 1615623.72 frames. ], batch size: 21, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:16:47,627 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230745.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:16:57,339 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-09 02:17:06,602 INFO [train.py:901] (2/4) Epoch 29, batch 4450, loss[loss=0.1789, simple_loss=0.2705, pruned_loss=0.04359, over 8105.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2822, pruned_loss=0.05726, over 1617904.54 frames. ], batch size: 23, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:17:36,258 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230810.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:17:43,245 INFO [optim.py:369] (2/4) 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 02:17:43,984 INFO [train.py:901] (2/4) Epoch 29, batch 4500, loss[loss=0.2147, simple_loss=0.2902, pruned_loss=0.06958, over 8455.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2822, pruned_loss=0.05735, over 1620653.58 frames. ], batch size: 27, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:17:49,307 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1027, 1.5578, 4.2744, 2.0914, 2.3984, 4.8225, 4.9827, 4.2242], + device='cuda:2'), covar=tensor([0.1325, 0.2030, 0.0298, 0.1899, 0.1360, 0.0186, 0.0529, 0.0564], + device='cuda:2'), in_proj_covar=tensor([0.0311, 0.0328, 0.0296, 0.0325, 0.0328, 0.0279, 0.0447, 0.0311], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:2') +2023-02-09 02:17:51,400 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=230830.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:17:54,896 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-09 02:18:19,714 INFO [train.py:901] (2/4) Epoch 29, batch 4550, loss[loss=0.2074, simple_loss=0.2921, pruned_loss=0.06134, over 7801.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2821, pruned_loss=0.05707, over 1620736.27 frames. ], batch size: 20, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:18:28,416 INFO [zipformer.py:1185] (2/4) 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] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=230887.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:18:37,202 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=230895.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:18:46,426 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=230908.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:18:46,491 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230908.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:18:55,062 INFO [optim.py:369] (2/4) 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 02:18:55,766 INFO [train.py:901] (2/4) Epoch 29, batch 4600, loss[loss=0.1938, simple_loss=0.2566, pruned_loss=0.06548, over 6777.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2815, pruned_loss=0.05658, over 1619908.51 frames. ], batch size: 15, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:19:13,589 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=230945.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:19:31,772 INFO [train.py:901] (2/4) Epoch 29, batch 4650, loss[loss=0.1595, simple_loss=0.2549, pruned_loss=0.03203, over 7656.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2818, pruned_loss=0.05642, over 1623499.48 frames. ], batch size: 19, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:19:33,908 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=230974.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:19:42,741 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.5072, 4.4387, 4.0427, 1.8592, 3.9853, 4.0409, 3.9886, 3.9466], + device='cuda:2'), covar=tensor([0.0649, 0.0502, 0.1012, 0.4447, 0.0833, 0.0926, 0.1276, 0.0886], + device='cuda:2'), in_proj_covar=tensor([0.0544, 0.0460, 0.0451, 0.0563, 0.0442, 0.0466, 0.0441, 0.0411], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 02:19:51,178 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=230999.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:19:53,283 INFO [zipformer.py:1185] (2/4) 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] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231010.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:20:06,387 INFO [optim.py:369] (2/4) 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 02:20:07,122 INFO [train.py:901] (2/4) Epoch 29, batch 4700, loss[loss=0.1582, simple_loss=0.2412, pruned_loss=0.0376, over 7922.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2805, pruned_loss=0.05597, over 1620409.55 frames. ], batch size: 20, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:20:43,870 INFO [train.py:901] (2/4) Epoch 29, batch 4750, loss[loss=0.1847, simple_loss=0.2696, pruned_loss=0.04985, over 7795.00 frames. ], tot_loss[loss=0.196, simple_loss=0.2804, pruned_loss=0.05579, over 1618895.68 frames. ], batch size: 20, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:21:00,617 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-09 02:21:02,795 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-09 02:21:09,909 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.47 vs. limit=2.0 +2023-02-09 02:21:18,720 INFO [optim.py:369] (2/4) 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 02:21:19,433 INFO [train.py:901] (2/4) Epoch 29, batch 4800, loss[loss=0.1582, simple_loss=0.2404, pruned_loss=0.03797, over 7699.00 frames. ], tot_loss[loss=0.1954, simple_loss=0.2794, pruned_loss=0.05572, over 1616074.65 frames. ], batch size: 18, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:21:43,700 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=231154.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:21:55,936 INFO [train.py:901] (2/4) Epoch 29, batch 4850, loss[loss=0.2196, simple_loss=0.2987, pruned_loss=0.0702, over 8731.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.2794, pruned_loss=0.05584, over 1616427.05 frames. ], batch size: 30, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:21:55,943 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-09 02:22:17,913 INFO [zipformer.py:1185] (2/4) 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] (2/4) 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 02:22:31,778 INFO [train.py:901] (2/4) Epoch 29, batch 4900, loss[loss=0.1609, simple_loss=0.2438, pruned_loss=0.039, over 7212.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2805, pruned_loss=0.05608, over 1617391.24 frames. ], batch size: 16, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:22:35,618 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231226.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:22:54,381 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=231252.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:22:59,244 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231258.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:23:04,608 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231266.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:23:06,602 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=231269.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:23:06,638 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231269.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:23:07,928 INFO [train.py:901] (2/4) Epoch 29, batch 4950, loss[loss=0.2075, simple_loss=0.2868, pruned_loss=0.06416, over 7986.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2819, pruned_loss=0.05696, over 1617134.11 frames. ], batch size: 21, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:23:17,261 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231283.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:23:22,755 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231291.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:23:43,160 INFO [optim.py:369] (2/4) 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 02:23:43,853 INFO [train.py:901] (2/4) Epoch 29, batch 5000, loss[loss=0.2362, simple_loss=0.3158, pruned_loss=0.07831, over 8033.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.282, pruned_loss=0.05719, over 1619604.86 frames. ], batch size: 22, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:23:45,410 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4017, 1.8914, 3.4697, 1.5671, 2.4228, 3.8538, 3.9678, 3.2851], + device='cuda:2'), covar=tensor([0.0926, 0.1527, 0.0294, 0.2212, 0.1076, 0.0244, 0.0737, 0.0538], + device='cuda:2'), in_proj_covar=tensor([0.0313, 0.0330, 0.0299, 0.0327, 0.0328, 0.0281, 0.0450, 0.0312], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:2') +2023-02-09 02:24:17,055 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231367.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:24:19,694 INFO [train.py:901] (2/4) Epoch 29, batch 5050, loss[loss=0.1874, simple_loss=0.2776, pruned_loss=0.04857, over 8234.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2822, pruned_loss=0.05693, over 1622192.61 frames. ], batch size: 22, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:24:37,785 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-09 02:24:55,716 INFO [optim.py:369] (2/4) 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] (2/4) Epoch 29, batch 5100, loss[loss=0.2078, simple_loss=0.292, pruned_loss=0.06183, over 8138.00 frames. ], tot_loss[loss=0.1991, simple_loss=0.2835, pruned_loss=0.0574, over 1624916.44 frames. ], batch size: 22, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:25:32,299 INFO [train.py:901] (2/4) Epoch 29, batch 5150, loss[loss=0.1726, simple_loss=0.2651, pruned_loss=0.04008, over 7908.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.283, pruned_loss=0.05716, over 1625516.59 frames. ], batch size: 20, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:25:52,007 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.24 vs. limit=2.0 +2023-02-09 02:26:07,336 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-09 02:26:08,162 INFO [optim.py:369] (2/4) 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 02:26:08,942 INFO [train.py:901] (2/4) Epoch 29, batch 5200, loss[loss=0.1763, simple_loss=0.284, pruned_loss=0.03427, over 8338.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2826, pruned_loss=0.05704, over 1620977.43 frames. ], batch size: 26, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:26:12,099 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231525.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:26:30,207 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231550.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:26:39,121 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-09 02:26:44,754 INFO [train.py:901] (2/4) Epoch 29, batch 5250, loss[loss=0.2279, simple_loss=0.3115, pruned_loss=0.07215, over 7814.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2824, pruned_loss=0.05741, over 1619571.75 frames. ], batch size: 20, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:27:15,803 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=231613.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:27:20,495 INFO [optim.py:369] (2/4) 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 02:27:21,220 INFO [train.py:901] (2/4) Epoch 29, batch 5300, loss[loss=0.1854, simple_loss=0.2718, pruned_loss=0.04947, over 8462.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2828, pruned_loss=0.05749, over 1616732.73 frames. ], batch size: 25, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:27:22,860 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231623.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:27:41,043 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=231648.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:27:46,394 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.9296, 1.7459, 6.0953, 2.2886, 5.5153, 5.1635, 5.5991, 5.5385], + device='cuda:2'), covar=tensor([0.0477, 0.4894, 0.0409, 0.4084, 0.1005, 0.0888, 0.0501, 0.0522], + device='cuda:2'), in_proj_covar=tensor([0.0682, 0.0660, 0.0742, 0.0653, 0.0739, 0.0632, 0.0640, 0.0717], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 02:27:47,749 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4491, 1.4540, 1.4330, 1.8353, 0.6556, 1.3218, 1.3416, 1.4954], + device='cuda:2'), covar=tensor([0.0844, 0.0773, 0.0954, 0.0493, 0.1115, 0.1342, 0.0737, 0.0670], + device='cuda:2'), in_proj_covar=tensor([0.0230, 0.0195, 0.0243, 0.0213, 0.0202, 0.0245, 0.0250, 0.0204], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 02:27:53,027 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=231665.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:27:54,364 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1788, 1.4769, 4.3587, 1.5571, 3.8654, 3.6373, 3.9241, 3.8401], + device='cuda:2'), covar=tensor([0.0629, 0.4776, 0.0553, 0.4366, 0.1125, 0.0967, 0.0612, 0.0693], + device='cuda:2'), in_proj_covar=tensor([0.0682, 0.0660, 0.0742, 0.0653, 0.0739, 0.0632, 0.0640, 0.0717], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 02:27:57,015 INFO [train.py:901] (2/4) Epoch 29, batch 5350, loss[loss=0.2331, simple_loss=0.308, pruned_loss=0.0791, over 7692.00 frames. ], tot_loss[loss=0.2001, simple_loss=0.2834, pruned_loss=0.05836, over 1611057.82 frames. ], batch size: 73, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:28:32,684 INFO [optim.py:369] (2/4) 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 02:28:33,430 INFO [train.py:901] (2/4) Epoch 29, batch 5400, loss[loss=0.2226, simple_loss=0.2981, pruned_loss=0.0735, over 7640.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.2829, pruned_loss=0.05799, over 1611617.89 frames. ], batch size: 19, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:28:38,406 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=231728.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:29:09,676 INFO [train.py:901] (2/4) Epoch 29, batch 5450, loss[loss=0.1694, simple_loss=0.2614, pruned_loss=0.03865, over 7439.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2819, pruned_loss=0.05734, over 1609752.62 frames. ], batch size: 17, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:29:24,004 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3272, 3.6999, 2.7813, 3.2328, 3.0190, 2.4141, 3.0952, 3.2942], + device='cuda:2'), covar=tensor([0.1594, 0.0471, 0.0954, 0.0659, 0.0722, 0.1386, 0.0984, 0.1032], + device='cuda:2'), in_proj_covar=tensor([0.0356, 0.0244, 0.0343, 0.0315, 0.0302, 0.0348, 0.0351, 0.0320], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 02:29:29,518 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-09 02:29:43,734 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7725, 1.9857, 2.0466, 1.4267, 2.1446, 1.5182, 0.7665, 1.9677], + device='cuda:2'), covar=tensor([0.0614, 0.0406, 0.0325, 0.0670, 0.0407, 0.0937, 0.1032, 0.0326], + device='cuda:2'), in_proj_covar=tensor([0.0476, 0.0417, 0.0371, 0.0465, 0.0400, 0.0556, 0.0408, 0.0445], + device='cuda:2'), 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:2') +2023-02-09 02:29:44,879 INFO [optim.py:369] (2/4) 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 02:29:45,631 INFO [train.py:901] (2/4) Epoch 29, batch 5500, loss[loss=0.1522, simple_loss=0.2349, pruned_loss=0.03476, over 6864.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2821, pruned_loss=0.05748, over 1609365.97 frames. ], batch size: 15, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:29:48,650 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8465, 1.4875, 1.7592, 1.4295, 0.9165, 1.4966, 1.6624, 1.5675], + device='cuda:2'), covar=tensor([0.0614, 0.1214, 0.1645, 0.1490, 0.0625, 0.1424, 0.0726, 0.0662], + device='cuda:2'), in_proj_covar=tensor([0.0100, 0.0155, 0.0191, 0.0163, 0.0102, 0.0165, 0.0114, 0.0148], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0007, 0.0008], + device='cuda:2') +2023-02-09 02:30:21,383 INFO [train.py:901] (2/4) Epoch 29, batch 5550, loss[loss=0.1787, simple_loss=0.2614, pruned_loss=0.04804, over 7665.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2823, pruned_loss=0.05768, over 1608338.78 frames. ], batch size: 19, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:30:56,572 INFO [optim.py:369] (2/4) 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] (2/4) Epoch 29, batch 5600, loss[loss=0.1633, simple_loss=0.2607, pruned_loss=0.03298, over 7811.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2815, pruned_loss=0.05734, over 1606234.81 frames. ], batch size: 20, lr: 2.59e-03, grad_scale: 8.0 +2023-02-09 02:31:34,506 INFO [train.py:901] (2/4) Epoch 29, batch 5650, loss[loss=0.1907, simple_loss=0.284, pruned_loss=0.04876, over 8024.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2822, pruned_loss=0.05764, over 1610877.26 frames. ], batch size: 22, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:31:38,805 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-09 02:31:43,799 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=231984.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:31:58,932 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.75 vs. limit=5.0 +2023-02-09 02:32:02,831 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=232009.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:32:02,956 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=232009.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:32:10,323 INFO [optim.py:369] (2/4) 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 02:32:11,030 INFO [train.py:901] (2/4) Epoch 29, batch 5700, loss[loss=0.2099, simple_loss=0.287, pruned_loss=0.06645, over 8358.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2819, pruned_loss=0.05712, over 1613367.79 frames. ], batch size: 24, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:32:45,197 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-09 02:32:47,262 INFO [train.py:901] (2/4) Epoch 29, batch 5750, loss[loss=0.2333, simple_loss=0.3144, pruned_loss=0.07613, over 8534.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2816, pruned_loss=0.05682, over 1615832.89 frames. ], batch size: 31, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:33:23,822 INFO [optim.py:369] (2/4) 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 02:33:24,554 INFO [train.py:901] (2/4) Epoch 29, batch 5800, loss[loss=0.1689, simple_loss=0.25, pruned_loss=0.04387, over 7808.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2805, pruned_loss=0.05621, over 1615712.49 frames. ], batch size: 19, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:33:26,838 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=232124.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:33:35,064 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.6144, 2.2646, 3.1631, 1.6713, 1.7980, 3.1353, 0.8826, 2.1258], + device='cuda:2'), covar=tensor([0.1447, 0.0977, 0.0277, 0.1414, 0.2058, 0.0358, 0.1844, 0.1259], + device='cuda:2'), in_proj_covar=tensor([0.0201, 0.0204, 0.0137, 0.0224, 0.0278, 0.0148, 0.0172, 0.0200], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 02:33:43,977 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.8882, 2.3570, 3.6373, 1.6344, 1.7732, 3.6146, 0.7704, 2.0498], + device='cuda:2'), covar=tensor([0.1303, 0.1035, 0.0208, 0.1666, 0.2260, 0.0251, 0.1875, 0.1366], + device='cuda:2'), in_proj_covar=tensor([0.0201, 0.0204, 0.0137, 0.0224, 0.0278, 0.0148, 0.0172, 0.0200], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 02:33:59,563 INFO [train.py:901] (2/4) Epoch 29, batch 5850, loss[loss=0.188, simple_loss=0.275, pruned_loss=0.05045, over 8473.00 frames. ], tot_loss[loss=0.1956, simple_loss=0.2796, pruned_loss=0.05577, over 1613489.56 frames. ], batch size: 29, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:34:31,415 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9280, 1.4869, 1.7441, 1.4031, 0.9037, 1.5259, 1.6931, 1.5676], + device='cuda:2'), covar=tensor([0.0540, 0.1315, 0.1641, 0.1491, 0.0617, 0.1480, 0.0713, 0.0654], + device='cuda:2'), in_proj_covar=tensor([0.0100, 0.0155, 0.0191, 0.0163, 0.0102, 0.0165, 0.0115, 0.0148], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0008, 0.0008], + device='cuda:2') +2023-02-09 02:34:34,659 INFO [optim.py:369] (2/4) 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 02:34:35,360 INFO [train.py:901] (2/4) Epoch 29, batch 5900, loss[loss=0.2149, simple_loss=0.2973, pruned_loss=0.0662, over 8657.00 frames. ], tot_loss[loss=0.1951, simple_loss=0.2795, pruned_loss=0.05531, over 1617119.42 frames. ], batch size: 39, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:34:44,585 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8514, 1.3384, 4.0024, 1.4752, 3.5616, 3.3704, 3.6460, 3.5247], + device='cuda:2'), covar=tensor([0.0738, 0.4669, 0.0655, 0.4356, 0.1190, 0.1015, 0.0701, 0.0805], + device='cuda:2'), in_proj_covar=tensor([0.0688, 0.0667, 0.0748, 0.0659, 0.0747, 0.0637, 0.0643, 0.0724], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 02:35:06,628 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=232265.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:35:10,677 INFO [train.py:901] (2/4) Epoch 29, batch 5950, loss[loss=0.1975, simple_loss=0.2754, pruned_loss=0.05982, over 8139.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.2803, pruned_loss=0.05539, over 1622444.53 frames. ], batch size: 22, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:35:26,278 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-09 02:35:46,609 INFO [optim.py:369] (2/4) 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 02:35:47,356 INFO [train.py:901] (2/4) Epoch 29, batch 6000, loss[loss=0.1742, simple_loss=0.2629, pruned_loss=0.04278, over 8734.00 frames. ], tot_loss[loss=0.1951, simple_loss=0.2797, pruned_loss=0.05518, over 1616446.95 frames. ], batch size: 30, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:35:47,357 INFO [train.py:926] (2/4) Computing validation loss +2023-02-09 02:36:01,202 INFO [train.py:935] (2/4) Epoch 29, validation: loss=0.1708, simple_loss=0.2701, pruned_loss=0.03577, over 944034.00 frames. +2023-02-09 02:36:01,203 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6724MB +2023-02-09 02:36:14,198 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.75 vs. limit=5.0 +2023-02-09 02:36:37,559 INFO [train.py:901] (2/4) Epoch 29, batch 6050, loss[loss=0.1959, simple_loss=0.2723, pruned_loss=0.05978, over 7655.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.2803, pruned_loss=0.05573, over 1618995.33 frames. ], batch size: 19, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:36:44,009 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=232380.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:36:57,840 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=232399.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:37:02,392 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=232405.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:37:12,722 INFO [optim.py:369] (2/4) 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 02:37:13,448 INFO [train.py:901] (2/4) Epoch 29, batch 6100, loss[loss=0.2036, simple_loss=0.2913, pruned_loss=0.05797, over 8333.00 frames. ], tot_loss[loss=0.1956, simple_loss=0.2803, pruned_loss=0.0555, over 1618038.94 frames. ], batch size: 26, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:37:27,120 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-09 02:37:49,993 INFO [train.py:901] (2/4) Epoch 29, batch 6150, loss[loss=0.1918, simple_loss=0.2602, pruned_loss=0.06174, over 7793.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2811, pruned_loss=0.05584, over 1614448.35 frames. ], batch size: 19, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:38:06,451 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=232494.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:38:25,293 INFO [optim.py:369] (2/4) 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 02:38:25,872 INFO [train.py:901] (2/4) Epoch 29, batch 6200, loss[loss=0.2618, simple_loss=0.3356, pruned_loss=0.09399, over 8409.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2818, pruned_loss=0.05697, over 1605807.17 frames. ], batch size: 49, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:38:36,676 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=232536.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:38:41,134 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.16 vs. limit=2.0 +2023-02-09 02:38:47,162 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8005, 1.4791, 1.7484, 1.4064, 0.9417, 1.5239, 1.7154, 1.5666], + device='cuda:2'), covar=tensor([0.0612, 0.1259, 0.1610, 0.1465, 0.0622, 0.1450, 0.0711, 0.0663], + device='cuda:2'), in_proj_covar=tensor([0.0100, 0.0154, 0.0190, 0.0162, 0.0102, 0.0164, 0.0114, 0.0148], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0007, 0.0008], + device='cuda:2') +2023-02-09 02:39:02,798 INFO [train.py:901] (2/4) Epoch 29, batch 6250, loss[loss=0.1694, simple_loss=0.2606, pruned_loss=0.03907, over 7973.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2811, pruned_loss=0.0569, over 1603196.84 frames. ], batch size: 21, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:39:30,162 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=232609.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:39:36,585 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2094, 2.1097, 1.6539, 1.9199, 1.6997, 1.4521, 1.6568, 1.6748], + device='cuda:2'), covar=tensor([0.1434, 0.0459, 0.1321, 0.0593, 0.0803, 0.1614, 0.1023, 0.0843], + device='cuda:2'), in_proj_covar=tensor([0.0358, 0.0246, 0.0347, 0.0316, 0.0304, 0.0352, 0.0352, 0.0322], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 02:39:37,831 INFO [optim.py:369] (2/4) 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 02:39:39,193 INFO [train.py:901] (2/4) Epoch 29, batch 6300, loss[loss=0.1851, simple_loss=0.276, pruned_loss=0.04708, over 8024.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.281, pruned_loss=0.05687, over 1602563.52 frames. ], batch size: 22, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:40:14,954 INFO [train.py:901] (2/4) Epoch 29, batch 6350, loss[loss=0.1494, simple_loss=0.2247, pruned_loss=0.03702, over 7694.00 frames. ], tot_loss[loss=0.196, simple_loss=0.2801, pruned_loss=0.05598, over 1607452.73 frames. ], batch size: 18, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:40:50,678 INFO [optim.py:369] (2/4) 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 02:40:51,324 INFO [train.py:901] (2/4) Epoch 29, batch 6400, loss[loss=0.207, simple_loss=0.2943, pruned_loss=0.05987, over 8319.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2807, pruned_loss=0.05655, over 1608382.37 frames. ], batch size: 25, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:40:53,608 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=232724.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:41:08,029 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=232743.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:41:28,293 INFO [train.py:901] (2/4) Epoch 29, batch 6450, loss[loss=0.1666, simple_loss=0.247, pruned_loss=0.04309, over 7191.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2801, pruned_loss=0.05642, over 1610686.48 frames. ], batch size: 16, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:42:05,012 INFO [optim.py:369] (2/4) 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 02:42:05,753 INFO [train.py:901] (2/4) Epoch 29, batch 6500, loss[loss=0.192, simple_loss=0.2752, pruned_loss=0.05437, over 8472.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.281, pruned_loss=0.05686, over 1618643.52 frames. ], batch size: 25, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:42:17,441 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=232838.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:42:31,253 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=232858.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:42:38,231 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6244, 2.1127, 3.0802, 1.4866, 2.3108, 2.0361, 1.7934, 2.4659], + device='cuda:2'), covar=tensor([0.1991, 0.2690, 0.0981, 0.4928, 0.2158, 0.3475, 0.2494, 0.2476], + device='cuda:2'), in_proj_covar=tensor([0.0544, 0.0641, 0.0566, 0.0675, 0.0665, 0.0614, 0.0569, 0.0647], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 02:42:40,786 INFO [train.py:901] (2/4) Epoch 29, batch 6550, loss[loss=0.2282, simple_loss=0.3217, pruned_loss=0.06738, over 8456.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2815, pruned_loss=0.05673, over 1620996.40 frames. ], batch size: 25, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:42:47,138 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=232880.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:42:47,798 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-09 02:43:00,699 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5185, 1.8007, 1.8811, 1.2319, 2.0051, 1.3054, 0.5940, 1.8265], + device='cuda:2'), covar=tensor([0.0831, 0.0451, 0.0360, 0.0743, 0.0486, 0.1078, 0.1113, 0.0394], + device='cuda:2'), in_proj_covar=tensor([0.0480, 0.0420, 0.0375, 0.0467, 0.0402, 0.0560, 0.0410, 0.0447], + device='cuda:2'), out_proj_covar=tensor([1.2700e-04, 1.0875e-04, 9.7622e-05, 1.2210e-04, 1.0533e-04, 1.5572e-04, + 1.0920e-04, 1.1695e-04], device='cuda:2') +2023-02-09 02:43:02,013 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7524, 2.6746, 2.0068, 2.4142, 2.2363, 1.8081, 2.2052, 2.3051], + device='cuda:2'), covar=tensor([0.1555, 0.0441, 0.1126, 0.0645, 0.0815, 0.1427, 0.0971, 0.0861], + device='cuda:2'), in_proj_covar=tensor([0.0359, 0.0247, 0.0347, 0.0317, 0.0304, 0.0352, 0.0353, 0.0322], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 02:43:08,164 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-09 02:43:11,912 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4473, 1.5313, 1.4278, 1.8655, 0.7809, 1.3385, 1.3486, 1.5444], + device='cuda:2'), covar=tensor([0.0853, 0.0779, 0.0930, 0.0437, 0.1070, 0.1391, 0.0737, 0.0699], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0196, 0.0246, 0.0215, 0.0204, 0.0247, 0.0251, 0.0206], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 02:43:16,620 INFO [optim.py:369] (2/4) 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 02:43:17,343 INFO [train.py:901] (2/4) Epoch 29, batch 6600, loss[loss=0.16, simple_loss=0.247, pruned_loss=0.03648, over 7815.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2812, pruned_loss=0.0568, over 1614472.35 frames. ], batch size: 20, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:43:40,635 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=232953.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:43:52,850 INFO [train.py:901] (2/4) Epoch 29, batch 6650, loss[loss=0.2048, simple_loss=0.2902, pruned_loss=0.05971, over 8130.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2809, pruned_loss=0.0564, over 1616009.56 frames. ], batch size: 22, lr: 2.58e-03, grad_scale: 16.0 +2023-02-09 02:43:59,983 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=232980.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:44:10,257 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=232995.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:44:17,495 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233005.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:44:19,556 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233008.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:44:28,482 INFO [optim.py:369] (2/4) 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] (2/4) Epoch 29, batch 6700, loss[loss=0.2092, simple_loss=0.2987, pruned_loss=0.05983, over 8461.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2809, pruned_loss=0.05657, over 1618983.86 frames. ], batch size: 25, lr: 2.58e-03, grad_scale: 8.0 +2023-02-09 02:44:40,611 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233036.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:45:06,280 INFO [train.py:901] (2/4) Epoch 29, batch 6750, loss[loss=0.1943, simple_loss=0.2758, pruned_loss=0.05637, over 8249.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.2799, pruned_loss=0.05593, over 1617183.39 frames. ], batch size: 22, lr: 2.58e-03, grad_scale: 8.0 +2023-02-09 02:45:30,225 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4122, 2.3490, 3.0080, 2.4822, 3.0852, 2.5801, 2.4210, 1.8624], + device='cuda:2'), covar=tensor([0.6054, 0.5462, 0.2327, 0.4119, 0.2611, 0.3269, 0.1884, 0.6125], + device='cuda:2'), in_proj_covar=tensor([0.0972, 0.1035, 0.0843, 0.1004, 0.1028, 0.0940, 0.0777, 0.0857], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 02:45:30,752 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-09 02:45:38,406 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233114.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:45:43,344 INFO [optim.py:369] (2/4) 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 02:45:43,365 INFO [train.py:901] (2/4) Epoch 29, batch 6800, loss[loss=0.2094, simple_loss=0.2803, pruned_loss=0.06929, over 8125.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2802, pruned_loss=0.05649, over 1609594.14 frames. ], batch size: 22, lr: 2.58e-03, grad_scale: 8.0 +2023-02-09 02:45:56,014 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233139.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:46:03,486 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7016, 1.4786, 4.8580, 1.8634, 4.3614, 4.0709, 4.3832, 4.2783], + device='cuda:2'), covar=tensor([0.0557, 0.5071, 0.0541, 0.4298, 0.1063, 0.0942, 0.0607, 0.0680], + device='cuda:2'), in_proj_covar=tensor([0.0691, 0.0667, 0.0751, 0.0660, 0.0748, 0.0639, 0.0644, 0.0728], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 02:46:19,946 INFO [train.py:901] (2/4) Epoch 29, batch 6850, loss[loss=0.1833, simple_loss=0.2755, pruned_loss=0.04556, over 8188.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2799, pruned_loss=0.0565, over 1609054.90 frames. ], batch size: 23, lr: 2.58e-03, grad_scale: 8.0 +2023-02-09 02:46:22,009 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-09 02:46:23,567 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8309, 3.7946, 3.4314, 1.8088, 3.4046, 3.5538, 3.3565, 3.2904], + device='cuda:2'), covar=tensor([0.0886, 0.0629, 0.1156, 0.4904, 0.1055, 0.1314, 0.1428, 0.0996], + device='cuda:2'), in_proj_covar=tensor([0.0548, 0.0461, 0.0454, 0.0563, 0.0446, 0.0472, 0.0446, 0.0415], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 02:46:30,301 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.51 vs. limit=2.0 +2023-02-09 02:46:46,499 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233209.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:46:55,110 INFO [optim.py:369] (2/4) 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 02:46:55,130 INFO [train.py:901] (2/4) Epoch 29, batch 6900, loss[loss=0.1849, simple_loss=0.2639, pruned_loss=0.05291, over 7416.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2811, pruned_loss=0.05699, over 1611472.11 frames. ], batch size: 17, lr: 2.58e-03, grad_scale: 8.0 +2023-02-09 02:47:04,465 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233234.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:47:16,261 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 29, batch 6950, loss[loss=0.2121, simple_loss=0.2989, pruned_loss=0.06271, over 8644.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2803, pruned_loss=0.0565, over 1612119.77 frames. ], batch size: 34, lr: 2.58e-03, grad_scale: 8.0 +2023-02-09 02:47:33,663 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-09 02:47:35,276 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233276.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:47:35,914 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233277.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:48:07,352 INFO [optim.py:369] (2/4) 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 02:48:07,372 INFO [train.py:901] (2/4) Epoch 29, batch 7000, loss[loss=0.1698, simple_loss=0.2559, pruned_loss=0.04185, over 8479.00 frames. ], tot_loss[loss=0.1952, simple_loss=0.2795, pruned_loss=0.05549, over 1611259.80 frames. ], batch size: 25, lr: 2.58e-03, grad_scale: 8.0 +2023-02-09 02:48:30,363 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=233352.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:48:40,382 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-09 02:48:44,307 INFO [train.py:901] (2/4) Epoch 29, batch 7050, loss[loss=0.1929, simple_loss=0.2719, pruned_loss=0.05694, over 7913.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.28, pruned_loss=0.0559, over 1607001.85 frames. ], batch size: 20, lr: 2.58e-03, grad_scale: 8.0 +2023-02-09 02:48:51,038 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=233380.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:49:09,793 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2594, 1.4003, 4.3456, 2.1040, 2.5615, 4.8935, 5.0312, 4.1744], + device='cuda:2'), covar=tensor([0.1269, 0.2219, 0.0276, 0.1979, 0.1225, 0.0187, 0.0535, 0.0607], + device='cuda:2'), in_proj_covar=tensor([0.0310, 0.0328, 0.0296, 0.0326, 0.0327, 0.0279, 0.0447, 0.0310], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:2') +2023-02-09 02:49:22,021 INFO [optim.py:369] (2/4) 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 02:49:22,041 INFO [train.py:901] (2/4) Epoch 29, batch 7100, loss[loss=0.1943, simple_loss=0.2888, pruned_loss=0.04989, over 8602.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.2802, pruned_loss=0.05602, over 1608883.09 frames. ], batch size: 34, lr: 2.58e-03, grad_scale: 8.0 +2023-02-09 02:49:34,962 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1978, 2.0851, 2.6577, 2.2166, 2.6694, 2.3183, 2.1860, 1.5863], + device='cuda:2'), covar=tensor([0.5730, 0.5073, 0.2095, 0.4322, 0.2870, 0.3284, 0.1996, 0.5746], + device='cuda:2'), in_proj_covar=tensor([0.0971, 0.1035, 0.0841, 0.1005, 0.1029, 0.0941, 0.0778, 0.0855], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 02:49:48,645 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.48 vs. limit=2.0 +2023-02-09 02:49:54,755 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 29, batch 7150, loss[loss=0.2271, simple_loss=0.3119, pruned_loss=0.07118, over 8581.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2809, pruned_loss=0.0562, over 1615409.80 frames. ], batch size: 31, lr: 2.58e-03, grad_scale: 8.0 +2023-02-09 02:50:00,189 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9916, 1.4797, 1.6756, 1.3813, 1.2100, 1.3837, 2.0291, 1.5253], + device='cuda:2'), covar=tensor([0.0578, 0.1314, 0.1700, 0.1510, 0.0581, 0.1605, 0.0668, 0.0703], + device='cuda:2'), in_proj_covar=tensor([0.0101, 0.0155, 0.0191, 0.0162, 0.0102, 0.0165, 0.0114, 0.0148], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0007, 0.0008], + device='cuda:2') +2023-02-09 02:50:14,747 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233495.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:50:34,190 INFO [optim.py:369] (2/4) 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 02:50:34,211 INFO [train.py:901] (2/4) Epoch 29, batch 7200, loss[loss=0.1649, simple_loss=0.2442, pruned_loss=0.04282, over 7428.00 frames. ], tot_loss[loss=0.196, simple_loss=0.2802, pruned_loss=0.05591, over 1614145.07 frames. ], batch size: 17, lr: 2.58e-03, grad_scale: 8.0 +2023-02-09 02:50:37,514 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.59 vs. limit=2.0 +2023-02-09 02:50:47,427 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7854, 2.0207, 2.0845, 1.4929, 2.2478, 1.5713, 0.7190, 2.0130], + device='cuda:2'), covar=tensor([0.0667, 0.0416, 0.0344, 0.0638, 0.0456, 0.0988, 0.1049, 0.0363], + device='cuda:2'), in_proj_covar=tensor([0.0477, 0.0417, 0.0374, 0.0464, 0.0400, 0.0560, 0.0408, 0.0446], + device='cuda:2'), out_proj_covar=tensor([1.2620e-04, 1.0794e-04, 9.7401e-05, 1.2124e-04, 1.0481e-04, 1.5582e-04, + 1.0882e-04, 1.1654e-04], device='cuda:2') +2023-02-09 02:51:10,563 INFO [train.py:901] (2/4) Epoch 29, batch 7250, loss[loss=0.1686, simple_loss=0.2556, pruned_loss=0.0408, over 7521.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2819, pruned_loss=0.05664, over 1616689.84 frames. ], batch size: 18, lr: 2.58e-03, grad_scale: 8.0 +2023-02-09 02:51:31,468 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 +2023-02-09 02:51:46,281 INFO [optim.py:369] (2/4) 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 02:51:46,302 INFO [train.py:901] (2/4) Epoch 29, batch 7300, loss[loss=0.1616, simple_loss=0.2335, pruned_loss=0.04487, over 7531.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2819, pruned_loss=0.05673, over 1614098.69 frames. ], batch size: 18, lr: 2.58e-03, grad_scale: 8.0 +2023-02-09 02:51:46,383 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=233621.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:52:13,024 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233657.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:52:22,375 INFO [train.py:901] (2/4) Epoch 29, batch 7350, loss[loss=0.2004, simple_loss=0.2889, pruned_loss=0.05599, over 8477.00 frames. ], tot_loss[loss=0.1993, simple_loss=0.2831, pruned_loss=0.05774, over 1613441.72 frames. ], batch size: 25, lr: 2.58e-03, grad_scale: 8.0 +2023-02-09 02:52:28,052 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-09 02:52:39,772 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-09 02:52:48,258 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-09 02:52:58,115 INFO [optim.py:369] (2/4) 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 02:52:58,135 INFO [train.py:901] (2/4) Epoch 29, batch 7400, loss[loss=0.205, simple_loss=0.2897, pruned_loss=0.06018, over 8612.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.282, pruned_loss=0.05709, over 1611940.10 frames. ], batch size: 34, lr: 2.58e-03, grad_scale: 8.0 +2023-02-09 02:52:59,682 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233723.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:53:03,228 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=233728.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:53:08,949 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=233736.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:53:18,933 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233748.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:53:21,096 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233751.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:53:32,421 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-09 02:53:36,088 INFO [train.py:901] (2/4) Epoch 29, batch 7450, loss[loss=0.1816, simple_loss=0.2638, pruned_loss=0.0497, over 7934.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2804, pruned_loss=0.05596, over 1612142.37 frames. ], batch size: 20, lr: 2.58e-03, grad_scale: 8.0 +2023-02-09 02:53:39,815 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=233776.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:54:12,675 INFO [optim.py:369] (2/4) 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 02:54:12,695 INFO [train.py:901] (2/4) Epoch 29, batch 7500, loss[loss=0.1781, simple_loss=0.2577, pruned_loss=0.04924, over 7792.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2811, pruned_loss=0.05617, over 1614169.62 frames. ], batch size: 19, lr: 2.57e-03, grad_scale: 8.0 +2023-02-09 02:54:48,750 INFO [train.py:901] (2/4) Epoch 29, batch 7550, loss[loss=0.1581, simple_loss=0.2423, pruned_loss=0.03694, over 5995.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2808, pruned_loss=0.05643, over 1610885.43 frames. ], batch size: 13, lr: 2.57e-03, grad_scale: 8.0 +2023-02-09 02:55:24,355 INFO [optim.py:369] (2/4) 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 02:55:24,375 INFO [train.py:901] (2/4) Epoch 29, batch 7600, loss[loss=0.1677, simple_loss=0.2592, pruned_loss=0.0381, over 7929.00 frames. ], tot_loss[loss=0.1962, simple_loss=0.2805, pruned_loss=0.05592, over 1614230.28 frames. ], batch size: 20, lr: 2.57e-03, grad_scale: 8.0 +2023-02-09 02:56:01,017 INFO [train.py:901] (2/4) Epoch 29, batch 7650, loss[loss=0.1955, simple_loss=0.2827, pruned_loss=0.05412, over 8322.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.28, pruned_loss=0.05552, over 1613479.78 frames. ], batch size: 25, lr: 2.57e-03, grad_scale: 8.0 +2023-02-09 02:56:16,601 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=233992.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:56:23,680 INFO [zipformer.py:1185] (2/4) 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] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234017.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:56:38,046 INFO [optim.py:369] (2/4) 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 02:56:38,066 INFO [train.py:901] (2/4) Epoch 29, batch 7700, loss[loss=0.1746, simple_loss=0.2635, pruned_loss=0.04278, over 8079.00 frames. ], tot_loss[loss=0.1946, simple_loss=0.2793, pruned_loss=0.05491, over 1615867.48 frames. ], batch size: 21, lr: 2.57e-03, grad_scale: 8.0 +2023-02-09 02:56:49,503 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-09 02:56:53,151 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234043.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:57:12,872 INFO [train.py:901] (2/4) Epoch 29, batch 7750, loss[loss=0.1662, simple_loss=0.2432, pruned_loss=0.04465, over 7417.00 frames. ], tot_loss[loss=0.1931, simple_loss=0.2776, pruned_loss=0.05428, over 1613544.79 frames. ], batch size: 17, lr: 2.57e-03, grad_scale: 8.0 +2023-02-09 02:57:13,650 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=234072.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:57:43,295 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.11 vs. limit=5.0 +2023-02-09 02:57:45,239 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234116.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:57:48,462 INFO [optim.py:369] (2/4) 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 02:57:48,482 INFO [train.py:901] (2/4) Epoch 29, batch 7800, loss[loss=0.1897, simple_loss=0.275, pruned_loss=0.05224, over 8034.00 frames. ], tot_loss[loss=0.1926, simple_loss=0.2774, pruned_loss=0.05391, over 1616587.31 frames. ], batch size: 22, lr: 2.57e-03, grad_scale: 8.0 +2023-02-09 02:58:09,969 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.3410, 2.0580, 2.5459, 2.1226, 2.5380, 2.3778, 2.2309, 1.4165], + device='cuda:2'), covar=tensor([0.6169, 0.5484, 0.2468, 0.4671, 0.3023, 0.3683, 0.2046, 0.6103], + device='cuda:2'), in_proj_covar=tensor([0.0973, 0.1036, 0.0845, 0.1009, 0.1034, 0.0945, 0.0780, 0.0859], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 02:58:24,431 INFO [train.py:901] (2/4) Epoch 29, batch 7850, loss[loss=0.1612, simple_loss=0.244, pruned_loss=0.03923, over 8080.00 frames. ], tot_loss[loss=0.1916, simple_loss=0.2763, pruned_loss=0.05349, over 1612505.73 frames. ], batch size: 21, lr: 2.57e-03, grad_scale: 8.0 +2023-02-09 02:58:26,120 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8090, 2.3906, 3.6852, 1.6710, 2.8506, 2.3199, 1.9607, 2.8910], + device='cuda:2'), covar=tensor([0.1924, 0.2645, 0.0958, 0.4679, 0.1881, 0.3224, 0.2490, 0.2451], + device='cuda:2'), in_proj_covar=tensor([0.0540, 0.0637, 0.0563, 0.0670, 0.0660, 0.0613, 0.0564, 0.0643], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 02:58:36,027 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234187.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:58:58,922 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234220.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:58:59,504 INFO [optim.py:369] (2/4) 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 02:58:59,525 INFO [train.py:901] (2/4) Epoch 29, batch 7900, loss[loss=0.1804, simple_loss=0.2705, pruned_loss=0.04514, over 8459.00 frames. ], tot_loss[loss=0.1917, simple_loss=0.2764, pruned_loss=0.0535, over 1614056.20 frames. ], batch size: 25, lr: 2.57e-03, grad_scale: 8.0 +2023-02-09 02:59:19,146 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234249.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 02:59:34,076 INFO [train.py:901] (2/4) Epoch 29, batch 7950, loss[loss=0.1945, simple_loss=0.2697, pruned_loss=0.05967, over 7917.00 frames. ], tot_loss[loss=0.1911, simple_loss=0.2757, pruned_loss=0.0533, over 1608266.11 frames. ], batch size: 20, lr: 2.57e-03, grad_scale: 8.0 +2023-02-09 03:00:10,396 INFO [optim.py:369] (2/4) 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 03:00:10,416 INFO [train.py:901] (2/4) Epoch 29, batch 8000, loss[loss=0.173, simple_loss=0.2523, pruned_loss=0.04691, over 7968.00 frames. ], tot_loss[loss=0.1926, simple_loss=0.2774, pruned_loss=0.05389, over 1611638.68 frames. ], batch size: 21, lr: 2.57e-03, grad_scale: 8.0 +2023-02-09 03:00:30,577 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1435, 1.6400, 1.8674, 1.4540, 1.0283, 1.5640, 1.8246, 1.7802], + device='cuda:2'), covar=tensor([0.0530, 0.1185, 0.1592, 0.1465, 0.0615, 0.1421, 0.0689, 0.0618], + device='cuda:2'), in_proj_covar=tensor([0.0101, 0.0154, 0.0191, 0.0163, 0.0102, 0.0165, 0.0114, 0.0148], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0007, 0.0008], + device='cuda:2') +2023-02-09 03:00:44,624 INFO [train.py:901] (2/4) Epoch 29, batch 8050, loss[loss=0.179, simple_loss=0.2542, pruned_loss=0.05188, over 7245.00 frames. ], tot_loss[loss=0.1913, simple_loss=0.2747, pruned_loss=0.05392, over 1584701.34 frames. ], batch size: 16, lr: 2.57e-03, grad_scale: 8.0 +2023-02-09 03:00:45,524 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234372.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:00:55,749 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=234387.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:01:02,568 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234397.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:01:20,235 WARNING [train.py:1067] (2/4) Exclude cut with ID 3488-85273-0017-111273_sp0.9 from training. Duration: 27.47775 +2023-02-09 03:01:23,761 INFO [train.py:901] (2/4) Epoch 30, batch 0, loss[loss=0.2019, simple_loss=0.2894, pruned_loss=0.05719, over 8451.00 frames. ], tot_loss[loss=0.2019, simple_loss=0.2894, pruned_loss=0.05719, over 8451.00 frames. ], batch size: 25, lr: 2.53e-03, grad_scale: 8.0 +2023-02-09 03:01:23,762 INFO [train.py:926] (2/4) Computing validation loss +2023-02-09 03:01:35,940 INFO [train.py:935] (2/4) Epoch 30, validation: loss=0.1704, simple_loss=0.27, pruned_loss=0.03537, over 944034.00 frames. +2023-02-09 03:01:35,941 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6724MB +2023-02-09 03:01:47,831 INFO [optim.py:369] (2/4) 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 03:01:52,033 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590_sp0.9 from training. Duration: 28.72225 +2023-02-09 03:02:04,493 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234443.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:02:12,495 INFO [train.py:901] (2/4) Epoch 30, batch 50, loss[loss=0.203, simple_loss=0.2897, pruned_loss=0.05815, over 8350.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2821, pruned_loss=0.05709, over 368506.84 frames. ], batch size: 24, lr: 2.53e-03, grad_scale: 8.0 +2023-02-09 03:02:23,084 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234468.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:02:28,106 WARNING [train.py:1067] (2/4) Exclude cut with ID 6709-74022-0004-57021_sp1.1 from training. Duration: 0.9409375 +2023-02-09 03:02:49,595 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234502.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:02:50,898 INFO [train.py:901] (2/4) Epoch 30, batch 100, loss[loss=0.2097, simple_loss=0.2949, pruned_loss=0.06221, over 8436.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2823, pruned_loss=0.05634, over 648631.30 frames. ], batch size: 27, lr: 2.53e-03, grad_scale: 8.0 +2023-02-09 03:02:54,581 WARNING [train.py:1067] (2/4) Exclude cut with ID 497-129325-0061-9566_sp1.1 from training. Duration: 0.97725 +2023-02-09 03:03:03,348 INFO [optim.py:369] (2/4) 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 03:03:28,106 INFO [train.py:901] (2/4) Epoch 30, batch 150, loss[loss=0.269, simple_loss=0.3257, pruned_loss=0.1062, over 6920.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2838, pruned_loss=0.05707, over 866739.75 frames. ], batch size: 71, lr: 2.53e-03, grad_scale: 8.0 +2023-02-09 03:03:35,359 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=234564.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:03:56,911 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=234593.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:04:04,594 INFO [train.py:901] (2/4) Epoch 30, batch 200, loss[loss=0.2082, simple_loss=0.2863, pruned_loss=0.06503, over 8451.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2834, pruned_loss=0.05649, over 1031108.05 frames. ], batch size: 27, lr: 2.53e-03, grad_scale: 8.0 +2023-02-09 03:04:13,775 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4110, 2.1953, 2.6679, 2.3146, 2.7691, 2.4566, 2.3248, 1.6855], + device='cuda:2'), covar=tensor([0.5663, 0.5295, 0.2372, 0.4237, 0.2613, 0.3500, 0.1963, 0.5716], + device='cuda:2'), in_proj_covar=tensor([0.0973, 0.1039, 0.0847, 0.1012, 0.1034, 0.0945, 0.0781, 0.0861], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 03:04:16,945 INFO [optim.py:369] (2/4) 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 03:04:40,208 INFO [train.py:901] (2/4) Epoch 30, batch 250, loss[loss=0.1823, simple_loss=0.2657, pruned_loss=0.04947, over 7648.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.283, pruned_loss=0.0561, over 1162805.44 frames. ], batch size: 19, lr: 2.53e-03, grad_scale: 8.0 +2023-02-09 03:04:48,468 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149_sp0.9 from training. Duration: 28.0944375 +2023-02-09 03:04:57,627 WARNING [train.py:1067] (2/4) Exclude cut with ID 4278-13270-0009-62705_sp0.9 from training. Duration: 25.45 +2023-02-09 03:04:58,544 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234679.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:05:04,641 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234687.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:05:16,613 INFO [train.py:901] (2/4) Epoch 30, batch 300, loss[loss=0.1807, simple_loss=0.2816, pruned_loss=0.03988, over 8324.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2814, pruned_loss=0.05592, over 1263368.80 frames. ], batch size: 25, lr: 2.53e-03, grad_scale: 8.0 +2023-02-09 03:05:19,746 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=234708.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:05:29,385 INFO [optim.py:369] (2/4) 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 03:05:45,711 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234744.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:05:46,509 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5951, 1.8459, 1.8443, 1.3544, 1.9469, 1.4052, 0.4393, 1.8237], + device='cuda:2'), covar=tensor([0.0686, 0.0422, 0.0397, 0.0630, 0.0573, 0.1067, 0.1076, 0.0327], + device='cuda:2'), in_proj_covar=tensor([0.0477, 0.0416, 0.0372, 0.0465, 0.0400, 0.0555, 0.0405, 0.0443], + device='cuda:2'), 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:2') +2023-02-09 03:05:52,651 INFO [train.py:901] (2/4) Epoch 30, batch 350, loss[loss=0.1966, simple_loss=0.2897, pruned_loss=0.05181, over 8032.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2814, pruned_loss=0.05619, over 1342170.60 frames. ], batch size: 22, lr: 2.53e-03, grad_scale: 8.0 +2023-02-09 03:05:55,592 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234758.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:06:14,215 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234783.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:06:27,357 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234800.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:06:30,081 INFO [train.py:901] (2/4) Epoch 30, batch 400, loss[loss=0.1919, simple_loss=0.2861, pruned_loss=0.04883, over 8512.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.2796, pruned_loss=0.05571, over 1400351.42 frames. ], batch size: 26, lr: 2.53e-03, grad_scale: 8.0 +2023-02-09 03:06:42,234 INFO [optim.py:369] (2/4) 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 03:07:06,518 INFO [train.py:901] (2/4) Epoch 30, batch 450, loss[loss=0.2001, simple_loss=0.2703, pruned_loss=0.06493, over 6782.00 frames. ], tot_loss[loss=0.1954, simple_loss=0.2799, pruned_loss=0.05546, over 1447790.92 frames. ], batch size: 15, lr: 2.53e-03, grad_scale: 8.0 +2023-02-09 03:07:38,957 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 30, batch 500, loss[loss=0.1901, simple_loss=0.2834, pruned_loss=0.04844, over 8333.00 frames. ], tot_loss[loss=0.194, simple_loss=0.2783, pruned_loss=0.05481, over 1486589.42 frames. ], batch size: 25, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:07:54,773 INFO [optim.py:369] (2/4) 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 03:08:04,863 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234935.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:08:18,288 INFO [train.py:901] (2/4) Epoch 30, batch 550, loss[loss=0.1795, simple_loss=0.2585, pruned_loss=0.05028, over 8088.00 frames. ], tot_loss[loss=0.1945, simple_loss=0.2787, pruned_loss=0.05516, over 1517518.76 frames. ], batch size: 21, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:08:22,856 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=234960.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:08:26,373 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=234964.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:08:34,797 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234976.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:08:36,243 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=234978.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:08:44,115 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 30, batch 600, loss[loss=0.2028, simple_loss=0.2954, pruned_loss=0.05512, over 8323.00 frames. ], tot_loss[loss=0.1943, simple_loss=0.2788, pruned_loss=0.05492, over 1542780.50 frames. ], batch size: 26, lr: 2.52e-03, grad_scale: 16.0 +2023-02-09 03:09:06,005 INFO [optim.py:369] (2/4) 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 03:09:10,828 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465_sp0.9 from training. Duration: 29.816625 +2023-02-09 03:09:13,652 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=235031.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:09:29,976 INFO [train.py:901] (2/4) Epoch 30, batch 650, loss[loss=0.2101, simple_loss=0.2947, pruned_loss=0.06271, over 8454.00 frames. ], tot_loss[loss=0.1954, simple_loss=0.2799, pruned_loss=0.05549, over 1561984.20 frames. ], batch size: 27, lr: 2.52e-03, grad_scale: 16.0 +2023-02-09 03:09:54,556 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=235088.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:10:05,672 INFO [train.py:901] (2/4) Epoch 30, batch 700, loss[loss=0.1929, simple_loss=0.2932, pruned_loss=0.04631, over 8191.00 frames. ], tot_loss[loss=0.1956, simple_loss=0.2804, pruned_loss=0.05536, over 1576301.73 frames. ], batch size: 23, lr: 2.52e-03, grad_scale: 16.0 +2023-02-09 03:10:12,448 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.55 vs. limit=2.0 +2023-02-09 03:10:17,689 INFO [optim.py:369] (2/4) 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 03:10:33,426 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=235144.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:10:34,916 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 30, batch 750, loss[loss=0.199, simple_loss=0.2718, pruned_loss=0.06309, over 7178.00 frames. ], tot_loss[loss=0.1948, simple_loss=0.2796, pruned_loss=0.05503, over 1582371.52 frames. ], batch size: 16, lr: 2.52e-03, grad_scale: 16.0 +2023-02-09 03:10:59,029 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994_sp0.9 from training. Duration: 30.1555625 +2023-02-09 03:10:59,124 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8378, 3.8081, 3.5400, 1.9602, 3.4147, 3.5842, 3.3871, 3.4332], + device='cuda:2'), covar=tensor([0.1024, 0.0670, 0.1075, 0.4262, 0.1019, 0.1066, 0.1537, 0.0886], + device='cuda:2'), in_proj_covar=tensor([0.0551, 0.0461, 0.0451, 0.0562, 0.0445, 0.0470, 0.0446, 0.0414], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 03:11:08,044 WARNING [train.py:1067] (2/4) Exclude cut with ID 8631-249866-0030-64025_sp0.9 from training. Duration: 26.32775 +2023-02-09 03:11:17,082 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235203.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:11:17,655 INFO [train.py:901] (2/4) Epoch 30, batch 800, loss[loss=0.1709, simple_loss=0.2611, pruned_loss=0.04035, over 8502.00 frames. ], tot_loss[loss=0.1943, simple_loss=0.2785, pruned_loss=0.055, over 1586220.36 frames. ], batch size: 28, lr: 2.52e-03, grad_scale: 16.0 +2023-02-09 03:11:30,473 INFO [optim.py:369] (2/4) 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 03:11:46,749 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=235244.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:11:53,490 INFO [train.py:901] (2/4) Epoch 30, batch 850, loss[loss=0.1917, simple_loss=0.2711, pruned_loss=0.05616, over 7536.00 frames. ], tot_loss[loss=0.1953, simple_loss=0.2795, pruned_loss=0.0556, over 1591737.81 frames. ], batch size: 18, lr: 2.52e-03, grad_scale: 16.0 +2023-02-09 03:11:57,057 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235259.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:12:07,383 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235273.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:12:20,619 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8027, 3.8008, 3.4677, 1.7851, 3.3217, 3.5212, 3.3474, 3.3282], + device='cuda:2'), covar=tensor([0.0991, 0.0699, 0.1041, 0.4714, 0.1086, 0.1119, 0.1497, 0.1016], + device='cuda:2'), in_proj_covar=tensor([0.0547, 0.0458, 0.0448, 0.0559, 0.0443, 0.0467, 0.0443, 0.0411], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 03:12:25,086 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.50 vs. limit=2.0 +2023-02-09 03:12:29,574 INFO [train.py:901] (2/4) Epoch 30, batch 900, loss[loss=0.2376, simple_loss=0.3171, pruned_loss=0.07901, over 8589.00 frames. ], tot_loss[loss=0.1951, simple_loss=0.2793, pruned_loss=0.05546, over 1598452.48 frames. ], batch size: 31, lr: 2.52e-03, grad_scale: 16.0 +2023-02-09 03:12:41,045 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=235320.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:12:41,602 INFO [optim.py:369] (2/4) 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 03:12:43,016 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=235322.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:13:05,251 INFO [train.py:901] (2/4) Epoch 30, batch 950, loss[loss=0.192, simple_loss=0.2932, pruned_loss=0.04538, over 8325.00 frames. ], tot_loss[loss=0.1956, simple_loss=0.2801, pruned_loss=0.05555, over 1604127.72 frames. ], batch size: 25, lr: 2.52e-03, grad_scale: 16.0 +2023-02-09 03:13:08,758 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235359.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:13:21,829 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235378.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 03:13:32,471 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3297, 2.0655, 1.6684, 1.9496, 1.7567, 1.4555, 1.6775, 1.6794], + device='cuda:2'), covar=tensor([0.1389, 0.0462, 0.1269, 0.0549, 0.0769, 0.1628, 0.0968, 0.0932], + device='cuda:2'), in_proj_covar=tensor([0.0357, 0.0243, 0.0344, 0.0314, 0.0302, 0.0349, 0.0350, 0.0321], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 03:13:32,983 WARNING [train.py:1067] (2/4) 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] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235402.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:13:40,158 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4928, 2.4549, 3.1055, 2.6252, 3.1813, 2.6345, 2.4861, 1.8659], + device='cuda:2'), covar=tensor([0.6566, 0.5597, 0.2414, 0.4289, 0.2842, 0.3318, 0.2052, 0.6398], + device='cuda:2'), in_proj_covar=tensor([0.0978, 0.1045, 0.0853, 0.1015, 0.1038, 0.0949, 0.0782, 0.0865], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 03:13:40,619 INFO [train.py:901] (2/4) Epoch 30, batch 1000, loss[loss=0.1625, simple_loss=0.2442, pruned_loss=0.04038, over 7657.00 frames. ], tot_loss[loss=0.1952, simple_loss=0.2792, pruned_loss=0.05559, over 1599712.48 frames. ], batch size: 19, lr: 2.52e-03, grad_scale: 16.0 +2023-02-09 03:13:52,267 INFO [optim.py:369] (2/4) 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 03:13:56,465 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235427.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:14:02,533 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235435.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:14:03,940 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235437.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:14:08,440 WARNING [train.py:1067] (2/4) Exclude cut with ID 6951-79737-0043-83149 from training. Duration: 25.285 +2023-02-09 03:14:15,934 INFO [train.py:901] (2/4) Epoch 30, batch 1050, loss[loss=0.2056, simple_loss=0.288, pruned_loss=0.06163, over 7974.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.2799, pruned_loss=0.05574, over 1602490.10 frames. ], batch size: 21, lr: 2.52e-03, grad_scale: 16.0 +2023-02-09 03:14:19,400 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235459.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:14:21,156 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403 from training. Duration: 29.735 +2023-02-09 03:14:36,650 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235484.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:14:41,812 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0 +2023-02-09 03:14:50,591 INFO [train.py:901] (2/4) Epoch 30, batch 1100, loss[loss=0.1961, simple_loss=0.2876, pruned_loss=0.0523, over 8289.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.28, pruned_loss=0.05588, over 1603700.90 frames. ], batch size: 23, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:14:59,106 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235515.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:15:03,871 INFO [optim.py:369] (2/4) 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 03:15:04,829 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5582, 1.8299, 1.8746, 1.2254, 1.9132, 1.3897, 0.4332, 1.7748], + device='cuda:2'), covar=tensor([0.0662, 0.0451, 0.0382, 0.0681, 0.0492, 0.1070, 0.1141, 0.0350], + device='cuda:2'), in_proj_covar=tensor([0.0483, 0.0421, 0.0375, 0.0470, 0.0404, 0.0561, 0.0410, 0.0448], + device='cuda:2'), 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:2') +2023-02-09 03:15:09,968 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.56 vs. limit=5.0 +2023-02-09 03:15:12,911 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.85 vs. limit=5.0 +2023-02-09 03:15:17,341 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 30, batch 1150, loss[loss=0.2269, simple_loss=0.307, pruned_loss=0.07341, over 8495.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2814, pruned_loss=0.05667, over 1609860.40 frames. ], batch size: 26, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:15:35,848 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467_sp0.9 from training. Duration: 27.8166875 +2023-02-09 03:15:39,057 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.48 vs. limit=2.0 +2023-02-09 03:16:03,359 INFO [train.py:901] (2/4) Epoch 30, batch 1200, loss[loss=0.1852, simple_loss=0.2756, pruned_loss=0.04738, over 8366.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2811, pruned_loss=0.05659, over 1608094.69 frames. ], batch size: 24, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:16:11,494 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235615.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:16:12,764 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=235617.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:16:16,060 INFO [optim.py:369] (2/4) 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 03:16:30,036 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235640.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:16:32,802 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2907, 1.9741, 2.4915, 2.1737, 2.4375, 2.3321, 2.2011, 1.3613], + device='cuda:2'), covar=tensor([0.5586, 0.4761, 0.2082, 0.3885, 0.2649, 0.3339, 0.2007, 0.5263], + device='cuda:2'), in_proj_covar=tensor([0.0970, 0.1037, 0.0846, 0.1008, 0.1030, 0.0943, 0.0777, 0.0860], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 03:16:39,493 INFO [train.py:901] (2/4) Epoch 30, batch 1250, loss[loss=0.1604, simple_loss=0.235, pruned_loss=0.04288, over 7697.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2808, pruned_loss=0.05627, over 1610532.47 frames. ], batch size: 18, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:17:06,191 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235691.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:17:07,624 INFO [zipformer.py:1185] (2/4) 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] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235697.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:17:15,249 INFO [train.py:901] (2/4) Epoch 30, batch 1300, loss[loss=0.1984, simple_loss=0.2889, pruned_loss=0.05391, over 8760.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2816, pruned_loss=0.05675, over 1612680.94 frames. ], batch size: 40, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:17:23,823 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235716.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:17:25,222 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=235718.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:17:27,375 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1834, 2.0426, 2.6049, 2.2006, 2.5506, 2.2677, 2.1337, 1.5083], + device='cuda:2'), covar=tensor([0.5954, 0.5255, 0.2164, 0.4224, 0.3003, 0.3409, 0.2027, 0.5616], + device='cuda:2'), in_proj_covar=tensor([0.0964, 0.1030, 0.0841, 0.1002, 0.1025, 0.0939, 0.0773, 0.0855], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 03:17:27,741 INFO [optim.py:369] (2/4) 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 03:17:27,828 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=235722.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 03:17:34,606 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235732.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:17:35,435 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.86 vs. limit=5.0 +2023-02-09 03:17:50,125 INFO [train.py:901] (2/4) Epoch 30, batch 1350, loss[loss=0.2429, simple_loss=0.3205, pruned_loss=0.08272, over 7223.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2823, pruned_loss=0.05685, over 1611054.75 frames. ], batch size: 71, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:18:26,587 INFO [train.py:901] (2/4) Epoch 30, batch 1400, loss[loss=0.1696, simple_loss=0.2575, pruned_loss=0.0408, over 8139.00 frames. ], tot_loss[loss=0.1979, simple_loss=0.2823, pruned_loss=0.05678, over 1617441.96 frames. ], batch size: 22, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:18:30,242 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=235809.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:18:39,090 INFO [optim.py:369] (2/4) 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 03:18:49,550 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=235837.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 03:19:00,841 INFO [train.py:901] (2/4) Epoch 30, batch 1450, loss[loss=0.1892, simple_loss=0.2788, pruned_loss=0.04983, over 8140.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2821, pruned_loss=0.05638, over 1617149.73 frames. ], batch size: 22, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:19:07,580 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0007-9590 from training. Duration: 25.85 +2023-02-09 03:19:22,539 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.61 vs. limit=2.0 +2023-02-09 03:19:38,191 INFO [train.py:901] (2/4) Epoch 30, batch 1500, loss[loss=0.1768, simple_loss=0.2576, pruned_loss=0.04802, over 7428.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.281, pruned_loss=0.0558, over 1618832.60 frames. ], batch size: 17, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:19:42,081 INFO [scaling.py:679] (2/4) 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] (2/4) 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 03:19:56,729 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.85 vs. limit=2.0 +2023-02-09 03:20:14,192 INFO [train.py:901] (2/4) Epoch 30, batch 1550, loss[loss=0.1885, simple_loss=0.2607, pruned_loss=0.05819, over 7711.00 frames. ], tot_loss[loss=0.196, simple_loss=0.2808, pruned_loss=0.05565, over 1617769.88 frames. ], batch size: 18, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:20:17,114 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9308, 1.5600, 3.5272, 1.5419, 2.5537, 3.8861, 4.0392, 3.4170], + device='cuda:2'), covar=tensor([0.1179, 0.1856, 0.0293, 0.2057, 0.0954, 0.0248, 0.0537, 0.0531], + device='cuda:2'), in_proj_covar=tensor([0.0309, 0.0329, 0.0296, 0.0329, 0.0327, 0.0282, 0.0447, 0.0310], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:2') +2023-02-09 03:20:22,056 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.18 vs. limit=2.0 +2023-02-09 03:20:27,554 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5036, 1.4366, 1.8423, 1.1900, 1.1008, 1.8182, 0.2736, 1.2410], + device='cuda:2'), covar=tensor([0.1521, 0.1190, 0.0401, 0.0867, 0.2459, 0.0422, 0.1744, 0.1193], + device='cuda:2'), in_proj_covar=tensor([0.0203, 0.0208, 0.0139, 0.0226, 0.0280, 0.0149, 0.0175, 0.0202], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 03:20:38,890 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=235988.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:20:51,280 INFO [train.py:901] (2/4) Epoch 30, batch 1600, loss[loss=0.2088, simple_loss=0.2876, pruned_loss=0.06501, over 8249.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2814, pruned_loss=0.05587, over 1619048.62 frames. ], batch size: 22, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:20:57,833 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=236013.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:21:00,082 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2166, 3.4815, 2.3324, 3.0910, 2.7760, 2.2146, 2.9342, 3.0191], + device='cuda:2'), covar=tensor([0.1510, 0.0369, 0.1200, 0.0631, 0.0783, 0.1395, 0.0926, 0.1048], + device='cuda:2'), in_proj_covar=tensor([0.0360, 0.0246, 0.0349, 0.0317, 0.0304, 0.0352, 0.0354, 0.0325], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 03:21:04,921 INFO [optim.py:369] (2/4) 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 03:21:15,451 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=236036.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:21:19,098 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=236041.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:21:28,384 INFO [train.py:901] (2/4) Epoch 30, batch 1650, loss[loss=0.2282, simple_loss=0.3093, pruned_loss=0.07356, over 8597.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2806, pruned_loss=0.05598, over 1614026.02 frames. ], batch size: 34, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:21:56,786 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=236093.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 03:22:04,813 INFO [train.py:901] (2/4) Epoch 30, batch 1700, loss[loss=0.1725, simple_loss=0.2419, pruned_loss=0.05153, over 7568.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2806, pruned_loss=0.05607, over 1613027.60 frames. ], batch size: 18, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:22:15,384 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=236118.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 03:22:17,725 INFO [optim.py:369] (2/4) 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 03:22:34,513 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.33 vs. limit=5.0 +2023-02-09 03:22:40,514 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 30, batch 1750, loss[loss=0.1733, simple_loss=0.258, pruned_loss=0.0443, over 8229.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.28, pruned_loss=0.05543, over 1615439.70 frames. ], batch size: 22, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:22:42,663 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=236156.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:23:07,917 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.1510, 1.9023, 2.2620, 2.0194, 2.3132, 2.1897, 2.0755, 1.2356], + device='cuda:2'), covar=tensor([0.5912, 0.5068, 0.2386, 0.4180, 0.2754, 0.3456, 0.2092, 0.5506], + device='cuda:2'), in_proj_covar=tensor([0.0972, 0.1037, 0.0848, 0.1008, 0.1032, 0.0946, 0.0778, 0.0860], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 03:23:16,214 INFO [train.py:901] (2/4) Epoch 30, batch 1800, loss[loss=0.2174, simple_loss=0.3029, pruned_loss=0.06594, over 8551.00 frames. ], tot_loss[loss=0.1951, simple_loss=0.2798, pruned_loss=0.05521, over 1613951.97 frames. ], batch size: 31, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:23:29,387 INFO [optim.py:369] (2/4) 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 03:23:52,508 INFO [train.py:901] (2/4) Epoch 30, batch 1850, loss[loss=0.2291, simple_loss=0.3104, pruned_loss=0.07386, over 7542.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.2805, pruned_loss=0.05547, over 1615894.93 frames. ], batch size: 18, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:24:03,968 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=236268.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:24:28,815 INFO [train.py:901] (2/4) Epoch 30, batch 1900, loss[loss=0.1469, simple_loss=0.2269, pruned_loss=0.03342, over 7427.00 frames. ], tot_loss[loss=0.1943, simple_loss=0.2791, pruned_loss=0.05469, over 1617021.45 frames. ], batch size: 17, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:24:41,169 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 +2023-02-09 03:24:41,424 INFO [optim.py:369] (2/4) 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 03:25:01,733 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0021-44397_sp0.9 from training. Duration: 27.511125 +2023-02-09 03:25:05,202 INFO [train.py:901] (2/4) Epoch 30, batch 1950, loss[loss=0.1899, simple_loss=0.2795, pruned_loss=0.05015, over 8493.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.2802, pruned_loss=0.05541, over 1620047.62 frames. ], batch size: 26, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:25:13,616 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390 from training. Duration: 27.92 +2023-02-09 03:25:24,201 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=236380.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:25:25,042 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0610, 1.2083, 1.1737, 0.8312, 1.1815, 1.0309, 0.1252, 1.2041], + device='cuda:2'), covar=tensor([0.0529, 0.0441, 0.0412, 0.0599, 0.0534, 0.1066, 0.1016, 0.0391], + device='cuda:2'), in_proj_covar=tensor([0.0477, 0.0418, 0.0373, 0.0465, 0.0400, 0.0555, 0.0405, 0.0442], + device='cuda:2'), 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:2') +2023-02-09 03:25:25,064 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4259, 1.6850, 2.1254, 1.3652, 1.5166, 1.6855, 1.5279, 1.5932], + device='cuda:2'), covar=tensor([0.1999, 0.2685, 0.1021, 0.4754, 0.2092, 0.3485, 0.2557, 0.2224], + device='cuda:2'), in_proj_covar=tensor([0.0546, 0.0646, 0.0569, 0.0681, 0.0670, 0.0622, 0.0573, 0.0651], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 03:25:33,317 WARNING [train.py:1067] (2/4) Exclude cut with ID 4964-30587-0040-138716_sp0.9 from training. Duration: 25.0944375 +2023-02-09 03:25:41,055 INFO [train.py:901] (2/4) Epoch 30, batch 2000, loss[loss=0.2323, simple_loss=0.3103, pruned_loss=0.07711, over 8114.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2816, pruned_loss=0.05562, over 1623574.30 frames. ], batch size: 23, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:25:43,360 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2152, 1.9244, 2.3255, 2.0711, 2.3029, 2.2454, 2.1242, 1.1809], + device='cuda:2'), covar=tensor([0.6378, 0.5379, 0.2441, 0.3912, 0.2729, 0.3580, 0.2066, 0.5727], + device='cuda:2'), in_proj_covar=tensor([0.0970, 0.1036, 0.0846, 0.1007, 0.1032, 0.0947, 0.0778, 0.0857], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 03:25:46,887 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=236412.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:25:53,568 INFO [optim.py:369] (2/4) 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 03:26:04,553 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=236437.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:26:16,285 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6398, 1.8170, 1.6144, 2.3147, 0.9514, 1.4437, 1.6856, 1.8634], + device='cuda:2'), covar=tensor([0.0842, 0.0736, 0.0921, 0.0452, 0.1076, 0.1291, 0.0759, 0.0741], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0194, 0.0244, 0.0215, 0.0201, 0.0247, 0.0251, 0.0206], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 03:26:16,759 INFO [train.py:901] (2/4) Epoch 30, batch 2050, loss[loss=0.2227, simple_loss=0.2969, pruned_loss=0.07422, over 7528.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.282, pruned_loss=0.05624, over 1622654.89 frames. ], batch size: 18, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:26:28,372 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2842, 2.1506, 1.6940, 1.9234, 1.7797, 1.4838, 1.6443, 1.6392], + device='cuda:2'), covar=tensor([0.1431, 0.0392, 0.1324, 0.0612, 0.0798, 0.1580, 0.1085, 0.0990], + device='cuda:2'), in_proj_covar=tensor([0.0359, 0.0245, 0.0348, 0.0316, 0.0304, 0.0351, 0.0354, 0.0323], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 03:26:47,590 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=236495.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:26:53,532 INFO [train.py:901] (2/4) Epoch 30, batch 2100, loss[loss=0.211, simple_loss=0.3003, pruned_loss=0.06084, over 8302.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2811, pruned_loss=0.05602, over 1621126.82 frames. ], batch size: 23, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:27:06,242 INFO [optim.py:369] (2/4) 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 03:27:07,829 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=236524.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:27:16,738 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0423, 1.5732, 4.4247, 1.9838, 2.4736, 5.1269, 5.1629, 4.4198], + device='cuda:2'), covar=tensor([0.1332, 0.2009, 0.0302, 0.1938, 0.1316, 0.0181, 0.0473, 0.0556], + device='cuda:2'), in_proj_covar=tensor([0.0310, 0.0330, 0.0298, 0.0329, 0.0329, 0.0282, 0.0449, 0.0312], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:2') +2023-02-09 03:27:25,241 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=236549.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:27:28,497 INFO [train.py:901] (2/4) Epoch 30, batch 2150, loss[loss=0.2412, simple_loss=0.3242, pruned_loss=0.07908, over 8575.00 frames. ], tot_loss[loss=0.1963, simple_loss=0.2808, pruned_loss=0.05589, over 1619958.30 frames. ], batch size: 31, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:27:59,356 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.98 vs. limit=5.0 +2023-02-09 03:28:04,582 INFO [train.py:901] (2/4) Epoch 30, batch 2200, loss[loss=0.2364, simple_loss=0.3243, pruned_loss=0.07427, over 8519.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2811, pruned_loss=0.0558, over 1621640.41 frames. ], batch size: 26, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:28:06,911 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.2732, 4.2042, 3.8956, 2.1331, 3.7632, 3.9178, 3.7924, 3.7368], + device='cuda:2'), covar=tensor([0.0825, 0.0635, 0.1027, 0.4557, 0.0926, 0.1060, 0.1315, 0.0847], + device='cuda:2'), in_proj_covar=tensor([0.0552, 0.0462, 0.0450, 0.0565, 0.0447, 0.0474, 0.0450, 0.0415], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 03:28:18,763 INFO [optim.py:369] (2/4) 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 03:28:40,981 INFO [train.py:901] (2/4) Epoch 30, batch 2250, loss[loss=0.1678, simple_loss=0.2526, pruned_loss=0.0415, over 7830.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2812, pruned_loss=0.05635, over 1621689.23 frames. ], batch size: 20, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:29:04,195 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.62 vs. limit=2.0 +2023-02-09 03:29:16,926 INFO [train.py:901] (2/4) Epoch 30, batch 2300, loss[loss=0.1797, simple_loss=0.2667, pruned_loss=0.04636, over 8332.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2816, pruned_loss=0.05664, over 1620613.01 frames. ], batch size: 26, lr: 2.52e-03, grad_scale: 8.0 +2023-02-09 03:29:29,246 INFO [optim.py:369] (2/4) 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 03:29:50,619 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=236751.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:29:52,585 INFO [train.py:901] (2/4) Epoch 30, batch 2350, loss[loss=0.1859, simple_loss=0.2647, pruned_loss=0.05362, over 7522.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2811, pruned_loss=0.05655, over 1617009.35 frames. ], batch size: 18, lr: 2.51e-03, grad_scale: 8.0 +2023-02-09 03:30:08,609 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 30, batch 2400, loss[loss=0.1993, simple_loss=0.2867, pruned_loss=0.05597, over 8249.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2809, pruned_loss=0.05613, over 1617563.76 frames. ], batch size: 22, lr: 2.51e-03, grad_scale: 8.0 +2023-02-09 03:30:36,998 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7320, 2.5724, 1.9368, 2.3595, 2.2605, 1.6581, 2.1046, 2.2631], + device='cuda:2'), covar=tensor([0.1553, 0.0453, 0.1221, 0.0711, 0.0827, 0.1641, 0.1142, 0.1058], + device='cuda:2'), in_proj_covar=tensor([0.0358, 0.0243, 0.0347, 0.0316, 0.0303, 0.0351, 0.0353, 0.0322], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 03:30:42,227 INFO [optim.py:369] (2/4) 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 03:30:43,211 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.51 vs. limit=2.0 +2023-02-09 03:30:53,422 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.22 vs. limit=2.0 +2023-02-09 03:31:05,203 INFO [train.py:901] (2/4) Epoch 30, batch 2450, loss[loss=0.1607, simple_loss=0.2449, pruned_loss=0.03822, over 7831.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2812, pruned_loss=0.05616, over 1619197.36 frames. ], batch size: 20, lr: 2.51e-03, grad_scale: 8.0 +2023-02-09 03:31:39,710 INFO [train.py:901] (2/4) Epoch 30, batch 2500, loss[loss=0.1908, simple_loss=0.2681, pruned_loss=0.05674, over 7517.00 frames. ], tot_loss[loss=0.1956, simple_loss=0.2799, pruned_loss=0.05561, over 1618774.99 frames. ], batch size: 18, lr: 2.51e-03, grad_scale: 8.0 +2023-02-09 03:31:43,936 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=236910.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 03:31:52,874 INFO [optim.py:369] (2/4) 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 03:32:13,802 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7933, 2.7370, 2.0609, 2.5244, 2.3708, 1.7920, 2.2924, 2.4222], + device='cuda:2'), covar=tensor([0.1511, 0.0405, 0.1187, 0.0665, 0.0773, 0.1595, 0.1031, 0.0896], + device='cuda:2'), in_proj_covar=tensor([0.0356, 0.0242, 0.0345, 0.0314, 0.0302, 0.0350, 0.0351, 0.0320], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 03:32:16,569 INFO [train.py:901] (2/4) Epoch 30, batch 2550, loss[loss=0.176, simple_loss=0.2607, pruned_loss=0.04561, over 7792.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.2802, pruned_loss=0.05544, over 1618742.16 frames. ], batch size: 19, lr: 2.51e-03, grad_scale: 8.0 +2023-02-09 03:32:54,185 INFO [train.py:901] (2/4) Epoch 30, batch 2600, loss[loss=0.1717, simple_loss=0.255, pruned_loss=0.04422, over 7918.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2809, pruned_loss=0.05615, over 1619035.52 frames. ], batch size: 20, lr: 2.51e-03, grad_scale: 8.0 +2023-02-09 03:33:06,912 INFO [optim.py:369] (2/4) 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 03:33:12,328 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.52 vs. limit=5.0 +2023-02-09 03:33:30,300 INFO [train.py:901] (2/4) Epoch 30, batch 2650, loss[loss=0.2105, simple_loss=0.3002, pruned_loss=0.06044, over 8553.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2819, pruned_loss=0.05679, over 1615766.63 frames. ], batch size: 34, lr: 2.51e-03, grad_scale: 8.0 +2023-02-09 03:34:06,433 INFO [train.py:901] (2/4) Epoch 30, batch 2700, loss[loss=0.2068, simple_loss=0.2851, pruned_loss=0.06423, over 7931.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2817, pruned_loss=0.05636, over 1616800.22 frames. ], batch size: 20, lr: 2.51e-03, grad_scale: 8.0 +2023-02-09 03:34:07,584 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.15 vs. limit=2.0 +2023-02-09 03:34:18,966 INFO [optim.py:369] (2/4) 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 03:34:41,480 INFO [train.py:901] (2/4) Epoch 30, batch 2750, loss[loss=0.2168, simple_loss=0.3009, pruned_loss=0.06632, over 8551.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2813, pruned_loss=0.05605, over 1615442.54 frames. ], batch size: 49, lr: 2.51e-03, grad_scale: 8.0 +2023-02-09 03:35:18,235 INFO [train.py:901] (2/4) Epoch 30, batch 2800, loss[loss=0.1849, simple_loss=0.2686, pruned_loss=0.05062, over 8031.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.281, pruned_loss=0.05587, over 1618231.56 frames. ], batch size: 22, lr: 2.51e-03, grad_scale: 8.0 +2023-02-09 03:35:20,499 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7468, 1.5123, 1.8509, 1.4533, 0.9612, 1.5740, 1.6468, 1.4598], + device='cuda:2'), covar=tensor([0.0574, 0.1255, 0.1547, 0.1441, 0.0612, 0.1474, 0.0712, 0.0681], + device='cuda:2'), in_proj_covar=tensor([0.0101, 0.0155, 0.0192, 0.0163, 0.0102, 0.0165, 0.0114, 0.0149], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0008, 0.0008], + device='cuda:2') +2023-02-09 03:35:31,346 INFO [optim.py:369] (2/4) 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 03:35:45,298 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.1475, 1.3972, 4.3426, 1.6188, 3.8511, 3.6307, 3.9188, 3.8331], + device='cuda:2'), covar=tensor([0.0633, 0.4528, 0.0564, 0.4245, 0.1009, 0.0921, 0.0627, 0.0679], + device='cuda:2'), in_proj_covar=tensor([0.0694, 0.0671, 0.0752, 0.0670, 0.0756, 0.0645, 0.0655, 0.0729], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 03:35:48,281 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.19 vs. limit=2.0 +2023-02-09 03:35:53,009 INFO [train.py:901] (2/4) Epoch 30, batch 2850, loss[loss=0.2307, simple_loss=0.3039, pruned_loss=0.07877, over 7979.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2813, pruned_loss=0.05606, over 1620512.21 frames. ], batch size: 21, lr: 2.51e-03, grad_scale: 8.0 +2023-02-09 03:35:53,088 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=237254.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 03:36:18,915 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-09 03:36:27,599 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.63 vs. limit=2.0 +2023-02-09 03:36:29,225 INFO [train.py:901] (2/4) Epoch 30, batch 2900, loss[loss=0.1829, simple_loss=0.2728, pruned_loss=0.04649, over 8081.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.2803, pruned_loss=0.05562, over 1617646.14 frames. ], batch size: 21, lr: 2.51e-03, grad_scale: 8.0 +2023-02-09 03:36:42,591 INFO [optim.py:369] (2/4) 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 03:37:04,021 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp1.1 from training. Duration: 0.7545625 +2023-02-09 03:37:05,367 INFO [train.py:901] (2/4) Epoch 30, batch 2950, loss[loss=0.1903, simple_loss=0.2709, pruned_loss=0.05486, over 7925.00 frames. ], tot_loss[loss=0.1949, simple_loss=0.2792, pruned_loss=0.05531, over 1612466.39 frames. ], batch size: 20, lr: 2.51e-03, grad_scale: 8.0 +2023-02-09 03:37:15,789 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=237369.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 03:37:36,195 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.8185, 3.8100, 3.4573, 1.6586, 3.3569, 3.4980, 3.3635, 3.4070], + device='cuda:2'), covar=tensor([0.0914, 0.0650, 0.1088, 0.5062, 0.0990, 0.1042, 0.1379, 0.0903], + device='cuda:2'), in_proj_covar=tensor([0.0550, 0.0461, 0.0450, 0.0562, 0.0445, 0.0473, 0.0447, 0.0415], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 03:37:40,336 INFO [train.py:901] (2/4) Epoch 30, batch 3000, loss[loss=0.191, simple_loss=0.2846, pruned_loss=0.04875, over 8036.00 frames. ], tot_loss[loss=0.1953, simple_loss=0.2799, pruned_loss=0.05535, over 1614902.12 frames. ], batch size: 22, lr: 2.51e-03, grad_scale: 8.0 +2023-02-09 03:37:40,336 INFO [train.py:926] (2/4) Computing validation loss +2023-02-09 03:37:54,062 INFO [train.py:935] (2/4) Epoch 30, validation: loss=0.1704, simple_loss=0.2697, pruned_loss=0.0356, over 944034.00 frames. +2023-02-09 03:37:54,063 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6724MB +2023-02-09 03:38:07,361 INFO [optim.py:369] (2/4) 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 03:38:31,173 INFO [train.py:901] (2/4) Epoch 30, batch 3050, loss[loss=0.2106, simple_loss=0.2963, pruned_loss=0.06245, over 8519.00 frames. ], tot_loss[loss=0.1965, simple_loss=0.2807, pruned_loss=0.05611, over 1612626.28 frames. ], batch size: 31, lr: 2.51e-03, grad_scale: 8.0 +2023-02-09 03:39:01,960 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6649, 1.8697, 1.9584, 1.3810, 2.0666, 1.5081, 0.6072, 1.9567], + device='cuda:2'), covar=tensor([0.0666, 0.0436, 0.0362, 0.0665, 0.0489, 0.1037, 0.1101, 0.0324], + device='cuda:2'), in_proj_covar=tensor([0.0483, 0.0422, 0.0376, 0.0467, 0.0403, 0.0562, 0.0409, 0.0446], + device='cuda:2'), out_proj_covar=tensor([1.2778e-04, 1.0912e-04, 9.7950e-05, 1.2182e-04, 1.0531e-04, 1.5642e-04, + 1.0891e-04, 1.1676e-04], device='cuda:2') +2023-02-09 03:39:07,127 INFO [train.py:901] (2/4) Epoch 30, batch 3100, loss[loss=0.1951, simple_loss=0.2779, pruned_loss=0.05611, over 7813.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2812, pruned_loss=0.05674, over 1614461.33 frames. ], batch size: 20, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:39:10,845 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=237509.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:39:14,041 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.34 vs. limit=2.0 +2023-02-09 03:39:19,646 INFO [optim.py:369] (2/4) 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 03:39:20,785 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.23 vs. limit=2.0 +2023-02-09 03:39:43,976 INFO [train.py:901] (2/4) Epoch 30, batch 3150, loss[loss=0.1786, simple_loss=0.2715, pruned_loss=0.04289, over 8029.00 frames. ], tot_loss[loss=0.198, simple_loss=0.2814, pruned_loss=0.05733, over 1610250.52 frames. ], batch size: 22, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:40:03,664 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=237581.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 03:40:21,026 INFO [train.py:901] (2/4) Epoch 30, batch 3200, loss[loss=0.1772, simple_loss=0.2667, pruned_loss=0.04389, over 7971.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2808, pruned_loss=0.05727, over 1608555.14 frames. ], batch size: 21, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:40:33,294 INFO [optim.py:369] (2/4) 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 03:40:35,496 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=237625.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 03:40:52,664 INFO [zipformer.py:1185] (2/4) 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] (2/4) Epoch 30, batch 3250, loss[loss=0.1684, simple_loss=0.2611, pruned_loss=0.03785, over 7805.00 frames. ], tot_loss[loss=0.1992, simple_loss=0.2824, pruned_loss=0.05797, over 1609218.81 frames. ], batch size: 20, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:41:32,176 INFO [train.py:901] (2/4) Epoch 30, batch 3300, loss[loss=0.1805, simple_loss=0.2716, pruned_loss=0.04472, over 7818.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2815, pruned_loss=0.05746, over 1606991.77 frames. ], batch size: 20, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:41:44,069 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.48 vs. limit=5.0 +2023-02-09 03:41:45,795 INFO [optim.py:369] (2/4) 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 03:42:07,977 INFO [train.py:901] (2/4) Epoch 30, batch 3350, loss[loss=0.2113, simple_loss=0.2925, pruned_loss=0.06504, over 8536.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2807, pruned_loss=0.05693, over 1609529.86 frames. ], batch size: 50, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:42:44,184 INFO [train.py:901] (2/4) Epoch 30, batch 3400, loss[loss=0.1783, simple_loss=0.2655, pruned_loss=0.04553, over 8139.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2806, pruned_loss=0.05712, over 1606524.97 frames. ], batch size: 22, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:42:47,123 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=237808.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:42:57,410 INFO [optim.py:369] (2/4) 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 03:43:00,484 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2864, 2.5878, 2.6751, 1.7202, 3.0628, 1.8392, 1.5228, 2.4088], + device='cuda:2'), covar=tensor([0.1012, 0.0472, 0.0443, 0.0941, 0.0593, 0.1034, 0.1214, 0.0588], + device='cuda:2'), in_proj_covar=tensor([0.0486, 0.0425, 0.0380, 0.0470, 0.0406, 0.0566, 0.0410, 0.0449], + device='cuda:2'), out_proj_covar=tensor([1.2850e-04, 1.0992e-04, 9.8809e-05, 1.2263e-04, 1.0614e-04, 1.5748e-04, + 1.0928e-04, 1.1749e-04], device='cuda:2') +2023-02-09 03:43:19,449 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=237853.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:43:20,120 INFO [train.py:901] (2/4) Epoch 30, batch 3450, loss[loss=0.1508, simple_loss=0.2338, pruned_loss=0.0339, over 7686.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2816, pruned_loss=0.05764, over 1606260.79 frames. ], batch size: 18, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:43:56,090 INFO [train.py:901] (2/4) Epoch 30, batch 3500, loss[loss=0.1994, simple_loss=0.2726, pruned_loss=0.06309, over 8086.00 frames. ], tot_loss[loss=0.1996, simple_loss=0.2827, pruned_loss=0.05825, over 1607807.20 frames. ], batch size: 21, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:44:08,739 INFO [optim.py:369] (2/4) 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 03:44:11,004 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=237925.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 03:44:11,726 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9670, 1.5323, 3.4549, 1.6541, 2.5111, 3.8373, 3.9103, 3.3027], + device='cuda:2'), covar=tensor([0.1189, 0.1874, 0.0311, 0.1967, 0.0985, 0.0214, 0.0499, 0.0502], + device='cuda:2'), in_proj_covar=tensor([0.0308, 0.0327, 0.0296, 0.0326, 0.0328, 0.0281, 0.0447, 0.0308], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:2') +2023-02-09 03:44:14,957 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=237930.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:44:22,349 WARNING [train.py:1067] (2/4) Exclude cut with ID 453-131332-0000-131866_sp0.9 from training. Duration: 25.3333125 +2023-02-09 03:44:32,808 INFO [train.py:901] (2/4) Epoch 30, batch 3550, loss[loss=0.1882, simple_loss=0.2773, pruned_loss=0.04957, over 8240.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.2813, pruned_loss=0.05802, over 1609645.68 frames. ], batch size: 24, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:44:43,165 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=237968.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:44:49,853 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.20 vs. limit=2.0 +2023-02-09 03:45:09,426 INFO [train.py:901] (2/4) Epoch 30, batch 3600, loss[loss=0.1924, simple_loss=0.281, pruned_loss=0.05196, over 8467.00 frames. ], tot_loss[loss=0.1972, simple_loss=0.2802, pruned_loss=0.05711, over 1608158.18 frames. ], batch size: 27, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:45:22,452 INFO [optim.py:369] (2/4) 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 03:45:35,416 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238040.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 03:45:40,892 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.9340, 3.8232, 3.5632, 1.7145, 3.4635, 3.5312, 3.4380, 3.3734], + device='cuda:2'), covar=tensor([0.0777, 0.0605, 0.0986, 0.4463, 0.0902, 0.1083, 0.1413, 0.0891], + device='cuda:2'), in_proj_covar=tensor([0.0551, 0.0463, 0.0452, 0.0563, 0.0445, 0.0475, 0.0451, 0.0416], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 03:45:44,902 INFO [train.py:901] (2/4) Epoch 30, batch 3650, loss[loss=0.1602, simple_loss=0.2475, pruned_loss=0.03643, over 7660.00 frames. ], tot_loss[loss=0.196, simple_loss=0.2795, pruned_loss=0.05625, over 1611665.53 frames. ], batch size: 19, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:46:05,855 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238082.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:46:12,778 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238092.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:46:20,739 INFO [train.py:901] (2/4) Epoch 30, batch 3700, loss[loss=0.2425, simple_loss=0.3271, pruned_loss=0.07891, over 8104.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2805, pruned_loss=0.05705, over 1612186.26 frames. ], batch size: 23, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:46:29,080 WARNING [train.py:1067] (2/4) Exclude cut with ID 2411-132532-0017-25057_sp1.1 from training. Duration: 0.9681875 +2023-02-09 03:46:33,265 INFO [optim.py:369] (2/4) 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 03:46:50,317 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238144.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:46:54,656 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6336, 1.2941, 2.8964, 1.4832, 2.3694, 3.1337, 3.2624, 2.6822], + device='cuda:2'), covar=tensor([0.1256, 0.1877, 0.0350, 0.2081, 0.0811, 0.0294, 0.0640, 0.0582], + device='cuda:2'), in_proj_covar=tensor([0.0310, 0.0330, 0.0298, 0.0328, 0.0331, 0.0283, 0.0451, 0.0310], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:2') +2023-02-09 03:46:56,063 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=238152.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:46:57,361 INFO [train.py:901] (2/4) Epoch 30, batch 3750, loss[loss=0.2192, simple_loss=0.2997, pruned_loss=0.06935, over 8300.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2803, pruned_loss=0.05669, over 1611559.09 frames. ], batch size: 23, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:47:19,947 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5519, 1.8596, 2.7633, 1.5385, 2.1472, 1.9644, 1.6378, 2.0481], + device='cuda:2'), covar=tensor([0.2053, 0.2734, 0.0939, 0.4969, 0.1873, 0.3563, 0.2636, 0.2218], + device='cuda:2'), in_proj_covar=tensor([0.0547, 0.0644, 0.0569, 0.0679, 0.0671, 0.0620, 0.0573, 0.0649], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 03:47:33,676 INFO [train.py:901] (2/4) Epoch 30, batch 3800, loss[loss=0.2045, simple_loss=0.2967, pruned_loss=0.05617, over 8463.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2806, pruned_loss=0.05657, over 1611613.08 frames. ], batch size: 48, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:47:46,035 INFO [optim.py:369] (2/4) 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 03:47:47,580 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238224.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:47:57,379 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([0.9782, 0.9158, 1.6216, 0.9423, 1.6046, 1.7914, 1.8485, 1.4981], + device='cuda:2'), covar=tensor([0.1051, 0.1305, 0.0596, 0.1733, 0.1224, 0.0393, 0.0753, 0.0609], + device='cuda:2'), in_proj_covar=tensor([0.0308, 0.0328, 0.0297, 0.0327, 0.0329, 0.0281, 0.0449, 0.0309], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:2') +2023-02-09 03:48:05,480 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238249.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:48:09,471 INFO [train.py:901] (2/4) Epoch 30, batch 3850, loss[loss=0.1829, simple_loss=0.278, pruned_loss=0.04383, over 8393.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2802, pruned_loss=0.05671, over 1602689.89 frames. ], batch size: 49, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:48:18,889 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238267.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:48:23,698 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=238274.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:48:38,379 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585_sp1.1 from training. Duration: 0.836375 +2023-02-09 03:48:39,907 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238296.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 03:48:45,173 INFO [train.py:901] (2/4) Epoch 30, batch 3900, loss[loss=0.1939, simple_loss=0.2801, pruned_loss=0.05384, over 8245.00 frames. ], tot_loss[loss=0.1944, simple_loss=0.2781, pruned_loss=0.05532, over 1602968.21 frames. ], batch size: 22, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:48:57,793 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238321.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 03:48:58,284 INFO [optim.py:369] (2/4) 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 03:49:18,556 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7460, 1.6131, 2.2819, 1.4960, 1.3071, 2.2296, 0.5296, 1.3958], + device='cuda:2'), covar=tensor([0.1384, 0.1218, 0.0298, 0.0949, 0.2261, 0.0361, 0.1644, 0.1153], + device='cuda:2'), in_proj_covar=tensor([0.0204, 0.0206, 0.0139, 0.0225, 0.0279, 0.0149, 0.0176, 0.0200], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 03:49:20,478 INFO [train.py:901] (2/4) Epoch 30, batch 3950, loss[loss=0.2101, simple_loss=0.2867, pruned_loss=0.06678, over 8035.00 frames. ], tot_loss[loss=0.1952, simple_loss=0.2791, pruned_loss=0.05566, over 1609160.97 frames. ], batch size: 22, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:49:46,425 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238389.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:49:56,962 INFO [train.py:901] (2/4) Epoch 30, batch 4000, loss[loss=0.1873, simple_loss=0.2852, pruned_loss=0.04468, over 8271.00 frames. ], tot_loss[loss=0.1948, simple_loss=0.2791, pruned_loss=0.05523, over 1609212.98 frames. ], batch size: 24, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:50:01,961 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5804, 1.6579, 2.0368, 1.6892, 1.1802, 1.7182, 2.2405, 2.1119], + device='cuda:2'), covar=tensor([0.0538, 0.1271, 0.1619, 0.1405, 0.0602, 0.1414, 0.0652, 0.0595], + device='cuda:2'), in_proj_covar=tensor([0.0101, 0.0155, 0.0192, 0.0162, 0.0102, 0.0165, 0.0114, 0.0148], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0008, 0.0008], + device='cuda:2') +2023-02-09 03:50:09,951 INFO [optim.py:369] (2/4) 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 03:50:12,701 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=238426.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:50:20,359 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=238436.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:50:21,790 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238438.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 03:50:28,648 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.0360, 2.2901, 3.8171, 2.0302, 1.9341, 3.7657, 0.7991, 2.3087], + device='cuda:2'), covar=tensor([0.1259, 0.1108, 0.0192, 0.1426, 0.2283, 0.0273, 0.1907, 0.1159], + device='cuda:2'), in_proj_covar=tensor([0.0204, 0.0206, 0.0139, 0.0225, 0.0279, 0.0149, 0.0175, 0.0201], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 03:50:32,677 INFO [train.py:901] (2/4) Epoch 30, batch 4050, loss[loss=0.1912, simple_loss=0.2878, pruned_loss=0.04732, over 8489.00 frames. ], tot_loss[loss=0.1948, simple_loss=0.279, pruned_loss=0.05532, over 1605800.52 frames. ], batch size: 27, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:50:38,558 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.0013, 2.2473, 1.8348, 2.7174, 1.3640, 1.5900, 2.1126, 2.1848], + device='cuda:2'), covar=tensor([0.0680, 0.0628, 0.0814, 0.0330, 0.1005, 0.1286, 0.0747, 0.0657], + device='cuda:2'), in_proj_covar=tensor([0.0230, 0.0193, 0.0243, 0.0214, 0.0201, 0.0245, 0.0248, 0.0203], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 03:50:57,400 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=238488.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:51:09,340 INFO [train.py:901] (2/4) Epoch 30, batch 4100, loss[loss=0.2144, simple_loss=0.297, pruned_loss=0.0659, over 8237.00 frames. ], tot_loss[loss=0.196, simple_loss=0.2801, pruned_loss=0.05597, over 1607209.14 frames. ], batch size: 22, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:51:21,790 INFO [optim.py:369] (2/4) 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 03:51:22,756 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238523.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:51:35,961 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238541.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:51:41,372 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238548.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:51:43,278 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238551.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:51:45,010 INFO [train.py:901] (2/4) Epoch 30, batch 4150, loss[loss=0.2086, simple_loss=0.3009, pruned_loss=0.05812, over 8286.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2809, pruned_loss=0.05626, over 1607798.01 frames. ], batch size: 23, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:51:57,581 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6289, 2.0031, 2.9238, 1.5172, 2.2217, 2.0793, 1.7027, 2.2670], + device='cuda:2'), covar=tensor([0.1947, 0.2711, 0.1055, 0.4821, 0.2099, 0.3254, 0.2509, 0.2530], + device='cuda:2'), in_proj_covar=tensor([0.0548, 0.0646, 0.0571, 0.0682, 0.0675, 0.0622, 0.0574, 0.0653], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 03:52:11,579 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2023-02-09 03:52:19,895 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238603.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:52:20,452 INFO [train.py:901] (2/4) Epoch 30, batch 4200, loss[loss=0.2051, simple_loss=0.2974, pruned_loss=0.05644, over 8444.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2824, pruned_loss=0.05714, over 1609722.30 frames. ], batch size: 49, lr: 2.51e-03, grad_scale: 16.0 +2023-02-09 03:52:30,416 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2276, 1.5484, 4.4984, 1.9136, 2.5380, 5.1466, 5.1927, 4.4949], + device='cuda:2'), covar=tensor([0.1194, 0.1965, 0.0233, 0.2112, 0.1127, 0.0149, 0.0358, 0.0471], + device='cuda:2'), in_proj_covar=tensor([0.0309, 0.0329, 0.0296, 0.0328, 0.0329, 0.0282, 0.0449, 0.0309], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:2') +2023-02-09 03:52:33,715 INFO [optim.py:369] (2/4) 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 03:52:41,502 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0005-9467 from training. Duration: 25.035 +2023-02-09 03:52:50,647 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238645.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:52:56,590 INFO [train.py:901] (2/4) Epoch 30, batch 4250, loss[loss=0.2107, simple_loss=0.2998, pruned_loss=0.06082, over 8291.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.2814, pruned_loss=0.05686, over 1605550.30 frames. ], batch size: 23, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 03:53:02,696 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5432, 1.4310, 1.8233, 1.1526, 1.1676, 1.7811, 0.2143, 1.1898], + device='cuda:2'), covar=tensor([0.1229, 0.0998, 0.0355, 0.0773, 0.2038, 0.0397, 0.1594, 0.0964], + device='cuda:2'), in_proj_covar=tensor([0.0204, 0.0206, 0.0139, 0.0224, 0.0279, 0.0149, 0.0176, 0.0201], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 03:53:05,091 WARNING [train.py:1067] (2/4) Exclude cut with ID 1914-133440-0024-53073_sp0.9 from training. Duration: 25.2444375 +2023-02-09 03:53:07,989 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238670.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:53:27,928 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238699.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:53:31,128 INFO [train.py:901] (2/4) Epoch 30, batch 4300, loss[loss=0.2408, simple_loss=0.3189, pruned_loss=0.08137, over 8226.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2811, pruned_loss=0.05659, over 1604939.36 frames. ], batch size: 22, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 03:53:44,805 INFO [optim.py:369] (2/4) 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 03:54:04,291 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2507, 3.0425, 2.3555, 2.7233, 2.6371, 2.2339, 2.4849, 2.8014], + device='cuda:2'), covar=tensor([0.1070, 0.0302, 0.0829, 0.0454, 0.0519, 0.1049, 0.0689, 0.0738], + device='cuda:2'), in_proj_covar=tensor([0.0356, 0.0243, 0.0345, 0.0314, 0.0301, 0.0348, 0.0349, 0.0318], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 03:54:06,887 INFO [train.py:901] (2/4) Epoch 30, batch 4350, loss[loss=0.2115, simple_loss=0.2938, pruned_loss=0.06462, over 8096.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.2808, pruned_loss=0.05651, over 1604844.26 frames. ], batch size: 23, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 03:54:27,361 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=238782.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 03:54:36,429 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425_sp0.9 from training. Duration: 28.638875 +2023-02-09 03:54:37,980 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238797.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:54:42,562 INFO [train.py:901] (2/4) Epoch 30, batch 4400, loss[loss=0.1949, simple_loss=0.2732, pruned_loss=0.0583, over 7942.00 frames. ], tot_loss[loss=0.1961, simple_loss=0.2798, pruned_loss=0.05618, over 1604675.51 frames. ], batch size: 20, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 03:54:44,443 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.91 vs. limit=5.0 +2023-02-09 03:54:44,937 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238807.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:54:53,640 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.87 vs. limit=2.0 +2023-02-09 03:54:55,879 INFO [optim.py:369] (2/4) 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 03:54:56,113 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238822.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:55:03,788 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238832.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:55:15,271 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp0.9 from training. Duration: 33.038875 +2023-02-09 03:55:18,601 INFO [train.py:901] (2/4) Epoch 30, batch 4450, loss[loss=0.222, simple_loss=0.3003, pruned_loss=0.07182, over 8525.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2817, pruned_loss=0.05695, over 1609558.37 frames. ], batch size: 28, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 03:55:22,515 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=238859.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:55:40,874 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=238884.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:55:50,462 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=238897.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 03:55:55,019 INFO [train.py:901] (2/4) Epoch 30, batch 4500, loss[loss=0.164, simple_loss=0.2482, pruned_loss=0.03987, over 7531.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2809, pruned_loss=0.05653, over 1611851.87 frames. ], batch size: 18, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 03:56:07,440 INFO [optim.py:369] (2/4) 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 03:56:08,185 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983 from training. Duration: 0.83 +2023-02-09 03:56:12,490 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5011, 1.3781, 1.7181, 1.2967, 0.8903, 1.4219, 1.4542, 1.3923], + device='cuda:2'), covar=tensor([0.0611, 0.1273, 0.1591, 0.1467, 0.0582, 0.1447, 0.0756, 0.0675], + device='cuda:2'), in_proj_covar=tensor([0.0101, 0.0155, 0.0192, 0.0163, 0.0102, 0.0165, 0.0114, 0.0148], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0008, 0.0008], + device='cuda:2') +2023-02-09 03:56:29,263 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.53 vs. limit=2.0 +2023-02-09 03:56:30,983 INFO [train.py:901] (2/4) Epoch 30, batch 4550, loss[loss=0.2023, simple_loss=0.3013, pruned_loss=0.05164, over 8460.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2813, pruned_loss=0.05703, over 1613681.36 frames. ], batch size: 49, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 03:56:31,139 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=238954.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 03:57:06,050 INFO [train.py:901] (2/4) Epoch 30, batch 4600, loss[loss=0.2478, simple_loss=0.3189, pruned_loss=0.08831, over 6429.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2813, pruned_loss=0.05698, over 1611304.22 frames. ], batch size: 72, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 03:57:19,158 INFO [optim.py:369] (2/4) 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 03:57:26,210 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.6960, 5.8614, 5.1091, 2.5732, 5.0925, 5.4613, 5.3149, 5.2603], + device='cuda:2'), covar=tensor([0.0426, 0.0264, 0.0708, 0.4016, 0.0704, 0.0749, 0.0969, 0.0529], + device='cuda:2'), in_proj_covar=tensor([0.0556, 0.0466, 0.0458, 0.0571, 0.0451, 0.0481, 0.0455, 0.0420], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 03:57:34,217 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=239043.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:57:40,144 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.92 vs. limit=5.0 +2023-02-09 03:57:41,725 INFO [train.py:901] (2/4) Epoch 30, batch 4650, loss[loss=0.2051, simple_loss=0.2817, pruned_loss=0.06428, over 8301.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2822, pruned_loss=0.05718, over 1617442.91 frames. ], batch size: 23, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 03:58:17,765 INFO [train.py:901] (2/4) Epoch 30, batch 4700, loss[loss=0.2065, simple_loss=0.2936, pruned_loss=0.05972, over 8334.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2817, pruned_loss=0.0574, over 1613490.41 frames. ], batch size: 25, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 03:58:25,196 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.0687, 1.3025, 1.2180, 0.7359, 1.2374, 1.0539, 0.1050, 1.2411], + device='cuda:2'), covar=tensor([0.0527, 0.0430, 0.0404, 0.0660, 0.0458, 0.1010, 0.0943, 0.0377], + device='cuda:2'), in_proj_covar=tensor([0.0481, 0.0420, 0.0378, 0.0467, 0.0402, 0.0560, 0.0407, 0.0449], + device='cuda:2'), out_proj_covar=tensor([1.2733e-04, 1.0839e-04, 9.8362e-05, 1.2200e-04, 1.0492e-04, 1.5570e-04, + 1.0844e-04, 1.1746e-04], device='cuda:2') +2023-02-09 03:58:30,952 INFO [optim.py:369] (2/4) 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 03:58:52,478 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=239153.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 03:58:52,925 INFO [train.py:901] (2/4) Epoch 30, batch 4750, loss[loss=0.1899, simple_loss=0.2821, pruned_loss=0.0489, over 8107.00 frames. ], tot_loss[loss=0.1994, simple_loss=0.283, pruned_loss=0.05791, over 1615109.88 frames. ], batch size: 23, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 03:58:55,954 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=239158.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 03:59:05,580 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.37 vs. limit=5.0 +2023-02-09 03:59:10,883 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=239178.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 03:59:12,054 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0003-9465 from training. Duration: 26.8349375 +2023-02-09 03:59:14,171 WARNING [train.py:1067] (2/4) Exclude cut with ID 5622-44585-0006-50425 from training. Duration: 25.775 +2023-02-09 03:59:21,387 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.76 vs. limit=5.0 +2023-02-09 03:59:28,648 INFO [train.py:901] (2/4) Epoch 30, batch 4800, loss[loss=0.2328, simple_loss=0.3013, pruned_loss=0.08218, over 8326.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2821, pruned_loss=0.05741, over 1615650.97 frames. ], batch size: 25, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 03:59:35,690 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7833, 3.1195, 2.5954, 4.1539, 1.7939, 2.2682, 3.0009, 2.9279], + device='cuda:2'), covar=tensor([0.0619, 0.0694, 0.0701, 0.0196, 0.1017, 0.1147, 0.0687, 0.0758], + device='cuda:2'), in_proj_covar=tensor([0.0233, 0.0195, 0.0245, 0.0215, 0.0203, 0.0247, 0.0250, 0.0205], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 03:59:41,698 INFO [optim.py:369] (2/4) 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 03:59:46,062 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7580, 2.1857, 3.5283, 1.7063, 1.7201, 3.4250, 0.8322, 2.1228], + device='cuda:2'), covar=tensor([0.1227, 0.1005, 0.0178, 0.1368, 0.2242, 0.0230, 0.1781, 0.1114], + device='cuda:2'), in_proj_covar=tensor([0.0203, 0.0205, 0.0138, 0.0223, 0.0276, 0.0148, 0.0174, 0.0200], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 04:00:04,581 INFO [train.py:901] (2/4) Epoch 30, batch 4850, loss[loss=0.1813, simple_loss=0.2751, pruned_loss=0.04377, over 8613.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.2811, pruned_loss=0.05671, over 1614597.28 frames. ], batch size: 34, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 04:00:06,729 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914 from training. Duration: 26.205 +2023-02-09 04:00:36,496 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=239298.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 04:00:40,436 INFO [train.py:901] (2/4) Epoch 30, batch 4900, loss[loss=0.1805, simple_loss=0.2782, pruned_loss=0.0414, over 8482.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2815, pruned_loss=0.05667, over 1617488.67 frames. ], batch size: 25, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 04:00:53,055 INFO [optim.py:369] (2/4) 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 04:01:15,934 INFO [train.py:901] (2/4) Epoch 30, batch 4950, loss[loss=0.217, simple_loss=0.3046, pruned_loss=0.06467, over 8194.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2804, pruned_loss=0.05639, over 1617013.18 frames. ], batch size: 23, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 04:01:43,675 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7239, 4.7514, 4.3081, 2.1725, 4.1832, 4.3844, 4.3104, 4.2338], + device='cuda:2'), covar=tensor([0.0660, 0.0502, 0.0928, 0.4391, 0.0918, 0.0867, 0.1173, 0.0697], + device='cuda:2'), in_proj_covar=tensor([0.0556, 0.0465, 0.0455, 0.0569, 0.0451, 0.0480, 0.0453, 0.0418], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 04:01:51,704 INFO [train.py:901] (2/4) Epoch 30, batch 5000, loss[loss=0.2193, simple_loss=0.2993, pruned_loss=0.06964, over 8279.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.281, pruned_loss=0.05678, over 1615245.45 frames. ], batch size: 49, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 04:01:58,093 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=239413.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 04:01:58,912 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=239414.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:02:04,998 INFO [optim.py:369] (2/4) 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 04:02:17,682 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=239439.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:02:29,042 INFO [train.py:901] (2/4) Epoch 30, batch 5050, loss[loss=0.1657, simple_loss=0.2446, pruned_loss=0.04344, over 7558.00 frames. ], tot_loss[loss=0.1978, simple_loss=0.2811, pruned_loss=0.05728, over 1614132.39 frames. ], batch size: 18, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 04:02:52,719 WARNING [train.py:1067] (2/4) Exclude cut with ID 5239-32139-0047-92994 from training. Duration: 27.14 +2023-02-09 04:03:05,847 INFO [train.py:901] (2/4) Epoch 30, batch 5100, loss[loss=0.1978, simple_loss=0.2826, pruned_loss=0.05651, over 8197.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2808, pruned_loss=0.05724, over 1610906.22 frames. ], batch size: 23, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 04:03:08,876 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8284, 1.6268, 2.4107, 1.4265, 1.3905, 2.3345, 0.4040, 1.4406], + device='cuda:2'), covar=tensor([0.1358, 0.1253, 0.0290, 0.1183, 0.2337, 0.0479, 0.1970, 0.1297], + device='cuda:2'), in_proj_covar=tensor([0.0203, 0.0207, 0.0138, 0.0225, 0.0279, 0.0149, 0.0175, 0.0201], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 04:03:20,027 INFO [optim.py:369] (2/4) 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 04:03:23,165 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.2045, 1.0765, 1.3037, 0.9594, 0.9758, 1.3007, 0.0730, 0.8966], + device='cuda:2'), covar=tensor([0.1393, 0.1241, 0.0533, 0.0652, 0.2171, 0.0583, 0.1865, 0.1158], + device='cuda:2'), in_proj_covar=tensor([0.0203, 0.0207, 0.0138, 0.0225, 0.0279, 0.0149, 0.0175, 0.0202], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 04:03:24,059 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.69 vs. limit=5.0 +2023-02-09 04:03:42,212 INFO [train.py:901] (2/4) Epoch 30, batch 5150, loss[loss=0.2139, simple_loss=0.2999, pruned_loss=0.06399, over 8098.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2819, pruned_loss=0.05787, over 1607862.76 frames. ], batch size: 23, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 04:03:44,421 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7109, 1.6206, 2.3126, 1.5186, 1.4124, 2.2223, 0.4094, 1.4359], + device='cuda:2'), covar=tensor([0.1487, 0.1105, 0.0330, 0.0989, 0.2202, 0.0393, 0.1789, 0.1216], + device='cuda:2'), in_proj_covar=tensor([0.0204, 0.0207, 0.0139, 0.0225, 0.0279, 0.0149, 0.0175, 0.0202], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 04:03:47,836 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6023, 2.4914, 1.9174, 2.3192, 2.1394, 1.6458, 2.0959, 2.0516], + device='cuda:2'), covar=tensor([0.1511, 0.0464, 0.1275, 0.0622, 0.0728, 0.1622, 0.0978, 0.1048], + device='cuda:2'), in_proj_covar=tensor([0.0361, 0.0247, 0.0348, 0.0317, 0.0304, 0.0350, 0.0352, 0.0323], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 04:03:49,473 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2023-02-09 04:04:18,743 INFO [train.py:901] (2/4) Epoch 30, batch 5200, loss[loss=0.2027, simple_loss=0.2896, pruned_loss=0.0579, over 8477.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2811, pruned_loss=0.05702, over 1611224.30 frames. ], batch size: 25, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 04:04:31,922 INFO [optim.py:369] (2/4) 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] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=239640.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:04:54,717 INFO [train.py:901] (2/4) Epoch 30, batch 5250, loss[loss=0.2159, simple_loss=0.2984, pruned_loss=0.06668, over 8289.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2812, pruned_loss=0.05685, over 1613065.99 frames. ], batch size: 23, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 04:04:56,126 WARNING [train.py:1067] (2/4) Exclude cut with ID 7859-102521-0017-21930_sp0.9 from training. Duration: 27.25 +2023-02-09 04:05:05,221 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=239669.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 04:05:23,272 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=239694.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 04:05:29,842 INFO [train.py:901] (2/4) Epoch 30, batch 5300, loss[loss=0.2171, simple_loss=0.3017, pruned_loss=0.06622, over 8764.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2819, pruned_loss=0.05713, over 1614816.44 frames. ], batch size: 30, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 04:05:43,733 INFO [optim.py:369] (2/4) 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 04:05:51,710 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.31 vs. limit=5.0 +2023-02-09 04:06:04,874 INFO [train.py:901] (2/4) Epoch 30, batch 5350, loss[loss=0.1696, simple_loss=0.258, pruned_loss=0.0406, over 8504.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2822, pruned_loss=0.05739, over 1614735.10 frames. ], batch size: 28, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 04:06:41,533 INFO [train.py:901] (2/4) Epoch 30, batch 5400, loss[loss=0.2232, simple_loss=0.3129, pruned_loss=0.06673, over 8343.00 frames. ], tot_loss[loss=0.1989, simple_loss=0.2827, pruned_loss=0.05756, over 1610162.89 frames. ], batch size: 26, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 04:06:55,041 INFO [optim.py:369] (2/4) 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 04:07:01,254 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.4725, 2.3664, 3.0336, 2.4360, 3.1254, 2.5120, 2.3950, 1.9435], + device='cuda:2'), covar=tensor([0.5720, 0.5298, 0.2344, 0.4414, 0.2788, 0.3241, 0.1940, 0.6026], + device='cuda:2'), in_proj_covar=tensor([0.0969, 0.1035, 0.0848, 0.1010, 0.1032, 0.0949, 0.0780, 0.0857], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 04:07:09,548 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5384, 2.3233, 1.7449, 2.2717, 1.9771, 1.5386, 1.9617, 1.9977], + device='cuda:2'), covar=tensor([0.1403, 0.0459, 0.1427, 0.0556, 0.0798, 0.1635, 0.0938, 0.0890], + device='cuda:2'), in_proj_covar=tensor([0.0356, 0.0244, 0.0343, 0.0313, 0.0299, 0.0346, 0.0347, 0.0318], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 04:07:17,645 INFO [train.py:901] (2/4) Epoch 30, batch 5450, loss[loss=0.2139, simple_loss=0.3057, pruned_loss=0.06112, over 8102.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2821, pruned_loss=0.05715, over 1610609.13 frames. ], batch size: 23, lr: 2.50e-03, grad_scale: 16.0 +2023-02-09 04:07:49,868 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0014-44390_sp0.9 from training. Duration: 31.02225 +2023-02-09 04:07:53,972 INFO [train.py:901] (2/4) Epoch 30, batch 5500, loss[loss=0.2153, simple_loss=0.2986, pruned_loss=0.06596, over 8494.00 frames. ], tot_loss[loss=0.1985, simple_loss=0.2822, pruned_loss=0.05738, over 1611064.96 frames. ], batch size: 26, lr: 2.50e-03, grad_scale: 8.0 +2023-02-09 04:08:08,780 INFO [optim.py:369] (2/4) 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 04:08:30,322 INFO [train.py:901] (2/4) Epoch 30, batch 5550, loss[loss=0.1712, simple_loss=0.2513, pruned_loss=0.04556, over 7542.00 frames. ], tot_loss[loss=0.1987, simple_loss=0.282, pruned_loss=0.05769, over 1609950.18 frames. ], batch size: 18, lr: 2.50e-03, grad_scale: 8.0 +2023-02-09 04:08:50,818 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=239984.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:09:07,157 INFO [train.py:901] (2/4) Epoch 30, batch 5600, loss[loss=0.1971, simple_loss=0.295, pruned_loss=0.04959, over 8345.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2826, pruned_loss=0.05775, over 1608414.39 frames. ], batch size: 25, lr: 2.50e-03, grad_scale: 8.0 +2023-02-09 04:09:21,032 INFO [optim.py:369] (2/4) 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 04:09:42,587 INFO [train.py:901] (2/4) Epoch 30, batch 5650, loss[loss=0.1888, simple_loss=0.2815, pruned_loss=0.04805, over 8580.00 frames. ], tot_loss[loss=0.199, simple_loss=0.2826, pruned_loss=0.05772, over 1610540.00 frames. ], batch size: 31, lr: 2.50e-03, grad_scale: 8.0 +2023-02-09 04:09:59,655 WARNING [train.py:1067] (2/4) Exclude cut with ID 6758-72288-0033-148662_sp0.9 from training. Duration: 25.988875 +2023-02-09 04:10:14,051 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=240099.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:10:16,487 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=3.88 vs. limit=5.0 +2023-02-09 04:10:17,315 INFO [train.py:901] (2/4) Epoch 30, batch 5700, loss[loss=0.1882, simple_loss=0.2806, pruned_loss=0.04793, over 8197.00 frames. ], tot_loss[loss=0.1995, simple_loss=0.283, pruned_loss=0.05796, over 1614680.18 frames. ], batch size: 23, lr: 2.50e-03, grad_scale: 8.0 +2023-02-09 04:10:32,005 INFO [optim.py:369] (2/4) 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 04:10:53,042 INFO [train.py:901] (2/4) Epoch 30, batch 5750, loss[loss=0.2201, simple_loss=0.2962, pruned_loss=0.07199, over 8464.00 frames. ], tot_loss[loss=0.1999, simple_loss=0.2831, pruned_loss=0.05833, over 1614041.00 frames. ], batch size: 25, lr: 2.50e-03, grad_scale: 8.0 +2023-02-09 04:11:04,197 WARNING [train.py:1067] (2/4) Exclude cut with ID 3972-170212-0014-103914_sp0.9 from training. Duration: 29.1166875 +2023-02-09 04:11:28,591 INFO [train.py:901] (2/4) Epoch 30, batch 5800, loss[loss=0.2007, simple_loss=0.2801, pruned_loss=0.0607, over 8111.00 frames. ], tot_loss[loss=0.1983, simple_loss=0.2815, pruned_loss=0.05753, over 1610555.71 frames. ], batch size: 23, lr: 2.50e-03, grad_scale: 8.0 +2023-02-09 04:11:42,452 INFO [optim.py:369] (2/4) 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 04:11:43,511 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.75 vs. limit=5.0 +2023-02-09 04:12:04,252 INFO [train.py:901] (2/4) Epoch 30, batch 5850, loss[loss=0.2281, simple_loss=0.3015, pruned_loss=0.07733, over 7289.00 frames. ], tot_loss[loss=0.1977, simple_loss=0.2813, pruned_loss=0.05708, over 1608351.21 frames. ], batch size: 73, lr: 2.50e-03, grad_scale: 8.0 +2023-02-09 04:12:39,827 INFO [train.py:901] (2/4) Epoch 30, batch 5900, loss[loss=0.224, simple_loss=0.308, pruned_loss=0.06995, over 8445.00 frames. ], tot_loss[loss=0.1968, simple_loss=0.2805, pruned_loss=0.05653, over 1608165.78 frames. ], batch size: 48, lr: 2.50e-03, grad_scale: 8.0 +2023-02-09 04:12:53,715 INFO [optim.py:369] (2/4) 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 04:13:13,640 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.9136, 2.0976, 1.7517, 2.8276, 1.3054, 1.6795, 2.1318, 2.1018], + device='cuda:2'), covar=tensor([0.0786, 0.0852, 0.0913, 0.0322, 0.1066, 0.1252, 0.0754, 0.0827], + device='cuda:2'), in_proj_covar=tensor([0.0230, 0.0193, 0.0244, 0.0213, 0.0201, 0.0245, 0.0247, 0.0203], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0005, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 04:13:15,499 INFO [train.py:901] (2/4) Epoch 30, batch 5950, loss[loss=0.2152, simple_loss=0.2921, pruned_loss=0.06916, over 7808.00 frames. ], tot_loss[loss=0.1982, simple_loss=0.2818, pruned_loss=0.05723, over 1610209.67 frames. ], batch size: 19, lr: 2.50e-03, grad_scale: 8.0 +2023-02-09 04:13:16,424 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=240355.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:13:18,458 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=240358.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:13:33,807 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=240380.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:13:50,594 INFO [train.py:901] (2/4) Epoch 30, batch 6000, loss[loss=0.2247, simple_loss=0.3097, pruned_loss=0.06986, over 8772.00 frames. ], tot_loss[loss=0.1975, simple_loss=0.281, pruned_loss=0.05702, over 1609916.44 frames. ], batch size: 30, lr: 2.50e-03, grad_scale: 8.0 +2023-02-09 04:13:50,594 INFO [train.py:926] (2/4) Computing validation loss +2023-02-09 04:14:04,302 INFO [train.py:935] (2/4) Epoch 30, validation: loss=0.1701, simple_loss=0.2695, pruned_loss=0.03536, over 944034.00 frames. +2023-02-09 04:14:04,303 INFO [train.py:936] (2/4) Maximum memory allocated so far is 6731MB +2023-02-09 04:14:07,896 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3546, 1.4564, 1.3536, 1.7980, 0.7287, 1.1920, 1.3062, 1.4739], + device='cuda:2'), covar=tensor([0.0856, 0.0755, 0.0939, 0.0454, 0.1129, 0.1385, 0.0738, 0.0686], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0194, 0.0244, 0.0213, 0.0202, 0.0246, 0.0248, 0.0203], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 04:14:17,955 INFO [optim.py:369] (2/4) 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 04:14:39,916 INFO [train.py:901] (2/4) Epoch 30, batch 6050, loss[loss=0.1591, simple_loss=0.2377, pruned_loss=0.04022, over 7687.00 frames. ], tot_loss[loss=0.1976, simple_loss=0.2811, pruned_loss=0.05709, over 1609950.23 frames. ], batch size: 18, lr: 2.50e-03, grad_scale: 8.0 +2023-02-09 04:15:16,411 INFO [train.py:901] (2/4) Epoch 30, batch 6100, loss[loss=0.1812, simple_loss=0.271, pruned_loss=0.04573, over 8468.00 frames. ], tot_loss[loss=0.1981, simple_loss=0.2818, pruned_loss=0.05725, over 1615429.95 frames. ], batch size: 25, lr: 2.50e-03, grad_scale: 8.0 +2023-02-09 04:15:30,263 INFO [optim.py:369] (2/4) 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 04:15:40,575 WARNING [train.py:1067] (2/4) Exclude cut with ID 3033-130750-0096-107983_sp0.9 from training. Duration: 0.92225 +2023-02-09 04:15:51,558 INFO [train.py:901] (2/4) Epoch 30, batch 6150, loss[loss=0.1819, simple_loss=0.2609, pruned_loss=0.05152, over 7788.00 frames. ], tot_loss[loss=0.1988, simple_loss=0.2822, pruned_loss=0.05774, over 1618486.56 frames. ], batch size: 19, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:16:24,186 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.8890, 1.3625, 3.1437, 1.5413, 2.4173, 3.3681, 3.5146, 2.9468], + device='cuda:2'), covar=tensor([0.1200, 0.1994, 0.0331, 0.2105, 0.0947, 0.0262, 0.0651, 0.0506], + device='cuda:2'), in_proj_covar=tensor([0.0309, 0.0330, 0.0296, 0.0330, 0.0330, 0.0284, 0.0452, 0.0309], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:2') +2023-02-09 04:16:28,465 INFO [train.py:901] (2/4) Epoch 30, batch 6200, loss[loss=0.1974, simple_loss=0.2836, pruned_loss=0.05563, over 8331.00 frames. ], tot_loss[loss=0.1986, simple_loss=0.2818, pruned_loss=0.05765, over 1616452.32 frames. ], batch size: 26, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:16:44,278 INFO [optim.py:369] (2/4) 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 04:17:05,850 INFO [train.py:901] (2/4) Epoch 30, batch 6250, loss[loss=0.2322, simple_loss=0.299, pruned_loss=0.08267, over 7968.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.28, pruned_loss=0.05691, over 1611735.74 frames. ], batch size: 21, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:17:06,787 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5747, 2.9509, 2.5163, 4.0560, 1.5917, 2.1751, 2.5120, 2.7651], + device='cuda:2'), covar=tensor([0.0669, 0.0744, 0.0675, 0.0201, 0.1104, 0.1196, 0.0895, 0.0763], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0194, 0.0245, 0.0214, 0.0203, 0.0247, 0.0248, 0.0205], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 04:17:16,721 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.37 vs. limit=2.0 +2023-02-09 04:17:39,922 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=240702.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:17:41,235 INFO [train.py:901] (2/4) Epoch 30, batch 6300, loss[loss=0.1797, simple_loss=0.2627, pruned_loss=0.04836, over 7519.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.2798, pruned_loss=0.05606, over 1608233.25 frames. ], batch size: 18, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:17:51,941 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=2.93 vs. limit=5.0 +2023-02-09 04:17:54,933 INFO [optim.py:369] (2/4) 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 04:18:17,086 INFO [train.py:901] (2/4) Epoch 30, batch 6350, loss[loss=0.1749, simple_loss=0.2624, pruned_loss=0.04374, over 7804.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2806, pruned_loss=0.05681, over 1610380.86 frames. ], batch size: 19, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:18:53,328 INFO [train.py:901] (2/4) Epoch 30, batch 6400, loss[loss=0.1796, simple_loss=0.2637, pruned_loss=0.0477, over 7534.00 frames. ], tot_loss[loss=0.1973, simple_loss=0.281, pruned_loss=0.05674, over 1613211.69 frames. ], batch size: 18, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:19:02,608 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=240817.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:19:07,074 INFO [optim.py:369] (2/4) 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 04:19:28,696 INFO [train.py:901] (2/4) Epoch 30, batch 6450, loss[loss=0.1796, simple_loss=0.2578, pruned_loss=0.05071, over 7418.00 frames. ], tot_loss[loss=0.1984, simple_loss=0.2822, pruned_loss=0.05728, over 1612937.77 frames. ], batch size: 17, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:20:03,783 INFO [train.py:901] (2/4) Epoch 30, batch 6500, loss[loss=0.1867, simple_loss=0.2741, pruned_loss=0.04967, over 8292.00 frames. ], tot_loss[loss=0.198, simple_loss=0.282, pruned_loss=0.05701, over 1609614.08 frames. ], batch size: 23, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:20:16,798 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7850, 1.9019, 1.7383, 2.2631, 0.9621, 1.5467, 1.7388, 1.9229], + device='cuda:2'), covar=tensor([0.0755, 0.0759, 0.0893, 0.0414, 0.1055, 0.1298, 0.0720, 0.0680], + device='cuda:2'), in_proj_covar=tensor([0.0232, 0.0194, 0.0245, 0.0214, 0.0203, 0.0247, 0.0249, 0.0204], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 04:20:17,923 INFO [optim.py:369] (2/4) 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 04:20:36,286 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=240950.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 04:20:38,871 INFO [train.py:901] (2/4) Epoch 30, batch 6550, loss[loss=0.1794, simple_loss=0.2616, pruned_loss=0.04862, over 8499.00 frames. ], tot_loss[loss=0.1974, simple_loss=0.2816, pruned_loss=0.05659, over 1611541.24 frames. ], batch size: 26, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:21:00,575 WARNING [train.py:1067] (2/4) Exclude cut with ID 3557-8342-0013-71585 from training. Duration: 0.92 +2023-02-09 04:21:15,890 INFO [train.py:901] (2/4) Epoch 30, batch 6600, loss[loss=0.2261, simple_loss=0.3191, pruned_loss=0.06659, over 8477.00 frames. ], tot_loss[loss=0.1969, simple_loss=0.281, pruned_loss=0.05639, over 1609323.68 frames. ], batch size: 27, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:21:16,033 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=241004.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:21:20,065 WARNING [train.py:1067] (2/4) Exclude cut with ID 4133-6541-0027-26893_sp1.1 from training. Duration: 0.9681875 +2023-02-09 04:21:27,612 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.39 vs. limit=2.0 +2023-02-09 04:21:29,847 INFO [optim.py:369] (2/4) 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 04:21:51,501 INFO [train.py:901] (2/4) Epoch 30, batch 6650, loss[loss=0.1904, simple_loss=0.2643, pruned_loss=0.05825, over 7711.00 frames. ], tot_loss[loss=0.1948, simple_loss=0.279, pruned_loss=0.05532, over 1606412.90 frames. ], batch size: 18, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:21:53,753 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5936, 1.6299, 4.3249, 2.0024, 2.7010, 4.8910, 5.0176, 4.2446], + device='cuda:2'), covar=tensor([0.1015, 0.2016, 0.0276, 0.1937, 0.1059, 0.0174, 0.0454, 0.0562], + device='cuda:2'), in_proj_covar=tensor([0.0311, 0.0331, 0.0299, 0.0331, 0.0332, 0.0285, 0.0455, 0.0311], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:2') +2023-02-09 04:22:04,820 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=241073.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:22:07,893 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.82 vs. limit=2.0 +2023-02-09 04:22:08,260 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5694, 1.3533, 2.3470, 1.3002, 2.2301, 2.4906, 2.7051, 2.1164], + device='cuda:2'), covar=tensor([0.1158, 0.1529, 0.0427, 0.2208, 0.0756, 0.0408, 0.0615, 0.0671], + device='cuda:2'), in_proj_covar=tensor([0.0312, 0.0332, 0.0299, 0.0331, 0.0333, 0.0286, 0.0456, 0.0312], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0003, 0.0003, 0.0003, 0.0002, 0.0004, 0.0002], + device='cuda:2') +2023-02-09 04:22:23,569 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=241098.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:22:27,617 INFO [train.py:901] (2/4) Epoch 30, batch 6700, loss[loss=0.1711, simple_loss=0.2491, pruned_loss=0.0465, over 7422.00 frames. ], tot_loss[loss=0.196, simple_loss=0.2803, pruned_loss=0.05591, over 1612543.87 frames. ], batch size: 17, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:22:28,144 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.40 vs. limit=2.0 +2023-02-09 04:22:42,111 INFO [optim.py:369] (2/4) 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 04:22:57,376 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=192, metric=1.65 vs. limit=2.0 +2023-02-09 04:23:04,034 INFO [train.py:901] (2/4) Epoch 30, batch 6750, loss[loss=0.182, simple_loss=0.2691, pruned_loss=0.04747, over 7649.00 frames. ], tot_loss[loss=0.1958, simple_loss=0.2803, pruned_loss=0.05562, over 1616486.60 frames. ], batch size: 19, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:23:39,180 INFO [train.py:901] (2/4) Epoch 30, batch 6800, loss[loss=0.2325, simple_loss=0.3177, pruned_loss=0.0737, over 8462.00 frames. ], tot_loss[loss=0.197, simple_loss=0.2816, pruned_loss=0.05616, over 1617884.34 frames. ], batch size: 29, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:23:42,693 WARNING [train.py:1067] (2/4) Exclude cut with ID 8291-282929-0024-9607_sp0.9 from training. Duration: 26.438875 +2023-02-09 04:23:53,781 INFO [optim.py:369] (2/4) 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 04:23:54,034 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5598, 1.8589, 2.9053, 1.4399, 2.1353, 1.9807, 1.6017, 2.2127], + device='cuda:2'), covar=tensor([0.2236, 0.3107, 0.0992, 0.5263, 0.2201, 0.3535, 0.2944, 0.2490], + device='cuda:2'), in_proj_covar=tensor([0.0549, 0.0651, 0.0571, 0.0681, 0.0675, 0.0624, 0.0576, 0.0653], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0003, 0.0003, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 04:24:05,764 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.5701, 5.6296, 5.0746, 2.5603, 5.0409, 5.3005, 5.1134, 5.0851], + device='cuda:2'), covar=tensor([0.0595, 0.0418, 0.0881, 0.4161, 0.0772, 0.0859, 0.1049, 0.0634], + device='cuda:2'), in_proj_covar=tensor([0.0556, 0.0466, 0.0458, 0.0568, 0.0451, 0.0480, 0.0454, 0.0419], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 04:24:15,459 INFO [train.py:901] (2/4) Epoch 30, batch 6850, loss[loss=0.1898, simple_loss=0.2609, pruned_loss=0.05939, over 7244.00 frames. ], tot_loss[loss=0.1953, simple_loss=0.2797, pruned_loss=0.05541, over 1617672.29 frames. ], batch size: 16, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:24:34,755 WARNING [train.py:1067] (2/4) Exclude cut with ID 2929-85685-0079-61403_sp1.1 from training. Duration: 27.0318125 +2023-02-09 04:24:42,628 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.7431, 2.1998, 3.7773, 1.5619, 2.6959, 2.2608, 1.8369, 2.7094], + device='cuda:2'), covar=tensor([0.1996, 0.2910, 0.0987, 0.4786, 0.2167, 0.3441, 0.2611, 0.2869], + device='cuda:2'), in_proj_covar=tensor([0.0548, 0.0650, 0.0569, 0.0679, 0.0674, 0.0623, 0.0575, 0.0653], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0003, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 04:24:43,872 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=241294.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 04:24:50,576 INFO [train.py:901] (2/4) Epoch 30, batch 6900, loss[loss=0.189, simple_loss=0.2557, pruned_loss=0.06116, over 7422.00 frames. ], tot_loss[loss=0.1951, simple_loss=0.2797, pruned_loss=0.05527, over 1617383.55 frames. ], batch size: 17, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:25:05,717 INFO [optim.py:369] (2/4) 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 04:25:22,107 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=241348.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:25:26,914 INFO [train.py:901] (2/4) Epoch 30, batch 6950, loss[loss=0.1888, simple_loss=0.279, pruned_loss=0.04935, over 8295.00 frames. ], tot_loss[loss=0.1945, simple_loss=0.2787, pruned_loss=0.05512, over 1615863.64 frames. ], batch size: 23, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:25:32,447 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2211, 1.8594, 2.5541, 1.5130, 1.7054, 2.5068, 1.3166, 2.0969], + device='cuda:2'), covar=tensor([0.1459, 0.1004, 0.0345, 0.1177, 0.1690, 0.0607, 0.1494, 0.1146], + device='cuda:2'), in_proj_covar=tensor([0.0202, 0.0206, 0.0137, 0.0222, 0.0276, 0.0149, 0.0174, 0.0200], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 04:25:46,728 WARNING [train.py:1067] (2/4) Exclude cut with ID 7255-291500-0009-9471_sp0.9 from training. Duration: 26.62775 +2023-02-09 04:25:53,380 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.5821, 3.0103, 2.3353, 4.0708, 1.7151, 2.1293, 2.6451, 2.8457], + device='cuda:2'), covar=tensor([0.0655, 0.0667, 0.0768, 0.0193, 0.1019, 0.1242, 0.0882, 0.0799], + device='cuda:2'), in_proj_covar=tensor([0.0231, 0.0193, 0.0244, 0.0213, 0.0201, 0.0246, 0.0247, 0.0203], + device='cuda:2'), out_proj_covar=tensor([0.0005, 0.0005, 0.0006, 0.0005, 0.0005, 0.0006, 0.0006, 0.0005], + device='cuda:2') +2023-02-09 04:26:04,041 INFO [train.py:901] (2/4) Epoch 30, batch 7000, loss[loss=0.2113, simple_loss=0.2873, pruned_loss=0.06767, over 8084.00 frames. ], tot_loss[loss=0.1945, simple_loss=0.2786, pruned_loss=0.05514, over 1614076.03 frames. ], batch size: 21, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:26:07,764 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=241409.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 04:26:17,959 INFO [optim.py:369] (2/4) 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 04:26:18,117 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=241424.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:26:30,777 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.1433, 1.2801, 1.2182, 0.9190, 1.2189, 1.0583, 0.1497, 1.2229], + device='cuda:2'), covar=tensor([0.0510, 0.0464, 0.0442, 0.0570, 0.0541, 0.1191, 0.0976, 0.0380], + device='cuda:2'), in_proj_covar=tensor([0.0483, 0.0418, 0.0377, 0.0468, 0.0405, 0.0560, 0.0406, 0.0447], + device='cuda:2'), out_proj_covar=tensor([1.2779e-04, 1.0769e-04, 9.8145e-05, 1.2203e-04, 1.0567e-04, 1.5574e-04, + 1.0832e-04, 1.1685e-04], device='cuda:2') +2023-02-09 04:26:40,280 INFO [train.py:901] (2/4) Epoch 30, batch 7050, loss[loss=0.1999, simple_loss=0.2873, pruned_loss=0.05621, over 8557.00 frames. ], tot_loss[loss=0.1951, simple_loss=0.279, pruned_loss=0.05561, over 1613609.28 frames. ], batch size: 31, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:26:46,391 INFO [scaling.py:679] (2/4) Whitening: num_groups=1, num_channels=256, metric=4.36 vs. limit=5.0 +2023-02-09 04:26:46,797 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=241463.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:27:16,738 INFO [train.py:901] (2/4) Epoch 30, batch 7100, loss[loss=0.1949, simple_loss=0.267, pruned_loss=0.06136, over 7530.00 frames. ], tot_loss[loss=0.1964, simple_loss=0.2803, pruned_loss=0.05624, over 1615291.19 frames. ], batch size: 18, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:27:28,169 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.3405, 2.1249, 1.6952, 2.0270, 1.7122, 1.4391, 1.6077, 1.6643], + device='cuda:2'), covar=tensor([0.1318, 0.0474, 0.1345, 0.0545, 0.0824, 0.1679, 0.1029, 0.0966], + device='cuda:2'), in_proj_covar=tensor([0.0363, 0.0247, 0.0349, 0.0318, 0.0305, 0.0352, 0.0353, 0.0323], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 04:27:30,723 INFO [optim.py:369] (2/4) 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 04:27:51,598 INFO [train.py:901] (2/4) Epoch 30, batch 7150, loss[loss=0.1617, simple_loss=0.2306, pruned_loss=0.04636, over 7416.00 frames. ], tot_loss[loss=0.1971, simple_loss=0.2812, pruned_loss=0.05655, over 1617978.54 frames. ], batch size: 17, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:28:28,702 INFO [train.py:901] (2/4) Epoch 30, batch 7200, loss[loss=0.1847, simple_loss=0.2623, pruned_loss=0.05358, over 8081.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2809, pruned_loss=0.05614, over 1620207.82 frames. ], batch size: 21, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:28:43,487 INFO [optim.py:369] (2/4) 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] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=241639.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:29:03,963 INFO [train.py:901] (2/4) Epoch 30, batch 7250, loss[loss=0.2025, simple_loss=0.2884, pruned_loss=0.05824, over 7940.00 frames. ], tot_loss[loss=0.196, simple_loss=0.2799, pruned_loss=0.05603, over 1611674.76 frames. ], batch size: 20, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:29:11,986 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=241665.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 04:29:30,386 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=241690.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 04:29:40,038 INFO [train.py:901] (2/4) Epoch 30, batch 7300, loss[loss=0.1904, simple_loss=0.2801, pruned_loss=0.05041, over 8459.00 frames. ], tot_loss[loss=0.1951, simple_loss=0.2794, pruned_loss=0.05546, over 1612661.89 frames. ], batch size: 25, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:29:50,473 INFO [zipformer.py:1185] (2/4) 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] (2/4) 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 04:30:08,887 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=241744.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:30:15,572 INFO [train.py:901] (2/4) Epoch 30, batch 7350, loss[loss=0.1804, simple_loss=0.2594, pruned_loss=0.05072, over 7556.00 frames. ], tot_loss[loss=0.1955, simple_loss=0.2796, pruned_loss=0.0557, over 1613680.95 frames. ], batch size: 18, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:30:25,405 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=241768.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:30:39,742 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0026-44402_sp0.9 from training. Duration: 25.061125 +2023-02-09 04:30:46,599 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([5.9678, 1.6789, 6.1269, 2.3317, 5.4823, 5.0294, 5.6193, 5.5230], + device='cuda:2'), covar=tensor([0.0486, 0.4753, 0.0460, 0.3804, 0.1068, 0.0897, 0.0512, 0.0571], + device='cuda:2'), in_proj_covar=tensor([0.0703, 0.0676, 0.0762, 0.0680, 0.0765, 0.0653, 0.0664, 0.0740], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 04:30:51,563 INFO [train.py:901] (2/4) Epoch 30, batch 7400, loss[loss=0.178, simple_loss=0.2738, pruned_loss=0.04109, over 8131.00 frames. ], tot_loss[loss=0.1952, simple_loss=0.2793, pruned_loss=0.05559, over 1609285.72 frames. ], batch size: 22, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:30:59,777 WARNING [train.py:1067] (2/4) Exclude cut with ID 774-127930-0014-48411_sp1.1 from training. Duration: 0.95 +2023-02-09 04:31:04,085 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=241821.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:31:05,936 INFO [optim.py:369] (2/4) 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 04:31:24,334 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=241849.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 04:31:27,627 INFO [train.py:901] (2/4) Epoch 30, batch 7450, loss[loss=0.2036, simple_loss=0.2865, pruned_loss=0.0604, over 8089.00 frames. ], tot_loss[loss=0.1957, simple_loss=0.2797, pruned_loss=0.05584, over 1611662.06 frames. ], batch size: 21, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:31:40,210 WARNING [train.py:1067] (2/4) Exclude cut with ID 7699-105389-0094-102071_sp0.9 from training. Duration: 26.6166875 +2023-02-09 04:31:48,184 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=241883.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:31:48,872 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2663, 3.4453, 2.1928, 3.0338, 2.8381, 2.0186, 2.8300, 2.9813], + device='cuda:2'), covar=tensor([0.1607, 0.0430, 0.1322, 0.0663, 0.0689, 0.1580, 0.1027, 0.1083], + device='cuda:2'), in_proj_covar=tensor([0.0361, 0.0245, 0.0348, 0.0317, 0.0304, 0.0350, 0.0351, 0.0322], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 04:32:02,782 INFO [train.py:901] (2/4) Epoch 30, batch 7500, loss[loss=0.18, simple_loss=0.2757, pruned_loss=0.04219, over 8546.00 frames. ], tot_loss[loss=0.1948, simple_loss=0.279, pruned_loss=0.05527, over 1613679.67 frames. ], batch size: 49, lr: 2.49e-03, grad_scale: 16.0 +2023-02-09 04:32:13,912 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.45 vs. limit=2.0 +2023-02-09 04:32:18,961 INFO [optim.py:369] (2/4) 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 04:32:28,836 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4864, 1.8396, 1.8614, 1.2239, 1.8812, 1.4411, 0.4677, 1.7283], + device='cuda:2'), covar=tensor([0.0851, 0.0441, 0.0437, 0.0785, 0.0644, 0.1227, 0.1181, 0.0387], + device='cuda:2'), in_proj_covar=tensor([0.0482, 0.0418, 0.0376, 0.0467, 0.0404, 0.0560, 0.0407, 0.0446], + device='cuda:2'), out_proj_covar=tensor([1.2741e-04, 1.0779e-04, 9.7851e-05, 1.2190e-04, 1.0548e-04, 1.5577e-04, + 1.0841e-04, 1.1653e-04], device='cuda:2') +2023-02-09 04:32:39,978 INFO [train.py:901] (2/4) Epoch 30, batch 7550, loss[loss=0.2159, simple_loss=0.3055, pruned_loss=0.06318, over 7970.00 frames. ], tot_loss[loss=0.1949, simple_loss=0.2793, pruned_loss=0.05529, over 1612062.64 frames. ], batch size: 21, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:32:46,886 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-09 04:33:01,507 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=241983.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:33:03,586 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.4108, 4.4239, 3.9858, 2.0570, 3.8505, 4.0185, 3.9821, 3.8728], + device='cuda:2'), covar=tensor([0.0710, 0.0500, 0.1042, 0.4531, 0.0975, 0.0922, 0.1195, 0.0793], + device='cuda:2'), in_proj_covar=tensor([0.0553, 0.0460, 0.0454, 0.0563, 0.0449, 0.0478, 0.0450, 0.0416], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0002, 0.0002, 0.0003, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 04:33:12,195 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.5067, 1.3927, 1.8031, 1.1717, 1.1405, 1.7676, 0.2382, 1.1102], + device='cuda:2'), covar=tensor([0.1403, 0.1129, 0.0371, 0.0825, 0.2242, 0.0425, 0.1714, 0.1169], + device='cuda:2'), in_proj_covar=tensor([0.0202, 0.0205, 0.0137, 0.0222, 0.0276, 0.0148, 0.0174, 0.0200], + device='cuda:2'), out_proj_covar=tensor([0.0003, 0.0003, 0.0002, 0.0003, 0.0004, 0.0002, 0.0003, 0.0003], + device='cuda:2') +2023-02-09 04:33:17,220 INFO [train.py:901] (2/4) Epoch 30, batch 7600, loss[loss=0.1823, simple_loss=0.2621, pruned_loss=0.05122, over 7933.00 frames. ], tot_loss[loss=0.1951, simple_loss=0.2792, pruned_loss=0.05548, over 1612124.84 frames. ], batch size: 20, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:33:32,888 INFO [optim.py:369] (2/4) 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 04:33:54,462 INFO [train.py:901] (2/4) Epoch 30, batch 7650, loss[loss=0.192, simple_loss=0.2706, pruned_loss=0.05668, over 7973.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2803, pruned_loss=0.05653, over 1610967.39 frames. ], batch size: 21, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:33:56,937 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.28 vs. limit=2.0 +2023-02-09 04:34:25,721 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242098.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:34:29,810 INFO [train.py:901] (2/4) Epoch 30, batch 7700, loss[loss=0.1711, simple_loss=0.2555, pruned_loss=0.04335, over 8138.00 frames. ], tot_loss[loss=0.1967, simple_loss=0.2806, pruned_loss=0.05645, over 1615597.74 frames. ], batch size: 22, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:34:39,658 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.4804, 1.3587, 1.4827, 1.3203, 0.9356, 1.4048, 1.4652, 1.4245], + device='cuda:2'), covar=tensor([0.0733, 0.0955, 0.1335, 0.1220, 0.0627, 0.1139, 0.0764, 0.0507], + device='cuda:2'), in_proj_covar=tensor([0.0101, 0.0154, 0.0191, 0.0162, 0.0102, 0.0164, 0.0113, 0.0148], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0007, 0.0008], + device='cuda:2') +2023-02-09 04:34:44,311 INFO [optim.py:369] (2/4) 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 04:34:54,353 WARNING [train.py:1067] (2/4) Exclude cut with ID 7357-94126-0009-44385_sp0.9 from training. Duration: 27.02225 +2023-02-09 04:34:55,206 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([4.7330, 1.4205, 4.9688, 1.9274, 4.4741, 4.1622, 4.4823, 4.3872], + device='cuda:2'), covar=tensor([0.0552, 0.4826, 0.0415, 0.4207, 0.0982, 0.0859, 0.0568, 0.0614], + device='cuda:2'), in_proj_covar=tensor([0.0699, 0.0672, 0.0757, 0.0675, 0.0758, 0.0646, 0.0659, 0.0733], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 04:34:55,291 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=242139.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:35:05,979 INFO [train.py:901] (2/4) Epoch 30, batch 7750, loss[loss=0.1943, simple_loss=0.2784, pruned_loss=0.0551, over 8191.00 frames. ], tot_loss[loss=0.1966, simple_loss=0.2802, pruned_loss=0.05656, over 1612312.38 frames. ], batch size: 23, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:35:13,195 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=242164.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:35:13,775 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=242165.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:35:18,607 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242171.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:35:34,906 INFO [zipformer.py:1185] (2/4) warmup_begin=666.7, warmup_end=1333.3, batch_count=242193.0, num_to_drop=1, layers_to_drop={0} +2023-02-09 04:35:42,648 INFO [train.py:901] (2/4) Epoch 30, batch 7800, loss[loss=0.1728, simple_loss=0.2568, pruned_loss=0.04438, over 7693.00 frames. ], tot_loss[loss=0.1959, simple_loss=0.2799, pruned_loss=0.05595, over 1616784.09 frames. ], batch size: 18, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:35:44,308 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([1.6091, 2.2744, 1.4435, 2.4124, 2.0432, 1.2317, 1.8820, 2.3901], + device='cuda:2'), covar=tensor([0.1394, 0.0478, 0.1551, 0.0595, 0.0871, 0.1947, 0.1180, 0.0690], + device='cuda:2'), in_proj_covar=tensor([0.0358, 0.0243, 0.0343, 0.0313, 0.0301, 0.0347, 0.0347, 0.0319], + device='cuda:2'), out_proj_covar=tensor([0.0004, 0.0003, 0.0004, 0.0003, 0.0003, 0.0004, 0.0004, 0.0003], + device='cuda:2') +2023-02-09 04:35:57,874 INFO [optim.py:369] (2/4) 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 04:36:11,177 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242244.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:36:18,002 INFO [train.py:901] (2/4) Epoch 30, batch 7850, loss[loss=0.1925, simple_loss=0.2793, pruned_loss=0.05284, over 8039.00 frames. ], tot_loss[loss=0.1952, simple_loss=0.2793, pruned_loss=0.05556, over 1620416.25 frames. ], batch size: 22, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:36:36,021 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242280.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:36:47,919 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.7574, 1.6499, 1.8590, 1.7376, 1.0670, 1.7111, 2.1977, 1.9610], + device='cuda:2'), covar=tensor([0.0509, 0.1252, 0.1721, 0.1448, 0.0609, 0.1438, 0.0668, 0.0653], + device='cuda:2'), in_proj_covar=tensor([0.0101, 0.0154, 0.0191, 0.0163, 0.0102, 0.0164, 0.0113, 0.0148], + device='cuda:2'), out_proj_covar=tensor([0.0007, 0.0009, 0.0010, 0.0010, 0.0006, 0.0010, 0.0007, 0.0008], + device='cuda:2') +2023-02-09 04:36:52,301 INFO [train.py:901] (2/4) Epoch 30, batch 7900, loss[loss=0.2065, simple_loss=0.2967, pruned_loss=0.05817, over 8507.00 frames. ], tot_loss[loss=0.1952, simple_loss=0.2794, pruned_loss=0.05552, over 1616573.08 frames. ], batch size: 26, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:36:54,992 INFO [zipformer.py:1185] (2/4) warmup_begin=2000.0, warmup_end=2666.7, batch_count=242308.0, num_to_drop=1, layers_to_drop={1} +2023-02-09 04:37:06,487 INFO [optim.py:369] (2/4) 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 04:37:08,036 INFO [zipformer.py:1185] (2/4) warmup_begin=1333.3, warmup_end=2000.0, batch_count=242327.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:37:20,020 INFO [scaling.py:679] (2/4) Whitening: num_groups=8, num_channels=96, metric=1.25 vs. limit=2.0 +2023-02-09 04:37:26,309 INFO [train.py:901] (2/4) Epoch 30, batch 7950, loss[loss=0.1805, simple_loss=0.2659, pruned_loss=0.04753, over 8087.00 frames. ], tot_loss[loss=0.1935, simple_loss=0.278, pruned_loss=0.05452, over 1618113.49 frames. ], batch size: 21, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:37:26,519 INFO [zipformer.py:1185] (2/4) warmup_begin=3333.3, warmup_end=4000.0, batch_count=242354.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:37:43,567 INFO [zipformer.py:1185] (2/4) warmup_begin=2666.7, warmup_end=3333.3, batch_count=242379.0, num_to_drop=0, layers_to_drop=set() +2023-02-09 04:38:00,600 INFO [train.py:901] (2/4) Epoch 30, batch 8000, loss[loss=0.2131, simple_loss=0.3019, pruned_loss=0.06214, over 8239.00 frames. ], tot_loss[loss=0.1937, simple_loss=0.2785, pruned_loss=0.05446, over 1618980.99 frames. ], batch size: 24, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:38:14,883 INFO [optim.py:369] (2/4) 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 04:38:24,833 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([3.5111, 1.7962, 3.6657, 1.9374, 3.3208, 3.0961, 3.3574, 3.2881], + device='cuda:2'), covar=tensor([0.0829, 0.3610, 0.0953, 0.4376, 0.1016, 0.0992, 0.0697, 0.0734], + device='cuda:2'), in_proj_covar=tensor([0.0705, 0.0677, 0.0764, 0.0680, 0.0764, 0.0652, 0.0663, 0.0740], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002], + device='cuda:2') +2023-02-09 04:38:28,339 INFO [zipformer.py:2431] (2/4) attn_weights_entropy = tensor([2.2807, 1.9833, 2.4494, 2.1191, 2.5172, 2.3211, 2.1522, 1.3651], + device='cuda:2'), covar=tensor([0.5683, 0.5101, 0.2325, 0.3973, 0.2826, 0.3463, 0.2001, 0.5801], + device='cuda:2'), in_proj_covar=tensor([0.0972, 0.1039, 0.0855, 0.1016, 0.1035, 0.0950, 0.0783, 0.0863], + device='cuda:2'), out_proj_covar=tensor([0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0002, 0.0001, 0.0002], + device='cuda:2') +2023-02-09 04:38:35,235 INFO [train.py:901] (2/4) Epoch 30, batch 8050, loss[loss=0.2452, simple_loss=0.3112, pruned_loss=0.08959, over 7039.00 frames. ], tot_loss[loss=0.1953, simple_loss=0.279, pruned_loss=0.0558, over 1601086.43 frames. ], batch size: 71, lr: 2.49e-03, grad_scale: 8.0 +2023-02-09 04:38:58,154 INFO [train.py:1165] (2/4) Done!